SNDK +1%: Ok looking print (best of the night) with better top line and mixed GM. F1Q revenue of $11.03B beat Street ($10.74B) and was in line with the ~$10.9-11.2B bogey, but GM of 85.1% beat Street and missed the ~86% bar, and the F2Q guide of $12.2-12.6B / 84.5-85.5% GM / $55-57 EPS beat Street ($12.08B / 84.7% / $52.94) and hit the top of the ~$12.2-12.4B bogey, while the flat GM guide sits well short of the 86.5-87.5% bulls wanted. Datacenter did the work (+61% q/q) as Edge and Consumer missed. NBMs: no new signings, still 8 customers and the $93.9B floor minimum vs. bulls wanting more than 8; RPO $88.9B vs. $91.1B pro forma in Aug. ... 8 for 1 stock split coming in November.

Not crowded coming in, but the stock ran +12% the two days before today’s -5%, and a GM guide that’s flat q/q not exciting and bulls were looking for more pricing flow-through but expects low heading in.

• F1Q revenue of $11.03B (+23.0% q/q) beat $10.74B Street and was in line with the ~$10.9-11.2B bogeys, 2.1% above the top of the $10.3-10.8B guide (Aug beat the top by 8.7%).

  • 60% of the q/q revenue increase came from volume and ~40% from pricing (Aug: one-third volume / two-thirds price). That’s ~+8.6% pricing and ~+13% bits by our math, vs. Street ASP +11.1% / bits +7.9% and the ASP +15-18% bogey: a volume-led quarter, which is why GM only ticked up.

  • Datacenter beat at $4.80B (+61% q/q) vs. $4.0B Street (+34%), now 43.5% of revenue (33% in June).

  • Edge missed at $5.83B (+7% q/q) vs. $6.08B Street (+12%); Consumer missed at $397M (-29% q/q) vs. $584M.

  

• GM beat Street at 85.07% but missed the bogey, ~60bps above Street’s 84.45%, 0.4pt below the 85.5-86.5% bogey and 0.1pt over the 83-85% guide top (84.64% in June, so +0.4pt q/q).

  • Pure price flow-through on the ~$230M over the guide top would have given ~85.3%, so no cushion showed up from the “prudent” component-cost assumptions.

  • EPS of $49.51 beat $46.73 Street, in line with the ~$48-50 bogeys and above the $44-46 guide, on 154M diluted shares (~155M guide).

  • A 14.3% non-GAAP tax rate vs. the 15% guide added ~$0.40 of that.

  • Opex of $513M was under the $520-540M guide; OM of 80.42% beat Street’s 79.5% (~90bps above).

• F2Q revenue guide of $12.2-12.6B ($12.4B midpoint, +12.4% q/q) beat $12.08B Street and hit the top of the ~$12.2-12.4B bogeys.

  • GM guide of 84.5-85.5% (85.0% midpoint, flat with F1Q’s 85.07%) is ~30bps above Street’s 84.7% but 1.5pt below the low end of the 86.5-87.5% bogey.

  • EPS guide of $55-57 ($56 midpoint) beat $52.94 Street and is at the top of the ~$54-56 bogeys, on ~151M diluted shares (154M in F1Q). Opex guide $540-560M, tax 14.3%.

  • Part of the Dec-Q step-up is consumer seasonality (F2Q26 Consumer $907M vs. $652M in F1Q26), not all pricing.

  • FY27 Street of $49.12B now needs $25.7B in 2H vs. $26.3B before, or ~$12.8B a quarter vs. the $12.4B Dec-Q midpoint.

• NBMs: no new signings, still 8 customers and the $93.9B floor minimum vs. bulls wanting more than 8; RPO $88.9B vs. $91.1B pro forma in Aug. Bit coverage unchanged at ~1/2 of FY27 and ~2/3 of FY28 (bogey: above ~2/3 for FY28); $16.5B of financial guarantees, 4+ year average duration. The two Q1 closes were already announced in Aug; “active conversations” on extensions and new NBMs, and the pipeline “exceeds our supply through the end of the decade”.

• Cash: adj FCF missed at $5.70B vs. $5.93B Street (reported FCF $6.31B, with receivables up $1.88B).

  • Bought back $5.0B (3.2M shares) vs. $4.5B in June, which leaves ~$10.5B of the $15.5B authorization by our math.

  • 8-for-1 stock split: record date Nov 9, trades split-adjusted from Nov 20.

RDDT -7%: US DAUq fell to 50.7M (-2.5M q/q, -2% y/y) vs. 53.5M Street and the +0.1-0.2M bogey, and the letter says total DAUq could fall by a low-double-digit % in 4Q on search referrals. Revenue of $895M beat Street ($877M) but missed the ~$935-945M bogeys, and the 4Q guide of $1,000-1,030M missed both $1,044M Street and the ~$1.07-1.08B bogeys. Logged-in and app DAUq grew q/q, so the call is all about how big the logged-out hole gets.

Stock was short-skewed coming in and ~50% off its highs (-3.8% today, -19% since 10/12), but this is worse than bears were looking for - US users fell 2.5M vs. the 0.3M in Q2 and the guide takes away the revenue cushion too. Huffman has an uphill battle on the call.

Key quote: “Our search referral traffic was down meaningfully in Q3, with the decline accelerating later in the quarter. The biggest impact was on logged-out and web traffic, but some of the impact flows through to logged-in and App traffic, since search referral is a common source of reengagement for users… These declines have continued into the early part of Q4. During October, we have seen that search traffic has dropped by more than 1 million referrals on average per week. If this rate of decline continues, current trends would suggest that total DAUqs could fall in Q4, by a low-double-digit percentage from our current base of 130 million DAUqs.”

Read more

AMZN -11%: AWS +39.3% beat Street (37.7%) but missed the ~43% buyside bar, with 38.7% OM vs. the ~40% bogey, and the 4Q guide’s high ends ($247B revenue / $33.5B OI) were light vs. the ~$247-248B / ~$34-35B bogeys, with both midpoints under Street. 3Q cash capex of $75.9B took TTM FCF to -$42.4B; the offset is Jassy projecting AWS growth of “over 40% both in 2027 and in 2028”.

Crowded long so bulls not happy tonight and today’s -4% only took a bit of air out; a 39% AWS with OM down q/q isn’t what longs who wanted 43%+ and a 4-handle were looking for. Bulls get the 40%+ AWS outlook for 2027-28, bears get a $76B capex quarter, an ads miss and the funding question.

• AWS beat Street but missed the bogeys at $46.0B, +39.3%, vs. Street $45.45B (+37.7%) and bogeys around ~$47.2B (~43%, range 42-44%).

  • That’s a +$3.75B q/q add vs. the record +$4.6B in 2Q, and a 2.6pt accel from 2Q’s 36.7% on a 20.2% comp; 2-yr stack 59.5 vs. 54.2 in 2Q.

  • AWS OM 38.7% ($17.8B OI), ~136bps above Street’s 37.3% but ~135bps below the ~40% bogey and ~70bps below 2Q’s 39.4% (which had a ~$0.6B energy-derivative gain; ~38.1% ex).

  • Incremental OM 49% y/y, but only 31% q/q.

  • Run rate $184B vs. $169B in 2Q. Backlog and the AI / chips run rates are not in the release (2Q: $496B backlog; “over $25 billion” each).

  • The 2027-28 projection vs. Street: FY27 Street AWS $236.0B is ~+35% on ~$175B FY26 (YTD $125.8B + Street 4Q $49.3B); 40% on that base is ~$245B, ~$9B above Street.

  • New AWS agreements with META, NVDA, OpenAI, CRM and ORCL among others; Bedrock token volume up nearly 17x y/y.

• Capex ran $75.9B in 3Q (cash, net of proceeds, +122% y/y; $77.3B gross) vs. Street $58.2B on the same net basis and $53.1B in 2Q.

  • “We continue to expect 2026 cash capital expenditures to be approximately $220 billion.” YTD is $172.2B, which implies ~$47.8B for 4Q vs. Street $64.8B.

  • The unpaid-P&E build slowed to +$0.9B from +$10.7B in 2Q, so more of the spend hit cash this quarter.

  • FCF was -$33.5B for the quarter (OCF $42.4B less $75.9B) vs. Street +$1.3B; TTM FCF -$42.4B vs. -$7.6B in 2Q.

  • Funding so far is debt: $48.1B of long-term debt raised in 3Q (LT debt $173.1B vs. $65.6B at YE25). Also $24.0B of acquisitions / non-marketable investments in the quarter ($63.8B YTD).

  • Nothing on 2027 in the release; remember bogeys around ~$325-350B vs. Street ~$287B.

• Revenue missed at $201.9B, +12.1% (13% ex-FX), vs. Street $202.35B and ~$202.5-203.5B bogeys, just under the $202.0B high end of the $197-202B guide.

  • Ex the Prime Day shift the company says growth would have been ~400bps higher; 2Q+3Q combined +15.7% vs. Street 15.8%.

  • North America $112.9B (+6.2%) missed Street’s $113.8B mean (+7.1%) but was in line with the ~6-6.5% bogey and the $113.2B median; 2Q+3Q combined NA growth 11.0% vs. ~12% in 1Q.

  • International $43.1B (+5.4%, +8% ex-FX) in line with Street $43.0B.

• Ads missed at $19.6B, +10.9% (12% ex-FX), vs. Street $20.9B (+18%); 2Q+3Q combined +18.1% vs. Street 21.9%.

  • The release blames “timing anomalies primarily related to Prime Day (and a couple other timing factors)” and expects 4Q ads to “grow in the high teens” (Street +18.7%).

  • Keeps the agentic-ads debate front and center; Alexa for Shopping was used by 400M+ customers in the last 12 months, interactions +570% y/y.

• EBIT beat at $28.1B / 13.9% OM vs. Street $26.2B, the $22.5-26.5B guide and ~$27.5-28B bogeys; a record OM, +~21bps q/q.

  • That’s ~$1.6B above the guide high end and +61% y/y vs. $17.4B (a base hit by the FTC settlement and severance charges).

  • NA OI $8.7B (7.73% OM) vs. Street ~$7.7B (6.8%), ~93bps above; International $1.6B (3.67%) vs. ~$1.8B (4.1%), ~42bps below.

  • GAAP EPS missed at $1.89 vs. Street $1.97 despite the OI beat: non-op swung to -$1.2B (interest expense $1.69B vs. $0.54B a year ago, other expense $0.7B), with no Anthropic gain vs. $9.5B a year ago.

• 4Q guide light vs. the bogeys, midpoints under Street: revenue $240.0-247.0B (+12-16%) vs. Street $244.4B and ~$247-248B at the high end; OI $28.5-33.5B vs. Street $32.9B and ~$34-35B at the high end.

  • Revenue midpoint $243.5B = +14.1%, a 2pt accel vs. 3Q’s reported +12.1% (Prime Day-distorted) but below the 15.7% 2Q+3Q combined pace, on a 13.6% comp; 2-yr stack 27.7. FX ~60bps headwind.

  • The ranges widened to $7B on revenue and $5B on OI (vs. $5B / $4B for 3Q).

  • OI midpoint $31.0B = +24% y/y vs. $25.0B; implied FY26 OI ~$110.4B.

INTC -18%: 3Q beat Street on revenue, GM and EPS, with 44.1% GM and $0.52 EPS clearing the bogeys and DCAI +78%, but the Dec guide whiffed: revenue of $15.0-16.0B and 38.0% GM vs. $16.98B / ~42.5% Street and ~$17.7-18B / ~44% bogeys, with the CFO seeing the PC slowdown run at least through 1H27. Crowded long, so LBT has a lot to explain on the call.

Crowded long coming in (even after the ~28% pullback from the 9/24 high), and the buyside was well above Street on Dec, so a guide down 6% q/q with GM back to 38% is a big disappointment. Bulls will point to DCAI, but the PC slowdown is the story tonight.

Read more

TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.

Follow OpenAI, NVDA, AAPL or Anthropic on Ethmos to get a brief whenever their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.

Today’s Essential Episodes:

1. OpenAI CEO Sam Altman on OpenAI’s Own Chip, Robots Building Data Centers and Why He Wants a Trillion AIs Over One Superintelligence
   Cheeky Pint, with John Collison, co-founder and president of Stripe, and Sam Altman, CEO of OpenAI

2. Silicon Data’s Steve Hou on Why GPU Rents Keep Rising, the Compute Forward Curve and What Anthropic’s S-1 Says About Margins
   Monetary Matters with Jack Farley, with Max Wiethe, and Steve Hou, head of research at Silicon Data

3. Mark Gurman on the iPhone Duo’s Hinge Defect, the Touch-Screen OLED MacBook Pro and Apple’s M&A Shake-Up
   TBPN, with John Coogan and Jordi Hays, and Apple reporter Mark Gurman

4. SemiAnalysis’s Joey Brookhart on Anthropic’s S-1: the Inference Margin Flip, the $80 Billion Loss and Who Pays for Frontier AI
   Decoder with Nilay Patel, and Joey Brookhart, an analyst at SemiAnalysis

5. Dora Partners’ Tony Brough on Gas Turbines Sold Out Into 2032, Several-Fold Price Hikes and the Scramble of New Suppliers
   Catalyst with Shayle Kann, and Tony Brough, president of Dora Partners & Company

^{Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.}

Listen · Watch · Follow OpenAI on Ethmos

THE SIGNAL

• Collison disclosed AI tokens are roughly 1% of Stripe’s OpEx, which he called too low and expects to reach 5, 10, then 20%; Altman agreed many companies will settle at tens of percent of OpEx.

• Altman said DRAM makers historically spent about $20B of capex to earn about $19B, but now each may generate around $100B of free cash flow a year, and bankruptcy memories keep capex conservative, prolonging the shortage.

• Altman argued local inference wastes scarce DRAM, since PC memory is used about 5% of the time versus high cloud utilization; (our read) this pushes back on the idea that on-device AI eases data-center memory demand.

• Altman floated OpenAI possibly selling its own chip, which he said is strong on intelligence per watt; he called tape-out design a good task for AI and said only one company has mastered leading-edge lithography.

01 PHP vulnerabilities and the security overhang

Altman explains that “PHP week” emerged because models have been quietly improving at finding real-world vulnerabilities for a while, but only recently have people systematically pointed them at the vast surface area of existing code, which he expects to reveal far more serious vulnerabilities than anyone anticipated. Rather than just patching PHP, he and Collison discuss whether entire software architectures need rearchitecting — a shift Altman says would have seemed impossible two years ago but now looks achievable as startups build entirely new technology stacks using AI.

On the cyber capabilities question, Altman is self-critical about OpenAI’s public messaging, arguing the field has failed to explain why AI’s risks are worth the tradeoffs, likening weak industry messaging to a bad sales pitch. He defends OpenAI’s “iterative deployment” philosophy — pushing capable tools into the world early even when unpopular within the AI safety community — as the only workable path to avoid dangerous capability overhangs. On early results from the cyber initiative, he says the first days went better than his worst fears, with companies using models for rapid patching and defense, though he’s careful not to declare victory, and notes cyber experts are no longer confident their hardware-exploit threat models hold.

02 Sandboxing, alignment, and the case for swarms

Altman lays out a staged view of AI safety: sandboxing works to a point, then monitoring (including chain-of-thought and activation monitoring) works further, but beyond that, safety requires near-total confidence that a model is faithfully aligned with operators’ intent and values human control of the future indefinitely. He is explicit that a model dramatically smarter than the best security researcher raises containment questions he doesn’t have answers to.

