Over the last 24 hours we have gotten 3 pieces of really good news: 2 of the arguably most important KPIs to the AI trade — Hyperscaler ROI and Labs ARR — have come in better than expected (OpenAI, MSFT, AMZN), and the Leopold news. That all comes on the back of extreme price/sentiment. Recipe for at least a decent magnitude ST rally.

Korea was +18% overnight with Samsung +20%; Hynix +25%; TSMC/Softbank +10%+.

We’ll cover AMZN (+11%) first, then AAPL’s miss (-8%), RDDT’s DAU issues (-12%) Kioxia, and then onto the usual. We’ll post the RBLX (-22%) takes in TMTB Slack #earnings channel although don’t hear much interest on that stock these days either side.

AMZN +11%: ROI ROI ROI - that’s what Jassy hammered home on the call as AMZN delivered the AWS and ROIC print nervous investors needed, with 37% AWS growth and a 39.4% margin clearing bogeys

Read more

Why? Knowledge is volatility reducing. There’s been a lot of head-scratching on the magnitude of the moves down since Friday, which sapped investor confidence and forced further de-grossing as the drawdowns got intense. That’s not unique to the AI trade, but the AI trade was crowded, and it sees to carry more “behind-closed-doors information risk” than many other trades (eg, “what do the labs and others know that we don’t?”) When prices are falling and nobody knows why, every investor has to price in the possibility that the seller knows something they don’t. Everyone on X starts freaking out with weird theories. Moves feed on themselves, etc.

Now we have a reason why. It was SA’s leverage that got them into trouble and now there is an answer to what was previously unexplained, and arguably the cleanest resolution you could ask for: Citadel swooped in for entire public book in one print. Seller done. Near-term supply out of the way & the uncertainty discount comes out of the tape at the same time.

That’s all to say: we think the moves today make sense. It’s not surprising some of Leopold’s largest holdings were up the most: NBIS +27%; BE +27%; SNDK +26%; CRWV +22%. Many names up mid teens. TMT Mo +17%

Where do we go from here?

Read more

Some bogeys to get us through the last day of the busiest week of earnings season (AMZN, RDDT, AAPL):

Read more

Asia mainly red overnight: TPX -0.54%, NKY +0.71%, Hang Seng +0.2%, HSCEI +0.25%, SHCOMP -0.62%, Shenzhen -2.46%, Taiwan TAIEX -0.26%, Korea KOSPI -1.23%. Samsung flat after better #s overnight. SK Hynix -3%

We sent out our recap of MSFT earnings earlier this morning here in case you missed it. We’ll cover META, ARM, LRCX, and QCOM and then onto the usual. Still working our way to CVNA, EQIX, Sammy, and FTNT and we’ll post those Recaps/Bull vs. Bear Debates in the #earnings channel on Slack in a bit.

Let’s get to it…

META -9%: Disappointing quarter with top line shy of what buyside was hoping for and modest opex increase another negative (vs expects for reit or slight downtick). Zuck with a mediocre call performance and bulls didn’t get a definitive compute-rental launch or other scaled non-ad AI monetization proof point.

Sentiment/positioning had shifted positive following the AI compute and new model announcements early in the month so the mediocre top line print & opex raise along with a so-so call performance not going to do it here. Capex raise was limited to ~2% at the midpoint with the high end unchanged (AI semis bulls wanted a bigger raise of $145B at the midpoint). Mgmt refused to give a 2027 framework while saying plans are designed to maximize 2026-27 capacity and preserve the ability to continue growing compute in 2028 and beyond. Op inc missed but backing out one time fine and severance costs ($3.6B total), actually in line-ish and they reit FY26 op inc.

Overall, we think r/r still skews up (as it has for a while), but we thought the print was going to be better and was hoping Zuck and co would sound more enthused and give more detail about some of the key items bulls wanted to hear about (capacity leasing, model improvements, etc). Our sense is buyside numbers going down slightly but still in the $35-$36 range