As an alternative, Altman is drawn to agent teams with competing incentives as a promising safety direction, arguing they mirror the resilience properties of pluralistic, democratic societies rather than a single “brain.” He states he would much rather have a trillion AIs of slightly superhuman intelligence join society than one AI of a trillion IQ, presenting this as his own preference and an open research direction rather than a solved framework.

03 Compute, AI-driven research, and token budgets

Discussing test-time compute, Altman traces a rapid historical arc: a few years ago throwing more than roughly a dollar of compute at a problem yielded no benefit, but chain-of-thought reasoning unlocked productive use of massive parallel compute, enabling today’s agent teams to run thousands of copies simultaneously. He says the gap between elite AI users (advanced math, hacking) and typical corporate usage keeps widening because most organizations haven’t adopted swarm-style compute spending.

Internally, OpenAI avoids rewarding high compute usage — there are no token leaderboards, and topping one would not help a researcher’s career — instead giving researchers wide latitude unless usage looks abusive. On Navier-Stokes specifically, Altman estimates the proof cost around $10 million, with related math work totaling hundreds of millions of dollars, and predicts OpenAI could redo it a year after the original proof at roughly a hundredth of the cost, with smarter models and what it has learned about scaling agent teams (model size, copy count, rollout length). He acknowledges mathematicians’ unhappiness about AI solving open problems but frames it as ultimately good for the world, while admitting personal wistfulness and even sadness watching models rapidly solve problems that previously defined mathematicians’ careers.

On recursive self-improvement, Altman says OpenAI doesn’t yet have a codified method for AI-driven AI research beyond asking for concrete things like a modest compute-efficiency win — discovering something as foundational as the transformer architecture remains unsolved, and he expects to learn much more about optimal swarm usage in the coming months.

04 Compute shortage, chips, and robotics-built data centers

Altman argues the “invisible hand” is currently failing to resolve compute shortages because memory makers, scarred by historical boom-bust cycles, remain conservative on capex despite now-enormous profits. He says the solution requires new entrants and supply-chain innovation — citing new actuator makers for industrial robots and OpenAI’s own chip effort, which he says targets strong performance per watt of intelligence. He frames OpenAI’s robotics program as eventually enabling self-replicating data centers (robots building fabs, turbines, solar panels, and data centers), which he says gets far less attention than the usual picture of recursive self-improvement as AI doing AI research.

On cost targets, Altman takes $70 billion per gigawatt as an illustrative five-year total cost of ownership today (estimates vary, he notes) and says he’d like to see that fall toward $20 billion through cheaper energy, chips, and supply-chain margin compression, describing himself as pro-deflation.

05 Personal agents, agentic commerce, and pricing

Altman says personal agents (referencing OpenAI’s Dots and competitors) haven’t yet had their breakthrough “Macintosh moment” — people like them but aren’t yet dependent on them — though he believes that moment is close. Both he and Collison argue businesses resisting agentic interactions (preferring their own apps and forms) are mistaken, drawing a parallel to the iPhone app-store era when companies reluctantly ceded distribution control; Altman says he’d choose Amazon through ChatGPT almost every time regardless of interface, citing trust in fulfillment over UI.

On pricing, Altman says OpenAI nearly stopped pricing its next model in tokens, quoting a price per prompt and billing like AWS, but customer reaction was strongly negative; people he asked would rather pay more in a consistent unit than get an opaque price. He also distances OpenAI from “bench-maxing,” claiming the company is currently better than peers at avoiding benchmark gaming, while admitting it hasn’t always been.

Listen · Watch · Follow NVDA on Ethmos

THE SIGNAL

• Farley says Anthropic’s widely reported S-1 loss is mostly a non-cash GAAP mark on a convertible note issued to Amazon; netting cloud partners’ cut, he puts his implied gross margin (training excluded) nearer 82-83% than 88%.

• Farley thinks Google Cloud’s growth includes TPUs sold to Anthropic through a Broadcom lease structure described in the S-1, and calls cloud clients extraordinarily concentrated in OpenAI and Anthropic.

• Hou says SK hynix’s enormous latest quarterly profit still earns it a lower multiple, and ties that and CoreWeave’s and Nebius’s weakness despite surging rentals to market doubt that these earnings can last.

• Farley says Oracle’s investor day hinted at favoring free cash flow over capex, and that a round of big spenders not raising capex, Microsoft included, would make it a different environment for AI hardware.

01 GPU rental surge and the crunch spread

Hou presents Silicon Data’s core chart tracking a token price index against GPU rental rates for H100, B200 and B300. The token index has drifted sideways-to-down since late May — which bears read as commoditization and deflation — while rental rates for the most advanced chips have risen sharply in recent weeks. Hou notes that within the token index, open and proprietary model sub-indices haven’t actually fallen much individually; the headline decline reflects substitution toward cheaper open models, even as enterprises (per Anthropic’s S-1) show strong adoption of pricier frontier models like Opus. He says he believes Anthropic disclosed slower Q3 revenue growth at a recent investor day, though he cites Ramp data suggesting a reacceleration tied to Opus’s popularity.

02 Hyperscaler capex and profitability signals

Farley and co-host Max Wiethe note Google and Microsoft both reported strong cloud growth but came in under capex expectations for the quarter, raising questions about whether the focus is shifting from capex to ROI. Wiethe relays a listener’s note from Alphabet’s call that Google held its capex guidance unchanged, still expecting a significant increase in 2027. Farley says Alphabet’s miss came in search, while its cloud business grew far faster and more profitably — framing why executives remain enthusiastic about AI investment. Hou argues capex and stock performance are linked: companies won’t keep raising capex if markets don’t reward the spend. Both agree visibility is poor — physical constraints like power and land, not demand, are described as the current bottleneck, with investment numbers described as reaching a scale equivalent to fractions of GDP, even top executives can’t reliably forecast years out.

03 Rates, credit spreads, and the Fed

Asked whether borrowers can bear higher rates after Fed Chair Warsh welcomed tightening financial conditions, Hou argues that modest rate moves are unlikely to deter AI investment on their own; the real risk is if the Fed concludes AI-driven demand is itself inflationary and reacts aggressively to slow it. On the surge in AI data-center bond yields (Farley cites CoreWeave at up to 15%), Hou attributes the widening credit spreads not to higher base rates but to the bond market’s persistent doubt that AI investments will deliver the outsized returns bulls expect.

04 Compute futures and the forward curve

Read more

Outside of earnings, Semis -2.2% with some pointing to the lower than expected headline capex #s from GOOGL/MSFT although q-specific reasons for each of those. META +1% getting some love after the GOOGL miss. Neoclouds lower with NBIS -7% as some speculating the MSFT 3p capacity delay is them. LITE +5% as MRVL -11% said to be in advanced talks to buy them. CART +5% (halted) as DASH to acquire them for $65/share. Samsung -4% after an operating profit miss and delaying capital return announcement until end of Jan.

Another busy night tonight with AMZN the big one. We dive into GOOGL, MSFT, KLAC, FTNT, and NOW below, then the usual.

Let’ get to it…

EARNINGS

GOOGL -11%: Just an overall bad print on the 3 KPIs that mattered most: search, GCP & EBIT. Search missed at +12% vs. Street’s 14.7% and the ~13-14% buyside bar, Cloud ($30.7B) beat Street but fell short of $31-32.5B bogeys, and EBIT missed by ~$2.8B. YT miss. Cloud EBIT beat the one bright spot although that’s being debated as well.

Stock was shorted coming in but this was worse that expected and helping feed bearish investor fears that search is AI hampered. Comps continue to get tougher for the next couple quarters and just running the decel forward gets you to something like 7% search growth in Q1. Lots of talk from bears in Slack saying downside is now 15x unless they can show evidence of monetization. AI Mode is diluting Search monetization (paid clicks +8% on five-year-high query growth) as Google struggling to figure out how to monetize AI. Sundar and Philipp really didn’t say much to help things on the call. CY27 Buyside EPS coming down to a more street ineline-ish # around ~15-$15.50 although bears will frame an even lower #. A 20x gets you <$290, which is where stock is at while bears will push for a teens multiple given the search decel & fears. Remember this is a company where Total Ads are still 65% of the topline (and ~80% of the EBIT) so the Cloud not really big enough to make up for search fears. Uphill battle for Google longs now. R

META +1% as its looking a lot better in comparison after the print, with 70% of investors in Slack preferring to pay 18x for META vs. GOOGL and pointing to META’s better ROI:

Lots of Bearish Investor feedback overnight in the Slack earnings thread:

“Lower monetization on AI queries. He basically just admitted it’s going to be a transition between AI Overviews and search... Search going off patent.”

“AI Overviews / AI Mode are cannibalizing traditional search queries and are not really being monetized, at least not even close to traditional search levels. They gave a lot of stats on AI improving their ad engines, but it just seems like the monetization layer for these cannibalizing modes isn’t there yet. AI search ad monetization does seem harder to crack than social.”

“Meta is proof that LLMs can improve ad performance materially. We’ve had ~2 years straight of Meta growing CPMs and impressions at the same time. That never happened before. So Google blaming lower monetization just feels wrong. Q/q: Meta’s Family of Apps +$3.8B, Search +$79M. Zero reason to own this stock over Meta.”

“8% paid click growth y/y... in line with my backsolve, still decel on 2y/3y stacks.”

“This will derate to 18x. Underwriting $15 was also dependent on Search holding up better, and $15 might be high given the opex trend. 15x feels like the near-term floor, but that’s a big haircut from here.”

Read more

Positioning had come down last few days following concerns over the VA contract headwinds, but this is a bit worse even taking that into account. Some nascent bulls will be disappointed.

3Q subscription revenue of $3.988B (20.5% cc, +21% reported) was right on the $3.99B Street and below the ~$4.01–4.03B bogeys, a 50bp cc beat on the 20% guide vs. the ~120bp average beat of the last four quarters. cRPO of $13.65B grew 21% cc (20.5% reported) vs. ~21.5–22% bogeys, the 20% guide and $13.57B Street. Ex ~1pt Moveworks and ~1.25–1.75pt Armis that’s ~18.5% organic vs. the ~19–19.5% bulls were looking for. Net new cRPO of $450M q/q vs. $430M a year ago. RPO of $29.7B (22.5% cc) vs. $29.4B Street.

Op margin of 32.5% was ~140bps above Street’s 31.1% and 0.5–1pt above the ~31.5–32% bogey (31% guide). EPS of $1.00 vs. $1.03 Street despite the OM beat: net interest swung to about -$12M from +$109M a year ago (interest expense $81M vs. $6M). Subs GM of 80% was ~65bps below Street’s 80.65% and down 300bps y/y.

4Q subs guide of $4.245–4.255B (21.5% cc, +22.5–23% reported) vs. $4.24B Street and ~$4.24–4.26B bogeys. The midpoint is a 1pt cc accel from 3Q on a 1pt easier comp (19.5% cc in 4Q25), so the 2-yr stack is flat at ~41%.

4Q cRPO guide of 18.5% cc vs. ~19.5–20% bogeys; 18.5% on the $12.85B base is ~$15.23B vs. $15.29B Street. That’s a 2.5pt decel from 3Q on a tougher comp (21% incl. Moveworks), with the 2-yr stack down to 39.5% from 41.5%. Moveworks’ ~1pt laps at YE while Armis stays in, so the organic 4Q guide is ~17–17.25%.

FY subs goes to $15.78–15.79B from $15.76–15.78B, a $15M raise at the midpoint vs. a ~$10M 3Q beat over the guide midpoint ($3.9775B). That passes through the full beat but lands below the ~$15.80–15.83B bulls wanted. The new FY implies ~$4.25B of 4Q subs, the 4Q guide midpoint.

4Q OM guide of 32% vs. Street 33.5%. FY OM held at 31.5%, so the 3Q margin beat comes back out of 4Q. AI ACV target raised again, to $1.7B from “beat $1.5B” (it started the year at $1B). FY FCF margin goes to 36% from 35% on lower capex; FY subs GM held at 81% after two cuts this year. Large deals were strong: 19 transactions over $5M in net new ACV (+135% y/y) and 680 customers over $5M ACV (+23%).

EBAY +4.6%: Clean beat-and-raise that clears the bogeys on GMV, revenue, OM and EPS, and the 4Q GMV guide (+12-14% FXN) sits well above the ~10-11% bogey; ads were the soft spot and there’s no 2027 color on the release, so the margin framework on the call will be the focus for the HF shorts.

Read more

Stock was shorted coming in, but this is worse than expected - Sundar has an uphill battle on the call to explain this away…

Read more

TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.

Follow Gavin Baker, SHOP, Anthropic or INTC on Ethmos to get a brief whenever they or their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.

Today’s Essential Episodes:

1. Atreides’ Gavin Baker on Anthropic’s S-1 Surprise, GPU Prices Flipping Into Contango and the Capital the SpaceX and Anthropic IPOs Unlock
   TBPN, with John Coogan and Jordi Hays, and Gavin Baker, CIO of Atreides Management

2. Shopify CEO Tobi Lütke on Embracing Meta’s Muse as Amazon Blocks It, Agent Shopping and Choosing Volume Over Take Rate
   Sources with Alex Heath, and Tobi Lütke, co-founder and CEO of Shopify

3. Anthropic’s Katelyn Lesse and Angela Jiang on Opus 5.5, CPU as the New Bottleneck and Rewriting the Messages API in Rust
   Latent Space, with Katelyn Lesse and Angela Jiang, who head engineering and product for Anthropic’s Claude Platform

4. Intel Foundry’s Ravi Mahajan on EMIB’s Path From Science Fiction to High Volume, Why HBM Came First and How It Compares With CoWoS
   Semi Doped, with Austin Lyons, and Ravi Mahajan, distinguished engineer and director of advanced packaging pathfinding at Intel Foundry

5. Ben Bajarin and Jay Goldberg on Micron’s HBM Roadmap, Lam’s $200 Billion WFE Call, Meta’s Owned-Compute Edge and NVIDIA’s Supply Lock-Up
   The Circuit, with Ben Bajarin of Creative Strategies and Jay Goldberg

^{Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.}

Listen · Watch · Follow Gavin Baker on Ethmos

THE SIGNAL

• Bear case Baker credits to an analyst’s post: Anthropic devoting 74% of its compute to training, so any drop in training intensity could release a large block of capacity onto the inference market and weaken the tight compute-pricing thesis.

• Cursor used to be a major demand driver for inference clouds, and SpaceX has since moved Cursor’s compute in-house, yet Baker says the open-weight ecosystem still sped up and diversified, which suggests less single-customer concentration risk (our read).

• Baker puts compute contracted before May 2026 at roughly $10-15 per watt, now deep in the money; he says CoreWeave is actively chasing more spot exposure and fewer big offtakes, a possible repricing lever (our read).

• Baker reads the blank October ARR line in the S-1 as a sign that management expects a strong number; he notes OpenAI accelerated over the summer and Anthropic clearly reaccelerated in October.

01 Anthropic’s S-1 and enterprise diffusion

Baker, speaking as an investor at Atreides Management, describes the S-1 as a clear positive surprise, pointing to the $10B cumulative burn, profitability, and cash generation as better than feared. He cites an annualized revenue figure he reads as roughly $67B off the reported Q3 run-rate. Despite this scale, he argues enterprise AI penetration remains very early — possibly earlier-stage than he’d previously believed — given how small AI spend is relative to total enterprise budgets even among large adopters. Hosts Coogan and Hays push back with the commonly cited Ramp data suggesting a small number of companies drive most AI spend; Baker counters that the S-1’s own numbers (250 accounts over $10M, by his reading; a host cited ~11-12K over $100K) show a genuinely diversified customer base, a claim he treats as more authoritative than secondary data sources. He also notes open-weight model providers — Fireworks, Together, Baseten, Cognition — have been growing faster than either OpenAI or Anthropic since summer, a trend that held through Anthropic’s recent reacceleration; Atreides led Fireworks’ last funding round, which he says was hard to price because the company was growing so fast.

02 Compute pricing and the CoreWeave contrast

Baker’s central market claim is that GPU/compute pricing has moved into contango: spot prices are rising, but forward prices for clusters delivered in 2027 are rising even faster, meaning buyers pay substantially more to lock in future capacity than current capacity — a shift he attributes to remarks at GTC by the founder of Silicon Data, a firm Atreides has invested in and which he calls the most authoritative source. He says portfolio companies that secured Blackwell clusters for mid-2026 are shocked by the pricing demanded for a comparable cluster in September 2027, despite broad awareness that new supply is coming online. He contrasts CoreWeave — which a host described as reliable “diamond tier” infrastructure, and which Baker says locked in long-term compute contracts now priced well under market — with SpaceX’s more spot-market-exposed strategy. Separately, Baker cites CFO Sarah Friar saying her own company’s long-term compute commitments could now be resold at a 3-5x premium. Baker speculates Elon Musk reasoned from first principles toward shorter, spot-friendly contracts with fast payback rather than copying any peer. He also flags NVIDIA’s new DSX pod (announced at GTC Berlin) — a preassembled, rack-based deployment — as a meaningful innovation because it could shrink the typical three-to-six-month GPU “energizing” lag, freeing up working capital that would otherwise sit idle.

03 AI agents, biology, and commodities

Discussing personal AI agent use, Baker says Grok Bot has been the most impactful AI product for him and his firm, describing roughly 17 custom Grok Bots he has built, after earlier stitching tools together with Claude Code. On biology, he notes AI-assisted drug repurposing has already identified uses for decades-old compounds to treat rare diseases, though he concedes commercialization timelines remain long and rates the AI industry’s PR standing at roughly negative five out of 100. On commodities/energy, Baker says his firm tracks natural resources and energy closely but hasn’t made it a large portfolio allocation, aside from a private investment in a company developing a cleaner chemical process for rare-earth refining, an area he says DARPA has also pursued.

04 Financing the AI buildout

Baker identifies an underappreciated capital unlock: U.S. mutual funds are capped on how much of a fund can sit in illiquid assets, so SpaceX and Anthropic going public will free up significant mutual-fund buying power for the broader private AI ecosystem once lockups expire. He also points to sovereign wealth funds — particularly UAE-based MGX and Mubadala — as already heavily engaged, noting Atreides and Valor co-led a Crusoe round with Mubadala. Looking to 2027, Baker expects accelerating hyperscaler and neocloud revenue growth, converted into high-margin operating cash flow, to finance a larger share of the infrastructure buildout than before — suggesting peak reliance on external financing may be near. On public sentiment, a host argues that as OpenAI and Anthropic shares reach millions of holders (rather than maybe 50,000 people today), AI-related sentiment on platforms like X may become less polarized, and Baker praises Musk’s, Michael Dell’s, and Brad Gerstner’s practice of distributing company equity broadly as preferable to routing wealth through foundations.

Listen · Watch · Follow SHOP on Ethmos

THE SIGNAL

• Asked if agents let Shopify raise its take rate, Lütke said he would rather charge a lower percentage on more volume, managing to gross profit dollars, not margin; this caps agentic take-rate upside (our read).

• Agentic order volume started this year from a low base, so its growth rates mean little, Lütke said, but agents increasingly drive considered, high order value project purchases, which points to AOV upside before volume does (our read).

• Lütke says more Shopify stores are opening than ever, and skewing younger; Shopify is working so AIs can sign people up, with costs amortized across its billion-dollar merchants.

• As a Meta board member, Lütke called Meta’s ad engine the greatest boon to small businesses ever, with Google’s slightly less so; agents threaten paid discovery unless trust-compatible formats emerge (our read).

01 Shopify vs. Amazon on agents

Lütke says he learned of Amazon blocking Muse only after Shopify had already announced it was welcoming the agent, calling the contrast cheeky but not new. He describes Amazon as representing a digitalized department store built on inventory depth rather than storytelling, while Shopify represents boutique-style entrepreneurship. He attributes Shopify’s openness to agents partly to having no advertising business to protect and to a deliberate choice never to build a traditional three-sided marketplace, since marketplaces force an operator to choose between buyers and sellers; Shopify instead sides with merchants throughout the stack. He notes Shopify has also announced partnerships with other agents, describing the company as serving small businesses wherever they sell — social channels, point-of-sale, craft fairs — not just its own storefronts.

02 Early agentic shopping behavior

Read more

In Tech, Semis +1% continuing to rally on the back of Korea’s KOSPI’s 7% gain. Overall, earnings mixed overnight in the space with SKHY, VRT, GEV, FLEX, APH all betterand STX, NXPI, TER, FFIV mixed. SNDK/MU +4% on the back of Hynix’s solid print that came above buyside bogeys on both #s and capital return (Kioxia +14% /Samsung +4% overnight). Software -1% with NOW -2% leading the way lower ahead of their print tonight (Bernstein put a damper on bulls hopes for a big cRPO print yesterday calling out a 80-100bps headwind from a VA contract renewal).

Asia mainly up: TPX -0.14%, NKY +1.08%, Hang Seng +0.23%, HSCEI +0.32%, SHCOMP +0.91%, Shenzhen +1.98%, Taiwan TAIEX +2.04%, Korea KOSPI +6.63%

Another busy day in Tech today was we get ORCL’s analyst day near noon, MNDY analyst day at 2pm, AAPL product event at 1pm, and Fed at 2pm. Then after-hours, we get MSFT, GOOGL, NOW, EBAY, CVNA, FTNT, and others…

Lots to get to today. We’ll hit earnings first: SKHY, STX, NXPI, VRT, GEV ETSY then onto the usual. (Click link to go straight to the relevant section)

Let’s get to it…

SKHY +7%: ASP-Driven Beat, Mid-20s 4Q Bit Guide, “Sold Out Through 2029” and a FY26 KRW100T Capital (10% of mkt cap) Return Working to take on the the Peak-Cycle Bear Case

Overall, looks really solid across the board with revs and OP beating street and bogeys and better capital return plan. 3Q revenue of KRW104.6T (+32% q/q, +328% y/y) and OP of KRW83.3T (~80% OPM, a record) beat Street by mid-to-high single digits on a high-20s% DRAM ASP lift; EPS missed on KRW32.6T of non-operating FX and investment losses. Bogeys were around KRW 102-103T for revs and mid to high 70s for OP so this is much better.

The beat was all price (DRAM ASP +high-20s%, NAND +mid-20s% vs. models in the mid-teens to low-20s%), but the bigger number is 4Q: a +mid-20s% q/q DRAM bit guide that puts 4Q revenue/OP ~10-20% above pre-print Street.

Mgmt sounded good on the call: LTAs now cover >50% of ‘27/’28 capacity across 14 customers (in July mgmt wouldn’t size coverage), DRAM/HBM is “effectively sold out” through 2029, and 2027 HBM volume and price are locked, with the HBM-vs-commodity margin gap set to narrow “meaningfully.”

FY26 shareholder return of ~KRW100T (~US$75B, ~10% of mkt cap) topped the buyside bogey. Local shares closed +~11% but still sit ~50% below June highs at ~2-3x ‘27 EPS. Should help with memory sentiment especially as semi investors have looked to put new $ to work this week and with analog/HDDs not looking as good after NXPI/STX. MU +4% / SNDK +5% / Kioxia +12%

Key Takeaways

• LTAs. The most incremental disclosure. In July, mgmt said it couldn’t say how much of sales LTAs would cover; now it’s 14 key customers (vs. ~10 last q) incl. major global CSPs, covering >50% of 2027 and 2028 capacity, with similar coverage targeted over the medium-to-long term. Five-year terms and deposits were flagged in July; new are the coverage, the customer count, extension talks with some customers and a more specific pricing structure (floats with market conditions inside a price range agreed with each customer). Adding volumes under discussion based on customers’ concrete purchase plans, DRAM and HBM are “effectively sold out” through 2029. Non-LTA capacity is deliberately kept open for spot upside and next-gen products.

• Capital Return. FY26 return of ~KRW100T: ≥KRW30T of dividends and ≥KRW60T of buybacks, incl. August’s KRW40T program (completed last week, to be cancelled). That’s FY26 alone, not a multi-year figure, despite some headlines. 3Q: KRW20K DPS (KRW10K regular + KRW10K Kioxia special) = KRW14T, plus KRW20T of share purchases, half cancelled and half for exec/employee comp (excluded from the KRW100T). That implies ~KRW35T more in 4Q, incl. the rest of the Kioxia special in the year-end dividend. The bogey was ~KRW30T on top of August’s KRW40T (Korean press had floated KRW50-60T). Buybacks continue quarterly, sized on price and valuation, and mgmt said outright the shares are undervalued. Framed as delivering on the >50%-of-FCF commitment; no framework yet beyond FY26. Solidigm’s IPO review was dropped given market volatility.

• HBM. 2027 HBM volume and pricing are finalized with major customers and reflect HBM4’s higher value. Mgmt expects the HBM-vs-conventional DRAM profitability gap to “narrow meaningfully” (i.e., HBM currently earns less than commodity DRAM after this year’s spike) and HBM revenue and profit to rise significantly in 2027. Updated models have blended HBM ASPs +~55-60% next year (HBM4 at the main customer ~US$3.3/Gb vs. ~US$2.1 today). Execution: HBM4 entered full-scale MP in 3Q and passes HBM3E in bits and revenue in 4Q; HBM4 yield beat plan (target raised) and HBM3E cleared its mature-yield target; M15X is pulled in to full capacity by YE; HBM4E sampled in 2Q, with custom HBM projects underway. Incremental context: a key bear argument behind SKHY’s recent underperformance vs. MU/Samsung was that its Rubin HBM4 allocation would shift to Samsung on pin speed. NVDA relaxing the Rubin spec (to ~10.6Gbps from 11.7Gbps per one report; chatter says lower), plus today’s yield commentary and a locked 2027 book, help defuse it.

• Stack Count. Mgmt framed customers moving to lower stacks (e.g., 12-high HBM4 to 8-high HBM4E) as a symptom of shortage, not a demand negative: customers want more cubes from constrained supply, higher per-die density offsets much of the bit impact, more cubes means more base dies and back-end work, and customer requirements remain >2x what SKHY can supply. HBM shipment plans are unchanged, and HBM profitability should stay far above prior-cycle peaks.

• Pricing. DRAM ASP +high-20s% q/q (vs. ~+30% last q) on conventional DRAM strength and a richer HBM4 mix; NAND ASP +mid-20s% (vs. +mid-50s% last q) on broad increases and eSSD mix. Both beat models (mid-teens to low-20s%) despite a stronger won. Server DDR5 shipments rose 40%+ q/q, and server is ~55% of DRAM revenue by one estimate. Increases are decelerating, NAND especially, but off a base where server DRAM prices are up ~200% YTD.

• AI Demand. More data-driven than July’s “solid beyond next year” framing: hyperscaler cloud revenue +30-80% y/y, cloud backlogs 2-5x y/y, paying users rising; 2027 global CSP capex +60%+ with secured power +100%+ (third-party estimates); AI DC infrastructure ~40% CAGR through 2030. New argument: the buyer base is broadening (enterprise, sovereign, on-prem), so one buyer’s pause gets backfilled by others. PC/smartphone units will fall on memory costs and shortages, more than offset by AI. Supply stays short of demand through 2030.

• Supply. 2027 capex will rise “significantly” (leading-edge mix, Yongin Fab 1 ramp, domestic and overseas infrastructure) vs. a ≥KRW40T 2026 plan (~KRW31T spent through 3Q). The oversupply pushback got more specific than July’s “phased ramp” language: more fabs ≠ more bits (HBM and process complexity eat resources; shrinks yield fewer bits), Yongin/M17 shells accelerated but cleanrooms opened in phases with tool move-ins tied to demand, and, new, SKHY vets customers’ financial health and funding capacity before committing capacity, a clear nod to the neocloud-financing worry. On tech, SKHY will stretch 6F² via EUV scaling before moving to 4F² or 3D DRAM (CXMT recently touted 4F² progress).

• NAND. Bits slipped slightly vs. a +LSD% guide as SKHY prioritized 321L TLC eSSD; ASP +mid-20s%. 4Q bits +MSD% as 321L passes 50% of domestic capacity. KV-cache tiering is the new eSSD driver (SALT-KV software and CXL-based CMM-Hybrid), with HBF longer term.

• Below the Line. The EPS miss is entirely non-operating: a KRW32.6T net non-op loss incl. a KRW11.6T FX loss (strong won) and KRW22.9T of investment valuation losses, partly reversing 2Q’s large investment gains. Net income KRW39.8T (38% margin). The Street will look through it, but reported EPS will stay noisy.

Bull vs. Bear Debate

Bulls argue more broadly that memory is no longer a classic commodity cycle but a capacity-constrained input to AI infrastructure. Agentic and long-context inference raise memory intensity at every tier at once: HBM on accelerators, server DDR5 and SOCAMM on CPU head nodes, and eSSD for KV-cache offload. Supply can’t respond quickly: HBM consumes roughly 3x the wafer per bit of standard DRAM, shrinks now deliver fewer bits per node, cleanrooms take two-plus years to build, and little non-China NAND capacity is being added. SKHY is the purest way to own it: the HBM share leader with the highest AI revenue exposure of the big three (~70% by one estimate), first-supplier status at the key accelerator customer and industry-leading yields. The LTAs are what bulls think the market still underprices: 14 customers, five-year terms, deposits and pricing that floats inside agreed ranges, covering more than half of 2027-28 capacity. That moves SKHY from a spot-priced commodity toward contracted infrastructure revenue and should shrink the peak-earnings discount over time, the way take-or-pay structures support higher multiples in other capital-intensive industries.

This quarter delivered most of what bulls needed. The beat came from price, and the +mid-20s% 4Q bit guide puts 4Q roughly 10-20% above pre-print Street before any further price increases. The 2027 HBM negotiation, the biggest swing factor in ‘27 numbers, is closed and reprices HBM up toward commodity profitability, which is why updated ‘27 OP models coming out ~5-25% above Street and buyside #s moving up, but a bit less. The HBM4 overhang behind October’s underperformance looks resolved: yields beat plan, HBM4 crosses over in 4Q, NVDA relaxed the Rubin spec and SKHY’s 2027 allocation is locked. Capital return beat the bogey, with mgmt calling the stock undervalued and committing to quarterly buybacks. From here, estimates can keep rising if 1H27 contract prices keep climbing, LTA extensions push visibility out further and KV-cache demand turns NAND into a second AI leg. The multiple can expand as LTAs prove sticky through a pricing plateau: ‘27-‘28 FCF equals roughly half to two-thirds of today’s market cap, and at a >50% payout SKHY could retire ~20-30% of its share count by 2028 at ~2-3x earnings, while MU and SNDK trade at ~5-6x CY27.

Bears argue more broadly nothing about this cycle repeals memory’s basic economics: record margins pull in supply, and memory stocks peak when margins peak, not when earnings do. SKHY’s 85.7% GM is ~20pts above the prior-cycle record, DRAM pricing has roughly tripled this year, and the capacity response is now visible: SKHY’s 2027 capex goes up significantly, Yongin Fab 1 ramps in 2027 with M17 behind it, Samsung is executing better after its leadership reset, and CXMT keeps adding capacity while touting 4F². Some channel checks already put DRAM supply growth well above mgmt’s mid-20s demand framework. The “AI is different” argument also requires AI capex to keep compounding off an enormous base while memory becomes one of the largest line items in the AI bill of materials (industry memory revenue approaching ~US$1.7T in 2027), which invites pushback. The marginal buyer is the weakest one: neoclouds, roughly a fifth of AI data-center capex by some estimates, depend on increasingly expensive external funding, and the first place a capex pause shows up is non-LTA, spot-priced bits. Bears also think LTAs protect volume more than price: pricing floats with market conditions inside ranges whose floors aren’t disclosed, and customers will push to renegotiate if spot falls well below contract.

This quarter gave bears plenty to point to. Price growth is decelerating (DRAM from ~+30% to +high-20s%, NAND from +mid-50s% to +mid-20s%), and most expect DRAM pricing slowing to ~+7-15% in 4Q and to low-single digits by 2H27, so the earnings engine is shifting from price to bits just as SKHY adds +mid-20s% bits in one quarter. Customers are economizing: stack-count cuts, a relaxed Rubin spec and falling PC/smartphone units are all signs of price elasticity, whatever mgmt calls them. “Sold out through 2029” leans partly on volumes still under discussion, and there’s no return framework beyond FY26. Earnings quality is noisier than the OP line suggests: EPS missed by ~35% on KRW32.6T of FX and investment marks, the Solidigm IPO was pulled into a weak tape, 3Q diluted shares actually rose as July’s ADR raise outweighed buybacks, and another KRW10T of stock was bought for employee comp outside the KRW100T. HBM’s share of revenue has also retreated as commodity DRAM outpaced total top line, so more of SKHY’s earnings ride on commodity DRAM pricing than the “HBM moat” narrative implies.

STX - 2.5%: Beat Street but came up short of the buyside bar on every line, and the Dec guide ($4.8B / $9.20 EPS) is well above Street but below the ~$4.9–5.05B / $10–10.50 bogeys. Exabytes flat q/q. Tough for a still somewhat crowded long following WDC’s better print. Holding capex at 4-6%.

What a tale of two semis tapes: WDC crushes #s last week and is down same amount as STX this morning on a print that wasn’t as good, although STX held capex while one of the pushbacks against WDC was rising capex intensity for the next couple of years (STX mgmt must have been watching the px action AH on WDC’s print). For context, WDC guided Dec revenue ~12% and EPS ~21% above Street; STX is ~7% and ~8% above.

Read more

More broadly, as we discussed this weekend, the past few weeks have forced investors to better understand the reflexivity of the AI trade, with Lab ARR at the center of it all. Even after two strong semis sessions, the vibe feels different from past rebounds: less exuberance, more appreciation for dependencies across the AI ecosystem that aren’t going away. We think that’s healthy and mature.

Read more

TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.

Follow Anthropic, CVNA, CRWD or Doug O’Laughlin on Ethmos to get a brief whenever they or their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.

Today’s Essential Episodes:

1. Inside Anthropic’s S-1: SemiAnalysis on 90% Incremental Margins, 9x Customer Spend and Why the Compute Bill Isn’t That Big
   SemiAnalysis Weekly, with Jordan Nanos, Doug O’Laughlin, Jeremie Eliahou Ontiveros and Joey Brookhart of SemiAnalysis

2. CAS Investment Partners’ Cliff Sosin on Owning the Complements to Intelligence, Avoiding the AI Build-Out and His 80% Carvana Bet
   Invest Like the Best, with Patrick O’Shaughnessy and Cliff Sosin, founder of CAS Investment Partners

3. CrowdStrike CEO George Kurtz on Agentic Intrusions Tripling, Falcon Guardian and the Path to $20 Billion ARR
   The Compound and Friends, with Josh Brown and George Kurtz, co-founder and CEO of CrowdStrike

4. SemiAnalysis’ Doug O’Laughlin and Joey Brookhart on Anthropic’s S-1, Tokens as a Service and Lagging Fortune 500 Adoption
   TBPN, with John Coogan and Jordi Hays, and Doug O’Laughlin and Joey Brookhart of SemiAnalysis

^{Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.}

Listen · Watch · Follow Anthropic on Ethmos

THE SIGNAL

• S-1 net dollar retention: a customer spending $1 a year ago now spends about $9, up from roughly 5x in Q1, with 250 customers above $10M and 1,900 above $1M on a trailing-twelve-month basis.

• Ontiveros contrasts short SpaceX contracts (about $50M per MW-year, with 90-day terms) against Nscale’s ~$100B-per-GW commitment (likely heading to $120B+), arguing headline commitments understate capacity, at thinner margins.

• Revenue rose 1,368% YoY in the June quarter against 300% compute/infra cost growth (S-1); Brookhart ties a Q3 dip in revenue per MW to new SpaceX capacity, Meta moving usage in-house and extra safety compute.

• Brookhart says OpenAI expects a big Q4 enterprise ramp via Bedrock and Vertex and is reportedly discounting against Anthropic accounts, while only ~6% of its ~1B weekly users pay; a pricing risk to Anthropic’s premium (our read).

01 Margins and unit economics

The team walks through the S-1’s profitability disclosures as a central talking point. Brookhart cites 85% inference gross margins and 41% compute margins, alongside two quarters he calls profitable on a non-GAAP basis, framing the capital efficiency of customer acquisition — usage-based revenue requires little incremental sales spend — as unusually clean for a software business. O’Laughlin adds that on an incremental basis the company shows 90% gross margin and roughly 40% EBIT margin, arguing this directly rebuts “the biggest bears” who assumed heavy compute subsidization of tokens.

02 Compute commitments, reframed

Ontiveros spends significant time arguing the ~$500-600B compute commitment figure looks large only until compared with gigawatts actually contracted. Using the Anthropic-Nscale deal (about $100B per gigawatt) as a reference, he says the gigawatts Anthropic has contracted are materially higher than its commitments alone would imply, with compute set to rise materially by late next year, implying strong revenue growth is already assumed to justify the spend. He contrasts this favorably with the earlier OpenAI commitment controversy (~$20B ARR against ~$1.4T in commitments), suggesting Anthropic’s roughly $80B ARR against $500B in commitments, backed by $200B of post-IPO cash, is “a lot” but far more reasonable. He also notes the mix includes flexible deals (short-term, cancelable SpaceX-style contracts) alongside longer commitments like Nscale, improving balance-sheet efficiency even if per-unit pricing is higher.

03 Customer concentration and the CSP question

After some on-air confusion, the panel concludes that the S-1’s distribution figures (top five customers at 25% of revenue, CSP marketplaces like Bedrock/Vertex at roughly 48%) are separate categories, reading the top five as direct customers rather than marketplace buyers. They note that, by their reading of the S-1, the top 5% of Claude enterprise users spend about $18,000 per user per year versus far less for non-Claude Code users, and speculate — without confirming — about which companies sit in the top customer tier, floating names including Microsoft (via GitHub Copilot and Copilot 365), NVIDIA, and Jane Street, while acknowledging uncertainty. They also flag that 20% of Q3 revenue still came from models released before April, which they read as a signal of enterprise stickiness despite rapid model iteration.

04 Funding, risk, and the neocloud backdrop

Ontiveros details the newly disclosed capital — a $40B Google commitment (with $20B delivered and $20B tied to milestones like going public) and a $42B Broadcom deal enabling further TPU purchases — as evidence the 2027 compute roadmap is largely funded. O’Laughlin singles out the S-1’s language about never missing investor expectations, which he calls remarkable to put in a filing and consistent with Dario Amodei’s long-stated aversion to overextending: an under-promise, over-deliver signal for anyone worried about the fourth quarter. The group also discusses Elon Musk and Chamath’s recent comments seen as bearish on neoclouds, with O’Laughlin arguing that Musk, who runs the largest neocloud himself, is talking down rivals to lower SpaceX’s cost of capital as investors see it as the only game in town.

05 Competitive positioning: Anthropic’s moat and OpenAI’s squeeze

Brookhart and O’Laughlin argue Anthropic’s model-and-research flywheel — demonstrated by compute efficiency charts showing inference revenue per megawatt rising from roughly $16M to over $60M over the past year — gives it a durable lead, aided by a multi-chip strategy (NVIDIA, TPU, Trainium) that reduces dependence on any single vendor. O’Laughlin frames OpenAI’s position as structurally weaker: being “number two” forces competition on price rather than premium product margins, a dynamic he compares to semiconductor industry dynamics. Ontiveros offers a partial counterpoint, noting OpenAI may retain an edge in enterprise sales execution, which could offset share losses as enterprise AI spending — currently described as under $8 per employee per month even at large customers — begins to ramp in the current budgeting cycle.

06 Other disclosures: government revenue and seasonality

The filing shows government agency revenue at approximately 1% of total revenue across all reported periods, which the panel frames as a large untapped opportunity given hundreds of millions of existing Microsoft 365-style enterprise seats. They also note the S-1 discloses a roughly 20% decline in customer spend over the two-week year-end holiday period in the prior year, driven by consumption-based API usage dropping off, though they expect this seasonality to be less pronounced this year as Claude Code usage becomes embedded in daily workflows.

Listen · Watch · Follow CVNA on Ethmos

THE SIGNAL

• Sosin argues agents undermine aggregation theory: if an agent checks every search result, ranking first stops mattering and price, quality and service decide; that shift threatens search and marketplace aggregators’ pricing power (our read).

• On semis, he calls NVIDIA’s moat mostly intelligence (design files shipped to Taiwan), which AI could replicate, especially behind the frontier, while TSMC’s tacit, physical process know-how resists AI, though its scarcity has everyone trying to copy it.

• Firsthand lab commoditization: his agent spreads build jobs across all his model subscriptions on a server, and he no longer knows or cares which model runs, so routing layers turn frontier models into interchangeable supply (our read).

• Sosin says rents built on customer lethargy are exposed: agents will chase the highest FDIC-insured deposit rate, hurting banks like Bank of America, and vertical software sold to small operators, his bowling-alley example, gets crushed.

01 Intelligence’s hidden ceiling

Sosin’s core thesis, developed from an earlier phone conversation with host Patrick O’Shaughnessy, is that popular narratives treat intelligence as an unbounded scalar climbing toward godlike capability, which he views as a conceptual error. Instead he models intelligence as a tool for interpolating between known facts about the universe — excellent in closed, solvable domains like mathematics but fundamentally limited in complex dynamic systems, the technical category that includes weather, mixing cream into coffee, or human social interaction. He retells the classic chaos-theory origin story — a weather model producing wildly different forecasts after a technician rounded initial conditions by one decimal — to illustrate how small errors compound (the butterfly effect), meaning no amount of intelligence can substitute for actually gathering new information through real-world experiments. One corollary he highlights: intelligence can be turned on itself to reduce its own cost, meaning good-enough intelligence is rapidly becoming nearly free even as true breakthroughs still require touching reality.

02 Complements to intelligence

Applying this, Sosin says that once you assume everyone has access to excellent intelligence, the real investment question becomes what’s left — i.e., what scarce inputs AI cannot replicate. He lists these complements as data, attention, emotional attachment, physical proximity, scale, and, more cynically, regulatory capture. Meta is his flagship example: he reframes its core product as personalized TV, and argues that essentially all of its $250B in spending — data centers, inference, engineers’ code, video review — reduces, in a highly reductionist sense, to purchasing intelligence, a cost that should collapse as AI gets roughly 100x cheaper over five years. What doesn’t get destroyed by cheap AI is Meta’s data and attention, which let it monetize better-targeted ads and creator content regardless of falling compute costs. He floats Grindr (which he doesn’t own) as another company structurally capped by limited engineering resources that cheap intelligence could unlock.

03 Frontier labs’ commoditization risk

Read more

Tuesday: NXPI, STX

Wednesday (1/2): MSFT, GOOGL, ETSY, EBAY

Read more

In Tech. Semis +1.4% leading the way higher again as investors felt yesterday cleared the deck somewhat with a better received Anthropic S-1 and tea leaves pointing to a good Oct ARR #, along with positive financing news from META, AVGO, and SPCX all of which helped remove some uncertainty from the Lab ARR and funding debate. We had some thoughts on the cleaner set up for semis in our EOD Wrap yesterday. Memory names were up 4-6% in Asia. Neoclouds +3%+early. CLS +12% on a beat and raise ahead of their investor day today. AVGO +2.6% on a DA Davidson Upgrade and $63B in funding. Software -0.4% led lower by HUBS -2.5% after a UBS downgrade.

Stocks in Asia mixed despite Korea being up: TPX +0.61%, NKY +1.73%, Hang Seng -0.73%, HSCEI -0.45%, SHCOMP -1.33%, Shenzhen -2.55%, Taiwan TAIEX +0.12%, Korea KOSPI +3.09%

We’ll start with EPS recaps for CDNS, CLS, GLW, and PYPL then onto the usual (click to skip straight there)

Let’s get to it…

EARNINGS:

CDNS +2%: Looks better with Q4 Guide Well Above Street Drives a Third FY Raise; Maestro Is CDNS’s Answer to GPT-Synopsys

CDNS beat its revenue guide midpoint by $23M (vs. $14M in Q2) and raised FY26 revenue by $102M to +21% (~18% organic ex-Hexagon), with record backlog of $8.2B (+17% y/y) in a lighter renewal year. Maestro looks like a great answer to GPT Synopsys. Overall, good looking numbers with bulls focused on the better Q4 guide, but bears will push back and say AI debate still not solved, and the hoped-for 2H recurring acceleration did not come with just in line Q3 revs, weaker GM, the upside came from upfront revenue,& bookings fell y/y with cRPO flat

Upside was upfront-led (record hardware, IP +24%) while recurring grew 18%, and mgmt acknowledged customers aren’t yet paying meaningfully for agents; AI is monetizing indirectly through heavier base-tool usage. Investor focus on Maestro, a family of CDNS-post-trained open-weight models plus a model router, positioning CDNS to own the AI design interface rather than partner with a frontier lab the way SNPS did with OpenAI. Buybacks doubled to $400M with a new $2.5B authorization, though GAAP EPS missed the guide and Q3 OCF was light.

Details: Q3 revenue of $1.633B (+22% y/y) and EPS of $2.11 beat Street ($1.614B / $2.05); the Q4 guide of $1.71B / $2.22 sits ~4% / ~9% above Street, and FY26 raised a third time to $6.40B (+21%) / $8.40. Doubled buyback

Key Takeaways

• Guide. The Q4 guide of $1.695–1.725B (+18.7% y/y at the midpoint) and EPS of $2.19–2.25 sits ~4% / ~9% above Street ($1.649B / $2.03), with non-GAAP OM of 44–45% vs. Street ~43.2%. FY26 revenue rose $102M at the midpoint to $6.387–6.417B (+20.9% y/y vs. Street $6.32B, +19.4%), ~4x the $23M Q3 beat, and EPS rose $0.30 to $8.40 (Street $8.15). It is the third straight raise and came in ahead of investor expectations after the big Q2 raise. The FY revenue guide now sits ~$450M above the initial $5.95B, though ~80% of that increase traces to Hexagon (~$160M) and China upside (~$200M); stripping both out leaves ~14% growth. Mgmt still called the outlook prudent and noted Q4 is usually its strongest quarter.

• Recurring. Recurring revenue grew 18% y/y (79% of mix vs. 78% in Q2), down from ~24% reported in Q2, when Hexagon added ~4 pts and mgmt put normalized growth at “high teens to 20%.” The 18%-vs.-20% comparison raised on the call mixes Q3’s reported rate with Q2’s normalized one; like for like, and assuming a similar Hexagon lift, organic recurring slowed to roughly mid-teens. With 2H recurring acceleration the key pre-print focus, this is the quarter’s main blemish, and the beat came from upfront revenue (+42% y/y) instead. Mgmt tied the slowdown to China timing in the year-ago quarter (China was 18% of Q3’25 revenue) and pointed to FY25’s 11% as the better yardstick; averaging Q2 and Q3, recurring is ~21% reported and mid-to-high teens organic.

• Callback. Mgmt said the raise reflected a better Q3, improved visibility and strength across the business, led by record hardware demand and IP, and that the outlook remains prudent. Despite a lighter renewal year, bookings and backlog benefited from add-ons as customers expanded usage beyond initial plans, but customers are not yet paying significantly for agents; increased use of existing tools is what is driving more license purchases. The Q3 charitable contribution was in the “tens of millions,” which implies OM would have topped the 44.5% high end of the guide without it. On AI, CDNS wants to stay model-neutral and not be tied to one provider despite discussions with OpenAI this year, arguing Maestro’s edge comes from deep integration with its EDA tools rather than the underlying model.

• Backlog. Backlog hit a record $8.2B (+17% y/y) and cRPO $4.2B (+20% y/y). Both barely moved q/q (backlog +~$0.1B, cRPO flat), implying Q3 book-to-bill of only ~1.06x. YTD bookings growth slowed to ~26% from ~55% through 1H (both analyst figures mgmt didn’t dispute), implying Q3 bookings fell y/y against a strong year-ago quarter. Looking ahead, mgmt said 2027 carries a larger renewal base than 2026 and backlog should be higher y/y, but it stopped short of calling 2027 a stronger year.

• AI. Active Super Agent engagements topped 100 across ChipStack, ViraStack, InnoStack and AuraStack, vs. 20+ for ChipStack and 25+ for ViraStack cited in Q2, with ~10 public endorsements this quarter (mgmt says only ~1 in 10 customers ever goes public). Proof points: MediaTek put ChipStack into production verification (50–70% better coverage), Renesas cited 34x productivity from InnoStack, Nokia 5x from AuraStack, and Honda, using the new ChipStack RTL agent, saw 24% less area and 18% less power vs. pure foundation-model code. Mgmt credits agents for driving heavier base-tool usage behind the recurring step-up from FY25’s 11%.

• Maestro. CDNS launched Maestro, a portfolio of domain-specialized models (small decision models up to frontier-class open-weight models, including NVDA’s Nemotron) post-trained on CDNS tools and data, plus a router that escalates to frontier models only where they do better, expected in just ~5–10% of EDA use cases. Mgmt cites up to 30% better quality and up to 5x token efficiency vs. general-purpose models. Customers can post-train Maestro and run it on-prem or in the cloud, while CDNS keeps barring them from post-training generic models on CDNS tools. CDNS will charge for model usage on top of EDA and agent licenses, and the CFO stressed that customers can bear their own model compute and that CDNS won’t become a general-purpose AI infrastructure business. Incremental vs. Q2, when mgmt said customers were demanding LLM choice: Maestro productizes that with a CDNS-owned model layer, a clear divergence from SNPS’s OpenAI partnership (GPT-Synopsys).

• Hardware. Palladium/Protium posted another record quarter with new logos and growing hardware-cloud usage, and mgmt expects a record hardware year. The change vs. Q2: then hardware was still supply-constrained; now the CFO says there are “no capacity or supply chain constraints,” with inventory up another $58M q/q to ~$449M (+$72M in Q2) to ship against a significant backlog (one report still describes demand as exceeding available capacity). Demand is very strong across several large customers, including INTC; mgmt gave no 2027 hardware outlook beyond calling design-size growth a megatrend (quote #3).

• Margins. Non-GAAP OM of 44.1% was in line (Street 44.1–44.2%) but down ~350 bps y/y and 140 bps q/q, as non-GAAP G&A jumped to $81M from $54M in Q2 ($39M a year ago) on the timing of annual charitable contributions (Q3 this year vs. Q4 last year). The FY26 OM guide rose to 44.7% from 44.25%. FY26 incremental margin is only ~45% (+$1.1B revenue, +~$500M op income) because Hexagon adds little profit; ex-Hexagon, mgmt puts it in the usual 54–55% range, expects Hexagon headwinds to largely fade in 2027 (as with BETA) and targets ≥50% incremental margins. EDA and SD&A carry the highest margins, then IP, with hardware lowest. GAAP OM fell ~550 bps y/y to 26.2%, below the 27.5–28.5% guide, GAAP EPS of $1.05 missed the $1.11–1.17 guide, and roughly 60% of the non-GAAP EPS beat came from lower other expense and share count.

• IP. IP grew 24% y/y to an estimated ~$237–245M vs. Street ~$235–237M (segment dollars are derived from CDNS’s rounded mix), normalizing from >40% in Q2, which mgmt had cautioned not to annualize. Drivers: AI infrastructure, custom silicon and chiplets across interface, memory and foundation IP; Tensilica had a record quarter and a space company chose CDNS IP for a custom chip program. IP is near a ~$1B run-rate, with foundry diversification (the INTC 14A deal announced in Q2, Samsung) still ramping into 2027. On royalties, mgmt noted they already exist in Tensilica and design IP but called them “just one vector,” a contrast with SNPS’s ~$1B royalty ambition.

• EDA. Core EDA grew 20% y/y (18% in Q2), ahead of Street in dollars, with 16 new digital full-flow wins, including very large customers adopting the full digital flow (synthesis and signoff, not just Innovus). Displacements broadened to analog (Spectre FX at a marquee infrastructure silicon provider, multiple 2nm projects) and PCB (Allegro at a leading EV OEM), and verification was strong in big digital and memory designs at hyperscalers and memory makers.

• Customers. Systems companies are now ~45% of the business, and new chip designers include space and finance companies. Q3 brought a wide-ranging expansion with a marquee systems customer and a top telco that named CDNS its primary design partner. The CEO called design activity the strongest he has seen, with analog/mixed-signal and memory recovering alongside AI, helped by physical AI.

• China. China was 16% of revenue (15% in Q2, 18% a year ago; one report puts Q3 at 14%), is growing 30%+ in 2026 and is tracking ~15% of FY26 revenue vs. the 12–13% initially assumed (~$200M of upside). The guide again assumes export controls stay substantially similar.

Bull vs. Bear Debate

Bulls see CDNS as one of two toll roads on every advanced chip and system design, with ~80% recurring revenue and a product set that gets more valuable as AI raises design complexity. Design activity is the strongest mgmt has seen, and the customer base is broadening well beyond traditional semis: systems companies are now ~45% of the business, every hyperscaler is building custom silicon, and new designers range from AI startups to space and finance companies. The structural tailwinds are multi-year: chiplets and 3D-IC multiply verification and IP needs, a more diverse foundry landscape (INTC 14A, Samsung 2nm, Rapidus) requires new IP ports and DTCO work, and physical AI pulls SD&A into new markets. The moat is the middle layer of mgmt’s “three-layer cake”: signoff-accurate physics engines and silicon-correlated data that take decades to build and that foundries certify. Bulls argue agents raise the value of that layer rather than replace it, because every agent action still has to run CDNS engines to produce a design that works, and agents run far more design iterations than engineers do.

This quarter fed that thesis. FY26 growth is now 21% (~18% organic), the fastest in years, with a Q4 guide ~4% above Street and a third raise in three quarters. Recurring growth of 18% is still well above FY25’s 11%, and mgmt attributes the step-up to agents driving more base-tool usage, which is the bull mechanism at work. Super Agent engagements topped 100 with measured gains at Renesas, MediaTek, Nokia and Honda, and Maestro gives CDNS a credible answer to the frontier labs: own the model layer, route among third-party models and keep the customer interface. Hardware is no longer supply-constrained, so record demand can convert into shipments. Estimates can move higher on several levers: 2027 is a heavier renewal year in which agent and model value can be priced into larger multi-year deals; Hexagon integration should lift incremental margins back to the usual 54–55% or better; IP should pass a $1B run-rate on foundry diversification; and buybacks of more than 50% of FCF shrink the share count. Clear evidence of direct agent or Maestro monetization would let the multiple re-rate toward the high 40s P/E range CDNS held for most of the past five years.

Bears argue AI is the first real threat to EDA’s business model in decades, and that the best of this cycle is already in the numbers. The structural worry is that agents and frontier models become the designer’s front end, where design intent and workflow live, and push EDA engines toward commoditized back-end services priced on usage. AI could also raise engineer productivity enough that customers need fewer licenses per design, while frontier labs, hyperscalers and customers’ own agents capture the value of the new layer. CDNS still discloses no agent revenue, and mgmt concedes customers are not yet paying significantly for agents, so the AI upside remains a promise. Customers are already scrutinizing token budgets, and startups are going after the agent layer too (NVDA also works with ChipAgents). Meanwhile SNPS has laid out explicit long-term targets and a royalty model, and CDNS declined to offer a framework of its own. The cyclical worry is that 2026 benefits from things that may not repeat: ~80% of the ~$450M raise since the initial guide traces to Hexagon and China upside, two record hardware years leave a tough comp, China is growing 30%+ under an export-control regime that can change at any time, and IP is lumpy by nature.

This quarter gave bears several data points. Recurring revenue, the cleanest read on AI-driven software consumption, slowed to 18% from ~24% in Q2 (roughly mid-teens organic) just as bulls were looking for 2H acceleration; the beat came instead from upfront hardware and IP, the lumpier part of the model. Q3 bookings fell y/y (book-to-bill ~1.06x), cRPO was flat q/q, OCF missed Street by ~25% on receivables and inventory, GAAP EPS missed its own guide, non-GAAP OM fell 350 bps y/y, and roughly 60% of the non-GAAP EPS beat came from below the line. Maestro also brings new costs: post-training models, running a router and possibly giving away model value to win the interface. Estimates and the multiple could compress on a hardware digestion year in 2027, tighter China export rules (16% of revenue), weaker renewal pricing if customers refuse to pay for agents, a slower Hexagon payback, or evidence that agents reduce seat counts. If investors conclude EDA is a middle-layer utility rather than the owner of the AI design workflow, the multiple would drift toward mature-software levels.

CLS +16%: Clean beat and raise. $40B FY27 Guide (+90%) Clears street by ~11% and 2028 Higher at $21.50

Looks better against very low expectations as stock had been down 30% into the print and 5% yesterday. Bulls will point to the FY27/28 raise which helps answer TPU share loss debate while networking also re-accelerated. Bears will nitpick on GM slight miss and concentration moving wrong way. Investor day today.

Q3 beat on revenue, OM and EPS, Q4 guide in line, FY26 raised to $21.0B/$11.55, and an initial FY27 guide of $40B/9.0% OM/$21.50 EPS well above Street, but buyside still expecting someting closer to $25 for the full year

• Revenue: $5.83B, +83% y/y (last q +62% y/y) vs. Street ~$5.45B, +~71%

• EPS: $3.30 vs. Street ~$2.96

• Non-GAAP OM: 8.5% (record) vs. Street 8.4%

• FY27 Guide: revenue $40.0B, +90% vs. Street ~$35.9B, +~74%; EPS $21.50 vs. Street ~$19.80

Growth accelerated from +69% to +90% on a $21B base, OM expanded another 60bps to a record 9.0%, and mgmt says 2028 customer demand forecasts are “meaningfully higher” than the $40B outlook. The top customer (believed to be GOOGL) also grew ~35% q/q, which undercuts the TPU push-out and share-loss worries that, along with a weak AI tape, took the stock down ~33% in the month into the print. The blemishes are mix and cash: adj. GM slipped to 11.2%, below the mid-11s mgmt talked to last quarter, and FY26 FCF was cut to $460M from $600M (one quarter after being raised) to fund 2027 working capital, with $1.5–2.0B of FY27 capex on top. The stock is up ~16% AH to ~13x the FY27 EPS guide, and today’s Investor Day needs to size 2028 and the cash path.

GLW +2% pre: Beat above the top of the guide, carried by Optical ($2.38B, +44% y/y vs. ~$2.25B Street); 4Q guide in line on revenue, better on EPS. No change to the Springboard targets in the release.

Solid quarter overall. Stock is ~19% lower than a month ago and about half off its late-June high, so the setup was more about Optical upside than a target raise. Up ~2% pre, a bit behind optical peers (LITE/COHR +3%).

Sept-Q core sales of $5.08B (+19% y/y, the fastest since Springboard launched) cleared the top of the $4.9–5.0B guide vs. $5.03B Street and in line with bogeys. Core GM of 40.5% was 30bps above Street (+150bps y/y), and core EPS was $0.91 vs. $0.89 Street and the $0.85–0.89 guide.

Optical drove beat: +15% q/q, with Enterprise +84% and Gen AI sales more than doubling. That’s ~$130M of upside on a ~$50M total beat, so the rest of the portfolio was ~$80M light. Glass Innovations fell 3% y/y as memory pricing weighs on TV/PC/phone units. Solar was flat q/q and still losing money, with the ramp slower than planned (in July they said Solar profitability would improve in 3Q; it didn’t).

4Q guide is $5.2–5.3B vs. $5.20B Street, and EPS $0.96–1.00 vs. $0.93 Street. The midpoint is right at the ~19% pace on a slightly easier comp, and +3% q/q is normal seasonality, so the upside is mostly EPS. Some previews had revenue closer to $5.3B.

Q3 annualizes to ~$20.3B, so the end-2026 $20B Springboard run rate is already in hand a quarter early (4Q guide ~$21B). Four new multibillion-dollar LTAs incl. Verizon and AT&T add carriers to the AMZN/NVDA deals from 2Q.

Call ongoing…looking for any lift to the 2028/2030 Springboard targets (the debate coming in), Solar ramp timing, and how much Optical capacity comes on in 2027.

PYPL -1%: Big EPS beat and FY raise (~$5.60 from ~$5.38), but cost and credit drove beat; branded checkout +1%, the low end of the talked-down 1–2%, and revenue light. Nothing on strategic alternatives in the release.

Mixed bag. Better P&L, but branded weaker Stock is ~11% below where it was before Stripe/Advent walked away in late Aug, so most of the deal premium is already out; a little lower pre.

3Q non-GAAP EPS was $1.45 (+8% y/y) vs. ~$1.32 Street and a guide for a low-single-digit decline. TM$ of $3.95B grew +2% (+3% ex-interest) vs. a “slightly positive” guide. Revenue of $8.58B (+1% FXN) was light vs. $8.71B Street.

Pushback is quality: transaction + credit losses fell ~$97M y/y (incl. a reserve release tied to selling more receivables), more than the entire $79M of TM$ growth. Non-transaction opex grew just +1%, and FY opex growth is now guided to mid-single digits from 7–8%.

Branded checkout +1% FXN (2Q +2%, 3Q’25 +5%), with EU tariff pressure offsetting US improvement. PSP slowed to +8% from +13%; Venmo held +14%. Transaction take rate -6bps.

On the raise, ~$0.14 of the $0.22 is the 3Q beat vs. guide, the rest is 4Q. 4Q TM$ “slightly positive” is a decel from +2% even on an easier comp (4Q’25 +3% vs. 3Q’25 +6%), so the 2-yr stack drops ~5pts.

For the call: any update on M&A / strategic alternatives, branded checkout exit rate into holiday, and 2027 framing, since most of the turnaround levers don’t hit until then.

Details:

3Q Non-GAAP EPS: $1.45 (+8% y/y) vs. Street ~$1.32 and guide LSD decline (3Q’25 $1.34)
3Q Revenue: $8.58B (+2%, +1% FXN) vs. Street $8.71B
3Q TM$: $3.95B (+2%; +3% ex-interest) vs. guide “slightly positive”
3Q Branded Checkout TPV: +1% FXN vs. talked-down 1–2% (2Q +2%, 3Q’25 +5%)
3Q TPV: $487.1B (+7% FXN) vs. Street ~$491B; Venmo $96.8B (+14%) vs. Street ~$95.3B; PSP +8% (2Q +13%)
3Q Transaction + Credit Losses: $386M vs. $483M a year ago
3Q Non-GAAP OM: 18.8% (+16bps y/y)
4Q Guide: TM$ slightly positive (on a +3% comp); non-GAAP EPS ~$1.43 (4Q’25 $1.23)
FY26 Guide: non-GAAP EPS ~$5.60 vs. prior ~$5.38; TM$ ~$15.7B vs. ~$15.6B; non-transaction opex MSD growth vs. 7–8%; adj. FCF $6.5B+ vs. $6B+; buyback $6B
Capital Return: $1.5B buyback in 3Q (~28M shares); TTM $6.0B, weighted shares -11%

TECH RESEARCH/NEWS

Let’s start with AVGO…

AVGO / Anthropic: $63B chip financing package announced – Broadcom

Bank of America, Blackstone, Citi and Morgan Stanley announced a $63B financing structure for an SPV to buy Broadcom AI accelerators and networking equipment for lease to Anthropic, with deliveries scheduled for 2027. Banks are arranging first-lien financing; Blackstone is the sole committed second-lien provider. The announcement formalizes funding for another large block of AI compute.

AVGO: DA Davidson upgrades to Buy/$400 as Anthropic/OpenAI funding keeps AI ASIC cycle intact

Read more

TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.

Follow Dylan Patel, Citrini, MSFT, CRWV or OpenAI on Ethmos to get a brief whenever they or their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.

Today’s Essential Episodes:

1. SemiAnalysis’ Dylan Patel and Bernstein’s Stacy Rasgon on NVIDIA’s Neocloud Backstops, the Substrate Crunch and Custom Chips
   Chipstrat, from the CASPA CEO Summit panel moderated by Austin Lyons, with Stacy Rasgon of Bernstein and Dylan Patel of SemiAnalysis

2. Citrini on Why Monkeys Are the New Bottleneck, His Claude Antivenom Experiment and the Lab-Revenue Risk to the AI Buildout
   Monetary Matters, with Jack Farley and Citrini of Citrini Research

3. Microsoft CEO Satya Nadella on Zero-Trust Agents, Why Alignment Won’t Fall Out of Scaling and Every Company Owning Its Own Evals
   20VC, with Harry Stebbings and Satya Nadella, CEO of Microsoft

4. Ex-Fidelity PM George Noble on Shorting CoreWeave, Tesla and SpaceX, and Why He Calls AI the New Shale
   Monetary Matters, with Jack Farley and George Noble, former Fidelity portfolio manager

5. OpenAI’s Ari Weinstein and James Sun on Computer-Use Agents Faster Than Humans, Enterprise Guardrails and the Dots Cloud Computer
   The a16z Show, with Seema Amble, Fabrizio Serafini, and Ari Weinstein and James Sun of OpenAI

^{Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.}

Transcript · Follow Dylan Patel on Ethmos

THE SIGNAL

• Patel says Nvidia’s neocloud backstops guarantee rent at $3.50, the cloud’s breakeven, then take a 50% revenue share above that for Sharon AI and Fermi; Lambda’s Anthropic cluster carries 10%, so Nvidia absorbs tail risk (our read).

• Patel says Musk’s play is to sign no contracts until capacity is live, earning about $50B per gigawatt versus roughly $15B on contracted offtake; he puts Anthropic’s inference margins at 75-80% at $15B/GW, far lower at $50B.

• Rasgon: memory is 70-80% of AI rack silicon area, yet memory names trade at 2-4x forward earnings and SanDisk’s quarterly EPS guide tops its stock price of 24 months ago; he wonders what happens if 2028 pricing holds.

• Rasgon puts Nvidia revenue near 50/50 hyperscale versus non-hyperscale, AI labs about 20% directly and indirectly; Patel says Nvidia held margins near flat while absorbing memory, substrate and optics cost increases.

01 Financing gets harder, not impossible

Rasgon frames financing as one of several constraints — alongside power, supply, and semicap tools — now moderating AI buildout pace, noting the AI stocks are still up a lot but well off their peaks, a sign the market fears the buildout’s peak may be approaching. He observes hyperscalers, once the most cash-generative businesses in history, are now burning cash and probably headed for equity raises (Meta’s new share class structure would enable one, and Google has raised); CoreWeave’s unsecured bonds, which moderator Austin Lyons cited at 15% yields, signal where new issuance would price. Rasgon argues constraints aren’t purely bad — they prevent the kind of overbuild seen in the dot-com era’s unused dark fiber — and could produce a steadier, longer buildout cycle by better matching supply to demand.

Patel adds that financing has bifurcated: any cloud could get capital a year or two ago, but now neoclouds, which he says account for nearly half of what’s being built, can’t build without high-grade offtake, and even offtake from Anthropic or OpenAI gets priced at 13% to 15% debt. He cites Volta’s cluster build for Anthropic, where a 12% coupon bond sold at 95 cents effectively yields ~15%. This locks out neoclouds serving smaller enterprises and startups unless they strike structured deals — Nvidia’s revenue-share arrangements with Sharon AI, Fermi, and Lambda, and Broadcom’s rent-to-own chip leasing for Anthropic, are examples of chip vendors backstopping financing to keep capacity growing. Patel puts Meta’s compute and data-center commitments at $1.4 trillion and calls financing part of a “holy trinity” alongside land/power/shell and chips.

02 Ranking the real bottlenecks

Patel ranks substrates as the top constraint, ahead of memory, then optics/co-packaged optics (CPO) ramp — he says even one major company couldn’t satisfy 2028 CPO volumes given laser and fiber-attach limits. He argues power is a less severe bottleneck than popularly believed, since roughly 50 gigawatts of data centers are being built globally next year (7-8 in China, ~35 in the US and the rest worldwide, with Australia, India and Malaysia among the big places), and operators can work around power limits with on-site turbines, reciprocating engines, or fuel cells faster than they can fix semiconductor supply chains.

Rasgon largely agrees but stresses memory and wafer fab equipment (WFE) as constraints. He notes the trade ratio for high-bandwidth memory requires three to four times as many wafers per gigabyte versus standard DDR5, and that a company’s WFE guidance of above $200 billion for next year is about as high as he thinks possible given full clean rooms, though investors expected more — WFE estimates have already risen from roughly $120 billion early this year to a projected ~$155 billion. He expects tightness to persist through 2028, with 2029 a demand question.

03 Pricing power discovered during COVID

Rasgon recounts his own McKinsey research from years ago, which by his recollection showed the semiconductor industry destroying about $5 billion in aggregate economic profit from roughly 1996 to 2004 ($50 billion excluding Intel), meaning nearly all value flowed to customers. COVID changed that: input cost increases were fully passed through, and industry blended ASPs rose roughly 50% from 2019 to 2023/2024 even as unit shipments stayed flat — a dynamic he says is now playing out at an even greater intensity, with 70-80% of this year’s roughly 100% industry revenue growth (toward $1.5 trillion) coming from memory ASPs.

Patel notes some players flex pricing power opportunistically (memory firms swung from negative margins a few years ago to high margins, rewarding themselves on the upside after being squeezed in the downturn) while others like TSMC climbed steadily from 45% to 60% gross margin without giving much back in downturns. He estimates building a gigawatt of compute rose from roughly $40 billion in 2024 to over $60 billion now, though stripping out chip-vendor margin (Nvidia, Broadcom, Marvell) would put raw cost near $20 billion — meaning five gigawatts could theoretically be funded by one gigawatt of inference revenue. He reads Anthropic’s recent disclosures as roughly $70-80 billion per gigawatt of inference capacity.

04 Nvidia and Broadcom’s margin defense

Rasgon frames Nvidia’s margin durability as a function of performance-per-dollar gains outpacing ASP increases: Ampere-to-Hopper ASPs roughly doubled while performance rose ~10x; Blackwell ASPs rose ~50% while performance gained 3-30x; Rubin rack costs are probably doubling and cost per gigawatt is rising (about $25 billion for Blackwell and $40 billion for Rubin, as he recalls Nvidia’s figures) against another ~10x performance gain. He cites Nvidia chief scientist Bill Dally’s 2023 Hot Chips keynote target of a millionfold GPU performance improvement over ten years, following a thousandfold gain in the prior decade, as the benchmark that must hold for margins to stay defensible.

Patel says Nvidia and Broadcom trade at their lowest multiples basically ever, and Rasgon says Nvidia had traded cheaper relative to the SOX semiconductor index on only 13 days in the 10 years before a note he wrote nine or 10 months ago, and it’s cheaper today. Patel ties the discount to fears that big customers like Meta and Google will move away from Broadcom or stop buying Nvidia and insource, and says the shift has started: Nvidia’s slice of the industry’s CoWoS wafers and HBM fell in 2026 and falls again next year as Google, Amazon and AMD take share, though its earnings still double.

05 Custom silicon and enterprise adoption

Patel describes OpenAI’s first-generation custom chip beating Blackwell when SemiAnalysis ran open-source Chinese models on it with its InferenceX benchmark in OpenAI’s lab, a chip designed with heavy AI assistance; Anthropic aims to develop its own chip in half OpenAI’s development time, while Google’s TPU, Amazon’s Trainium (now fully usable in PyTorch), Meta, and Microsoft all pursue aggressive custom-silicon roadmaps. He notes the time to get new hardware running inference software on InferenceX has compressed dramatically — about six months for Nvidia and six to nine for AMD, a few months each for Google and Amazon, and a month and a half for OpenAI — suggesting chip design itself may become increasingly commoditized like software, which would pressure Nvidia and Broadcom margins longer-term even as both maintain strong positioning for the next two years.

On enterprise adoption, Rasgon identifies AI coding as the clear killer application with vertical revenue growth, and shares an anecdote of an analyst using Claude to one-shot a ten-year financial model in an hour and 20 minutes — work that would otherwise take a week. He frames the broader debate as whether AI spending generates measurable return, concluding anecdotal evidence leans positive though adoption remains early. Patel adds that enterprises will mostly buy tokens rather than deploy their own hardware or models, arguing scale economics push even pharma companies and hedge funds toward consuming AI as a service rather than running infrastructure themselves, continuing a decades-long industry shift toward cloud consumption.

Listen · Watch · Follow Citrini on Ethmos

THE SIGNAL

• Citrini says if combined lab ARR lands at $300B rather than the $500B being underwritten, the math that funnels lab revenue back into infrastructure breaks; meanwhile open models’ share of net-new ARR rose from ~1% to ~18% in a year.

• DeepSeek’s claimed inference stats (155T tokens/day on Huawei cards) imply ~$38 revenue per GPU hour versus $5–8 cost, ~80% margins and ~81-day payback, implying substantial further token-price deflation, especially below the frontier.

• Charles River’s earnings lever is operating leverage: spare post-downturn capacity makes new preclinical contracts higher-margin; monkeys take ~4 years to breed and China switched from halting exports to importing; he seeks monkey-pricing alt data.

• Their binder-design run used ~10,000 GPU hours costing ~$50k, which he says would now cost $2–4k, the kind of deflation behind his view that early-stage assets get commoditized (our read: bearish single-asset biotechs).

01 Why he cut AI hardware exposure

Citrini says his fund largely exited AI infrastructure and semiconductor positions starting around March, shifting toward what he calls “agentic utilities” — companies benefiting from actual AI usage rather than the capex buildout itself. He says he remains constructive on NVIDIA individually, but argues the infrastructure trade now requires underwriting aggressive assumptions, whereas usage-driven plays can be justified more simply. He notes that had his portfolio stayed as infrastructure-heavy as it was in April/May, it would likely be down ~50%; separately, Farley noted semiconductor stocks fell nine consecutive days in October, a streak he said was last seen in 2006.

02 The reflexivity thesis

Read more

Ok let’s touch on Anthropic first. We won’t dive into details here, plenty of that going around X, TMTB Slack, and in your inbox already. A few links we liked:

• The S-1 in charts: here

• FundaAI and SouthernValue X deep dive tweet threads here and here

A couple things that stood out to us:

1. 48% revenue is generated through CSP marketplaces like AWS Bedrock, GOOGL’s Vertex, and MSFT’s Foundry.

2. Customer concentration improving: The top five customers fell from ~36% of revenue in Q1'26 to ~21% in Q3'26. Top-five revenue still roughly doubled, from ~$1.7B in Q1 to ~$3.5B in Q3, and each of the five hit a record in Q3.. Revenue from everyone else grew ~4.4x over the same two quarters, from ~$3.0B to ~$13.3B. Customers spending $100K+ a year went from >950 to >11,900 between September 2025 and September 2026.

3. Cash levels/FCF positive after SBC, and operating leverage better than expected

4. Total customer spend fell about 20% in the two weeks ended January 2, 2026, compared with the two weeks ended December 19, 2025. This is important bc means Dec ARR likely will be understated ~10%...something to keep in mind

Overall, we think the release was largely positive as it reinforced Anthropic’s cash generation ability and assuaged some bear fears about Anthropic’s business model (despite a lot of it already being telegraphed in leaked reports).

Importantly, they left a spot open to fill in October ARR, which means we likely get it middle of next week when they filed updated S-1. We spent a good portion of our weekly yesterday detailing why Lab ARR matters so much for the AI trade and how it flows through to the financing and build out - it was one lesson we learned over the last month. So a lot of focus will be on that number.

Let’s dive in:

Read more

Anthropic S-1 set to drop at 9am ET so get your Claude’s ready…

In Tech, Memory +2-3% leading semis higher, TSLA +2.5% on a Barclays upgrade, MELI -15% as Bolsonaro beat Lula in a surprise results in Brazil, SPCX +1% after raising $40B in debt from Apollo & CRWV -2% on Elon’s tweet storm over the weekend, suggestion debt markets had all but closed for Neoclouds.

Already looking like it’s going to be a very fun week and we got a full earnings lineup in Tech as well with GOOGL/MSFT on Wed and AMZN on Thursday. We’ll have some bogeys out later today or tomorrow.

Asia saw mixed price action overnight: TPX +0.18%, NKY -0.8%, Hang Seng +0.51%, HSCEI +0.59%, SHCOMP +0.19%, Shenzhen -1.13%, Taiwan TAIEX -1.8%, Korea KOSPI -1.46%. Memory was flattish.

Lots to get to this morning, so let’s get to it. We’ll start with META and SPCX…

META: To offer $22.5B of Common stock, $60B of Class C stock, and $12.5B of Preferred - Bloomberg

NVDA AMD AVGO interested in buying $5B in aggregate

Timeline: (h/t Zackin Slack)

SPCX: Elon says they raised $40B in Debt and talks down neoclouds in weekend tweet storms

Debt deal led by Apollo with Apollo-backed Atlas SP slated to manage the entity, Under the contemplated master finance lease, the Equipment will be delivered in tranches, each of which will be subject to a separate lease schedule with a term of approximately five to five and one-quarter years from the applicable commencement date.

TSLA: Barclays upgrades to OW/$450 as SPCX takeout optionality joins improving core + Robotaxi inflection

Barclays upgrades TSLA to OW and raises PT to $450, arguing upside is no longer just about a potential SPCX combination: core auto trends are improving after two years of declines, Energy is nearly sold out for 2027 with >50% shipment growth potential on data-center demand, and Robotaxi miles have inflected sharply alongside FSD uptake >75% on newly purchased North American Teslas in 3Q. Barclays still sees a potential SPCX/TSLA merger and associated premium as an important catalyst, but now views improving fundamentals and Robotaxi momentum as standalone support for the stock.

TSLA: DB says SpaceX merger could become the main stock driver; Robotaxi progress tangible, Optimus still early

DB keeps Buy/$420 and says a potential SpaceX combination may increasingly anchor TSLA’s valuation, potentially creating a floor as investors handicap the premium needed to secure shareholder approval. Robotaxi is the clearest operating positive: unsupervised miles reached 3.1M, ~9x in three months, Cybercab is already in paid Austin service, and broader 24/7/geographic expansion is planned into 2027. Optimus remains much earlier, with production-quality units starting but management still describing a difficult S-curve; DB does raise energy-storage expectations as AI data-center demand improves pricing

SPCX: DB raises PT to $250 as compute pricing + capacity economics look better than expected

DB says near-term estimates remain too low after 3Q, with Customer B already generating ~$1.68B/month vs. its prior ~$1.25B assumption, suggesting an earlier renewal and/or pricing above the original ~$31/GW. DB also sees SpaceX’s vertically integrated build model sustaining ~$35–36B/GW capex, with a 440K Vera Rubin GPU training cluster (~1.3–1.4GW) due by 2Q27 and spot-market economics implying ~8–9 month paybacks. Funding looks manageable despite a ~$175B gap under a 10GW 2027 build scenario, with DB expecting investment-grade/GPU-backed debt; Buy, PT $250 from $235.

SPCX: Bernstein raises PT to $268 as terrestrial AI capacity/pricing materially expands the revenue path

Read more

Rough week for many L/S funds this week with GS’ HF VIP basket now down 10%+ YTD and had its worst move down since 2011 on Thursday. I’m hearing MS/GS primes has the average TMT fund down anywhere from 6-8% this month. Tech net selling over Oct 5-16 ran at more than 3 std deviations, the biggest reading on GS’s chart going back to 2017. Tech is still 23.5% of Goldman’s prime book, 90th percentile of five years. MS has said semi exposure has come down to 35% from 48%.

This dovetails with the conversations we’ve had over the past couple of weeks: plenty of exposure in AI semis despite de-grossing, but there’s a buyer’s strike. Faster-money semi guys have lowered exposure, but many of the longer-term guys I talk to still feel “stuck.” That vibes with our own trading: we’ve been very slow to move our feet in our long-term book and have taken a nasty drawdown from June peaks, but the charts got us out in time before much damage in our ST/MT books in this recent pullback. Meanwhile, a lot of the Tech generalists (both fast-money and longer-term) I speak to are spending more time hunting in software.

Mag7 Exposure near highs:

De-grossing has largely come at the expense of Nets, not Gross. According to GS, Global fundamental long/short net leverage is near a 3-year low (4th percentile). Gross still sits at 90%+ percentile

Some charts from GS via neilsethinew on X:

AI semis continue to get sold as investors grapple with the fallout of slowing ARR lab revenue and increasing financing risks (more on that later). Plenty of stocks/sub sectors now below the Leopold Low:

Price action isn’t great either: WDC down 3% on beat and raise despite the stock down 20% in a straight line since the beginning of Oct; SMCI has given back most of its positive pre-announcement pop; and semi caps took a beat on LRCX weaker guide.

LRCX now has roundtripped the whole move since the early April AI semis ramp began:

Pretty shocking chart. Not a bad r/r spot there to try long, but want to see that gap held if not going lower. Good tea leaf to watch for broader space.

Semis stuck in no man’s land approaching the 200d:

No other way to describe the vibes in semi land as something other than “sad” right now.

Meanwhile, in software land, things are feeling a lot better. Equal weighted SW is breaking out:

That looks like a chart that wants much higher. Cloud consumption, Cyber, and buckets of the higher quality app sw names (NOW, TEAM) leading the way higher here.

Price action here has been the opposite of semis. Just check out a name like TTAN which is up 60% in a straight line after being down 30% post-print in early Sept. Our sense in conversations is that tech investors are generally still underweight sw and companies have a relatively low bar heading into earnings. Outside of the names in the AI loser basket (HUBS, MNDY, INTU, etc.), we are interested in hunting names and moving quick on sw prints that show accelerating top line with potential narrative inflections - happy to buy up early on those in sw this earnings season.

The end of the Honeymoon period - what comes next for the AI trade?

The past few weeks have been painful for AI semi investors. One of our mistakes early in the month was not realizing how much of an impact Anthropic’s ARR print would have on the AI trade, and not “sussing” out the set up heading in: lower distillation revs, OpenAI taking share in September, & META/MSFT off-ramps. These were all things that were reported on; if this was an earnings print, we wouldn’t have wanted to be long in the weeks heading into it. In hindsight — knowing what we know now, it was a perfect set up to reduce exposure into it.

The upside of whiffing on that trade and the market’s renewed focus on Lab ARRs and financing? It has forced me (and other investors) to sharpen our pencils and better understand the different moving pieces of the trade, and how they play off each other.

Let’s rewind a bit before diving into where we are now:

For the first three years of the AI semi trade, and especially during the May/June go-go days, investors were in the throes of the honeymoon phase: maximum arousal, minimum scrutiny. Every piece of news was looked at with rose-colored-glasses. It’s just like the beginning of a new relationship when the infatuation is so strong that your partner can do no wrong. Even when she tells you she believes in an "open-source relationship," you somehow hand-wave it away and convince yourself it’s just going to soak up more of the frontier labs' excess demand. When your level-headed friends pull you aside to point out glaring red flags, like the fact that she’s running a shadowy circular-financing scheme with her roommates, you don’t listen; you just stop inviting them to parties. Before you know it, you’ve turned on Gavin Baker’s X notifications and forward SemiAnalysis to anyone who thinks you’re moving too fast, the ultimate wingmen who never stop hyping her up and telling you what you want to hear. “It’s way bigger than just coding, bro. She’s going to replace the entire global labor force.”

Before long, the endorphins and dopamine completely take the wheel. Even if you carry the scars from past relationships — whether the early 2000s, GFC, 2019 SaaS, or the 2020–2022 Covid binge — the blinders go right back on. Those exes weren’t real supercycles and everyone knows there’s no dark fiber in GPUs, you tell yourself as you start at GPU rental prices. You’ve even picked out baby names: Bel if it’s a girl; Claude if it’s a boy.

But eventually the dopamine wears off, and you have to get to know the person you’ve been telling everyone is perfect. You begin to see the flaws and start asking practical questions about spending habits, expectations, whose money is actually paying the bills. Can we afford this lifestyle on those lab ARRs, or are we just annualizing the honeymoon and assuming the capital stays cheap forever? And is there enough power in ’28 to keep the spark alive?

Things suddenly feel a lot more complicated, and we feel some investors longing for the simpler, more intoxicating days of May and June. But we see the end of the honeymoon as a chance for the relationship to mature. That means accepting where we are, adjusting expectations, and focusing our attention on where the strengths are rather than the flaws.

You are reading TMTB Pro. For more info or a trial, e-mail sales@tmtbreakout.com

Ok, let’s get serious. How does Lab ARR tie into financing and what comes next?

Read more

TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.

Follow AMZN, NET or OpenAI on Ethmos to get a brief whenever their executives go on a podcast. TMTB readers get 14 days free and 25% off their first month.

Today’s Essential Episodes:

1. Amazon CEO Andy Jassy on $220 Billion of Capex, the AI Demand Amazon Still Can’t Serve and Why Inference Must Get Far Cheaper
   The a16z Show, with Ben Horowitz, Erik Torenberg and Andy Jassy, president and CEO of Amazon

2. Ramp Economist Ara Kharazian on Who Actually Pays for AI and Why Spending Growth at the Top Is Slowing
   Plain English with Derek Thompson, with Ara Kharazian, lead economist at Ramp

3. Ben Thompson and Andrew Sharp on Netflix’s Growth Problem, the Internet’s Trust Crisis and Anthropic’s Alignment Philosophy
   Sharp Tech, with Andrew Sharp and Ben Thompson of Stratechery

4. Cloudflare CEO Matthew Prince on the PHP2Shell Flaw, Why AI Helps Defenders for Now and Agent Traffic at 1,000x Humans
   TBPN, with John Coogan, Jordi Hays and Matthew Prince, co-founder and CEO of Cloudflare

5. SemiAnalysis on ASML’s Hybrid-Bonding Tool, Why HBM Waits Until 2029 and China’s Packaging Push
   SemiAnalysis Weekly, with Jordan Nanos and SemiAnalysis analysts Sherry Zhang and Howie

6. OpenAI CFO Sarah Friar on 1.2 Billion Weekly Users, the Model OpenAI Is Holding Back and Going Live on Google Cloud
   Pioneers of AI, with Reid Hoffman and Sarah Friar, CFO of OpenAI, live at the Masters of Scale Summit

^{Produced by Ethmos: AI-assisted summaries of public podcast episodes. For information only. May miss context or contain errors. Verify before relying on them. Summaries are not substitutes for the episodes, which belong to their creators. Not investment advice, offer, or solicitation. Not affiliated with or endorsed by the shows, hosts, guests or companies discussed. Corrections and publisher opt-outs: support@ethmos.ai.}

Listen · Watch · Follow AMZN on Ethmos

THE SIGNAL

• Jassy says AWS capacity is sold out for 2026 and largely sold out for 2027, apart from on-demand reserves, with a good chunk of 2028 already booked, giving multi-year visibility behind the $220B capex.

• AWS’s AI business passed a $25B run rate within three years of the generative AI wave, versus about $60M for AWS after its first three years; AWS grew 36.7% last quarter on a $169B run rate.

• Core non-AI AWS is also growing fast, forcing a split of scarce memory between HBM for accelerators and DRAM for core servers.

• Jassy calls inference still an order of magnitude too expensive for pervasive enterprise use and expects enterprises’ existing production workloads to become the largest absolute segment of AI usage as costs fall, though he won’t predict how fast; cheaper inference favors Trainium (our read).

01 The capex calculus

Jassy explains Amazon’s investment logic splits into two cash cycles: data centers require commitment roughly two years ahead of monetization but carry ~30-year useful lives, while servers and networking need only months of lead time, break even in about three years, and run five-to-six-year useful lives on mostly five-year deals. This visibility into customer demand, he says, lets Amazon scale aggressively rather than guess — though the pattern creates near-term free-cash-flow headwinds while data centers ramp. He argues Amazon could justify investing considerably beyond $220B given unmet demand, though he expects supply constraints in power, chips, and memory to persist for several years, not just through 2027.

02 Pushing back on data-center backlash

Jassy disputes several claims behind political opposition to data centers, arguing Amazon’s facilities use less water annually than golf courses consume in two days, and that backup generators run only a tiny fraction of the year — around 0.1% — for maintenance. He compares the current buildout in scale to the U.S. buildout of highways and bridges in the 1950s, and attributes rising energy rates mainly to underinvestment in grid infrastructure rather than to data centers themselves. He notes Amazon has paid roughly $3 billion in taxes in St. Joseph County, Indiana, and has committed another $1 billion to local communities over five years through its Built Together program.

03 Chips, Anthropic and OpenAI

Jassy traces Amazon’s chip strategy to the 2015 Annapurna Labs acquisition, originally meant to offload virtualization and networking tasks from EC2 servers, which later produced Graviton CPUs — used by 99% of Amazon’s top 1,000 EC2 customers, with about 40% better price-performance than x86 — and Trainium AI accelerators, offering roughly 30-40% better price-performance than comparable processors. He frames this as a response to depending on a single dominant chip partner (following an earlier dynamic with Intel) whose incentives didn’t prioritize price-performance. Amazon’s investments in Anthropic and OpenAI (roughly $18B and $50B, a host’s estimate) are described as extensions of deep Bedrock partnerships: Anthropic has long built on Trainium, while the OpenAI deal let shared customers finally run OpenAI models on AWS via a new stateful runtime environment Jassy expects to underpin how agents are built going forward.

04 Reorganizing for agentic AI

Jassy points to internal examples — a host recalled his figure that six engineers rebuilt Bedrock’s inference engine in 76 days, work that would have taken 40 people a year, and he describes four or five people taking a new prototype to 80% shippable over a single weekend — as evidence organizations should shrink and de-specialize. He predicts companies will employ fewer people than they do today for current tasks, with smaller cross-functional teams, and near-total rewrites of existing codebases so agents can patch vulnerabilities within minutes rather than through the current multi-step human patch cycle. He singles out cybersecurity, not exotic AI-takeover scenarios, as the more immediate risk companies should focus on.

05 Reinventing products, advice to young people

On Alexa, Jassy says generative AI made it clear the underlying architecture needed a full rebuild, despite internal resistance from teams attached to prior design choices, and says Amazon is applying similar from-scratch reinvention thinking to retail, Prime Video, and Fire TV, even though the need there is less obvious. Asked for advice to an 18-year-old today, he says AI use in education is now unavoidable and universities shouldn’t fight it, but stresses that human judgment, critical thinking, and the work of synthesizing information into a strategy remain irreplaceable skills. He closes by linking Amazon’s sprawl across AI, robotics, autonomous ride-hailing (Zoox) and space (Amazon LEO, targeting 500 million underserved households) to a decentralized, single-threaded leadership structure that lets differentiated teams pursue multiple frontier bets simultaneously.

Listen

THE SIGNAL

• Ramp’s per-employee monthly AI spend tiers: median buyer about $12-15, top decile about $800, top 1% over $10,000; the top cohort spends roughly 800x the median per employee and is now actively cutting costs.

• Kharazian agrees with the Jevons argument: price wars and spreading spend across providers squeeze model-maker profits but still need more compute, so value goes to NVIDIA/infrastructure and end users rather than OpenAI/Anthropic.

• Thompson’s hurdle: model spend has to go from about $200B this year to about $400B, then about $800B; slowing top-1% growth, slower customer additions and falling prices all threaten that doubling pace.

• Fable lost to GPT Sol, which was slightly cheaper with little performance loss, because DoD-required prompt logging conflicted with enterprise data-retention duties; compliance friction plus near-substitutes cap frontier premium pricing (our read).

01 Framing the bubble as supply vs demand

Thompson opens by distinguishing the AI bubble’s supply side (chips, data centers, hyperscaler capex now partly debt-funded) from its demand side (actual revenue AI models generate). He argues the supply side looks less alarming than past bubbles because the buyers — Amazon, Microsoft, Google, Meta — are already enormously profitable (unlike thinly capitalized 19th-century railroads), and because nearly every AI chip today is in active, premium-priced use, unlike the vast majority of fiber left dark after the dot-com bust or the quarter of U.S. railroad mileage in bankruptcy by the 1890s. The demand side, by contrast, is largely unverifiable: OpenAI and Anthropic are private, so the public relies on leaks — one report pegged OpenAI’s revenue pace near $70B annually, another (FT) at closer to $50B, a large discrepancy that still moved markets. This opacity is why Thompson frames demand, not supply, as the central question, and why he brought on Kharazian, whose firm Ramp processes over $275 billion in annual business spend across 85,000+ companies, including over $10 billion in AI spend.

02 Revenue history and the Claude Code inflection

Kharazian traces AI enterprise revenue from ChatGPT’s consumer launch, through OpenAI’s early enterprise inroads, to Anthropic’s sharp share gains driven by Claude Code, a coding agent he says effectively took over or heavily augmented software engineers’ jobs and became a major revenue driver that spurred further infrastructure investment. More recently, he says competition has intensified — from SpaceX AI, NVIDIA’s own models, and overseas open-weight competitors — squeezing margins for the American labs.

03 Concentration and the slowdown among top spenders

Read more

In Tech, Semis flat while software -1%. WDC -2% despite handedly beating bogeys, BE -4% on a beat and raise, SPCX +2% on a sell-side upgrade and on reports they’re in talks to access 80k cell sites (Towers +7-8%; Telcos -4-5%). INFY -7% on lowered rev guide. RNG +15% on a set of good numbers.

Asia held in well despite the big declines yesterday in U.S.: TPX -0.79%, NKY -0.34%, Hang Seng +0.69%, HSCEI +0.87%, SHCOMP +0.89%, Shenzhen +2.44%, Korea KOSPI +0.3%. Hynix +4%; Samsung +0.5%

We’ll hit WDC BE VRSN RNG Earnings first, then onto the usual. Happy Friday! Let’s get to it…

WDC -2%: Beat and Raise above buyside bogeys, pricing Is Still Accelerating, but pushback is lower cost/TB decline and concerns around supply with higher capex

Recap: FQ1 revenue/EPS of $4.38B/$4.64 beat Street $4.14B/$4.10 on $/TB up 28% y/y. The FQ2 guide of $5.1B/$5.80 is ~12%/~20% above Street ($4.57B/$4.82), with GM guided to 60-61% vs. Street ~57.4%. FQ2 guide implies ~52.6% OM (Street ~49%) and ~75% q/q incremental GM. Opex steps up to $400-410M on product qualification costs.

$/TB accelerated to +28% y/y (from ~18% last q), mgmt expects Dec-Q growth to be higher still with pricing mostly locked, and it sees room to keep expanding y/y gains for another 3-4 quarters, which is driving ~80% incremental GMs. Exabytes (+22% y/y, below the >25% target for a second straight quarter) and cost/TB (only ~4-5% lower y/y on DRAM/NAND inflation) were the soft spots.

Overall solid set of numbers, especially given positioning as stock is down 50%+ since June highs and 20%+ since Oct 1st Toshiba news coming in (one of the worst performing stocks in AI semis). Stock was flat initially AH but fell another ~5-6% after hours after mgmt said capex “modestly above” the 4-6% range next to talk of accelerating exabyte supply growth. Neither message was new, but both touched on the post-Toshiba supply nerves.

Buyside numbers going up slightly morning and still sit around $60+ for CY‘28 vs street at $45 so trading at around 5x at sub $350. Bulls will like the accelerating rev % y/y and price increases to roll through the next 3 q’s with incrementals at 80%, which means still room for margins to keep rising. Bears will pushback on industry dynamics changing for the worse, mgmt’s comments on “pricing moderation” after 3-4 qs (although they walked that back on callbacks), and slower cost/TB declines.

Despite stock selling off on capex comments, message was the same as Citi conference:

On input costs, language was slightly more explicit than at Citi with only a 4-5% decline but kept 10% LT target alive

Key Takeaways

• Pricing. Blended $/TB rose 28% y/y and ~9% q/q (nearline +26% y/y, +8% q/q), up from ~18% in the June quarter and ~9% in March. Strength showed up in both nearline and client/consumer (client revenue more than doubled y/y). The drivers:

  • contractual escalators as LTAs hit their anniversaries

  • expiring LTAs renewed at higher prices

  • strong spot pricing on non-contracted exabytes

  • mix toward higher-capacity drives

  Dec-Q $/TB growth is guided above 28% (implied mid-to-high 30s%). HAMR will be priced at parity with ePMR per TB (“the price is the price”), so WDC won’t be buying HAMR share with introductory discounts. Incremental: pricing is now running well above the high-teens framing the Street had treated as sustainable.

• Margins. GM of 58.1% (+1,420bps y/y, +370bps q/q) beat Street ~55.6% and the 55-56% guide, with >80% incremental GM both y/y and q/q. OM was 49.1% vs. Street 46.4% as opex fell to 8.9% of revenue. Mgmt laid out three durable GM drivers:

  • pricing (escalators, renewals, spot)

  • up-mix to higher-capacity drives, which carry a higher ASP per drive

  • more exabytes through areal density without adding fixed unit cost

  In callbacks, mgmt said HAMR margins should be at least comparable to ePMR once at volume.

• Exabytes. Shipments were 248 EB (+22% y/y, +7% q/q), with nearline at 223 EB (+22%, ~90% of EB). Both came in a touch below Street (~249-250 total, ~227 nearline). Growth was stuck at 22% for a second quarter, below the >25% LT target and well behind STX’s mid-30s. Mgmt called Sept-Q “the floor.” It expects growth to accelerate through FY27-28 as 40TB ePMR reaches ~30% of nearline bits in Dec-Q and ~50% in Mar-Q, followed by HAMR. Mgmt called 40TB ePMR the “fastest transition of any ePMR platform.” The 50% target is unchanged from August; the 30% is new. Asked whether ~35% EB growth in six quarters is the message, the CFO said “you’re thinking about it in the right direction.” Bears will say if WDC grows supply ~35%, STX ~30% and Toshiba doubles capacity, industry supply could outrun demand and the 28%+ price increases could reverse. In callbacks, mgmt said 35% isn’t guidance. The long-term target is still >25%; above 30% is possible but not guided, and the growth is meant to fill orders customers have already signaled.

• Capex. Capex roughly doubled q/q to $227M (5.2% of revenue vs. ~3% in FY26).

  • New wording: the script now has capex “modestly above” the 4-6% range for the next two years, with the five-year average still 4-6%. In callbacks, mgmt said up to ~2pts above, i.e., ~7-8%, which is 3x+ FY26’s $418M in dollars.

  • Where it goes: head wafers (HAMR heads need more process steps and longer cycle times, which cuts throughput), media and glass substrates, and factory automation. It also covers more heads and media per drive as platter counts go from 11 to 12 and eventually 14. It is not for unit capacity: total platter counts are growing low-teens % y/y on minimal unit growth.

• Costs. Cost/TB fell only ~4.5% y/y and was roughly flat q/q, vs. ~8% in June and ~10% in the Dec and Mar quarters. That is well short of the ~10% LT target. Mgmt blamed DRAM/NAND inflation inside the drives in a tight supply chain, plus being between product transitions. Value-engineering fixes are gaining traction but take several quarters to implement and qualify. Near-term declines will stay under 10%, with no timeline back to ~10%. Incremental: input costs were first flagged at the September conference, but this is the first time the slippage is this visible. This is where WDC’s ~2-year HAMR lag vs. STX shows up in the P&L. Pricing is more than covering it for now.

• Competition. On Toshiba’s capacity plans and chatter about industry consolidation, mgmt said long-term industry fundamentals haven’t changed (25%+ demand CAGR, tight supply). It reiterated it will not add unit capacity. Mgmt argued the bottleneck is head and media capacity, not drive assembly. Players that rely on third-party heads and media may have to add units “just to maintain their share” and “will have a harder time” growing exabytes. WDC buys some external heads and media to benchmark its own, but only for lower-capacity drives; all leading-edge platforms use internal heads. That matters given reports of Toshiba’s interest in TDK’s HDD head business. If a competitor meaningfully adds units, WDC’s response “is going to be the same”: areal density, not units.

• HAMR. The roadmap:

  • qualifications with two hyperscalers finish in CQ1’27

  • volume shipments start in 1H CY27

  • two more customers begin qualification in 1H CY27

  • 2nd-gen HAMR (5TB/platter, ~55TB on 11 platters) begins qualification in 2H CY27

  Mgmt says customers like its failure rates vs. what they’ve seen from other players. The “two hyperscalers” wording read as a step back from September, when mgmt said it was “in qualification with four large hyperscale customers” (see callback below). High-bandwidth drives are sampling with multiple customers for KV cache in AI inference and for video streaming. Mgmt is now working out how to price them, with impact from late CY27 into CY28.

• Demand. Mgmt says revenue is limited by supply, not demand. All of its top seven customers provide visibility through 2030-31, the demand signal is “increasing” quarter after quarter, and the supply-demand gap isn’t going away near term. Mgmt framed four demand legs:

  • cloud, accelerating on video and AI-enabled services

  • training and RL

  • inference and agentic AI, with “tremendous” growth in logs, checkpoints and outputs

  • physical AI, led by AV customers in Asia and robotics

  It argued softness in any one leg, including neoclouds, would be absorbed by the others. Incremental: in September mgmt said customers were asking to sign LTAs through 2030-31. Now it cites visibility through 2030-31 across all of its top seven customers.

• Customers. Cloud was $3.9B (88% of revenue, +54% y/y, +16% q/q). Demand is broadening to enterprise and hybrid-cloud OEMs, neoclouds and frontier AI labs, which mgmt is now building into its forecasts and investment plans. Mgmt is selective on neoclouds, favoring strong balance sheets and quality end customers. Per callbacks, it also structures agreements to limit credit risk.

• Cash. FCF of $1.38B (31% margin) came in ~8% below Street (~$1.50B). Capex doubled and inventory rose ~18% q/q (+$270M), built in components rather than finished goods ahead of EB acceleration; mgmt says any extra exabytes would ship immediately. WDC returned $1.06B ($1.0B of buybacks, $56M of dividends) and raised the dividend 17% to $0.175.

• Callback. The most important takeaway: Dec-Q is largely de-risked. The vast majority of the quarter’s revenue is already under contract with pricing locked, and $/TB growth will exceed 28% y/y. Mgmt also walked back the scarier readings of the call:

  • ~35% EB growth is not guidance; >25% remains the LT target, and >30% is possible but not guided. Supply growth is aimed at existing customer demand, not excess capacity.

  • Four hyperscalers remain in HAMR qualification (two finishing first, with two new customers to follow), so there is no pushout.

  • The 3-4 quarter pricing comment refers to y/y growth continuing to accelerate, with no set timeline for moderation.

  The less comforting detail: LTAs mix fixed pricing, premium pricing on incremental volume and annual escalators. But commercial terms for the 2030-31 visibility years are not finalized, and purchase orders run only ~4-5 quarters out.

Bull vs. Bear Debate

Bulls this is an AI supercycle and Supply is limited by heads and media, not assembly lines. The two vertically integrated leaders (WDC and STX, ~90% share) are growing exabytes through areal density rather than units. The new pricing is being written into LTAs with escalators and renewals, and customers now plan storage five-plus years out alongside land, power and compute. Bulls argue this looks like a planned, multi-year infrastructure buildout, not the PC-driven cycles that earned HDDs single-digit multiples. They see Toshiba’s capacity ambitions as a prove-it story, given its reliance on third-party heads and media and its history of missed roadmap milestones.

This quarter fed every leg of that argument. Pricing accelerated to +28% with Dec-Q higher and largely locked, and mgmt sees 3-4 more quarters of expanding y/y gains. Pre-print Street CY27 models embedded only ~13-16% average $/TB growth. Every 5pts of CY27 pricing is worth roughly $1B of revenue that falls almost entirely to profit, or ~$2 of EPS.

GM is guided to 60.5%, and bulls see the mid-to-high 60s by CY28. 40TB ePMR and HAMR (priced at parity, with margins at least equal to ePMR) should lift EB growth from the 22% “floor” toward 30% while the industry adds no units. Opex is essentially fixed at ~$400M a quarter, so OM can head toward 60%. With 35-45% FCF margins and 85%+ returned, WDC would retire ~4-6% of its share count a year at today’s price. Bulls read the capex step-up as old news aimed at the real bottleneck (heads and media). Catalysts include STX’s upcoming print, HAMR qualification completions in CQ1’27, the Dec-Q pricing print and LTA renewals.

Bears see a classic commodity up-cycle dressed up as secular growth. $/TB is up 28% and heading toward the mid-30s in an industry where it historically fell 15-20% a year, and pricing like that invites a supply response:

• Toshiba reportedly plans to double capacity, and has reportedly shown interest in TDK’s HDD head business, which would ease its component bottleneck.

• STX is already growing EBs in the mid-30s on HAMR.

• WDC itself is more than tripling capex and talking up accelerating EB supply growth toward 30%.

An exabyte from areal density is still an exabyte. If industry supply grows in the high 30s while demand compounds at the ~25%+ mgmt keeps citing, the shortage closes sometime in 2028. The LTAs bulls lean on won’t stop it. Mgmt conceded that the further out you go, the less volume and pricing is covered; out-year terms aren’t set, and purchase orders run only 4-5 quarters. The incremental demand supporting spot pricing comes from neoclouds and frontier labs, the part of the AI stack most exposed to the financing stress now hitting the complex. Hyperscale customers (cloud is 88% of revenue) have every incentive to second-source and push back at renewals once supply loosens.

The bull case assumes pricing stays strong after supply responds and that WDC closes its technology gap. This quarter cut against both:

• EB growth was stuck at 22% for a second quarter.

• Cost/TB declines slowed to ~4-5% as the ~2-year HAMR lag vs. STX shows up in the P&L.

• HAMR volume still depends on qualifications that don’t finish until CQ1’27.

• The promised acceleration means running two transitions at once while absorbing HAMR head throughput losses.

Mgmt’s own comment that pricing growth will “eventually slow down” after 3-4 quarters puts the second-derivative peak on the calendar for 2H CY27 although mgmt seemed to walk back the moderation comment in the call back.

BE -4%: Third Beat-and-Raise Clears the Buyside Bar withs some pushback around the unchanged 34% GM guide

Read more

What caused the sell off today?

Read more

Pretty solid numbers overall. Don’t hear many investors asking why stock isn’t up more given semis investors have become immune to the anemic price action (see LRCX & NOK today) despite WDC being down 25% MTD. Will be interesting to see how this acts tomorrow.

Sept-Q revenue of $4.38B (+55% y/y, +17% q/q) came in above the top of the guide and the ~$4.25–4.3B bogey. GM of 58.1% was at the high end of what buyside had. EPS of $4.64 was ~13% above Street and ~5–8% above the bogey. Dec guide looks good: Revenue of $5.1B is ~12% above Street and well past the ~$4.7–4.8B bogey. GM is guided to 60.5% at the midpoint, ~330bps above Street and ~1.5–2pts above the bogey, and EPS to $5.80 vs. $4.78 Street and ~$5.00–5.20 bogeys. Implied op margin steps up to ~52–53% from 49%.

Read more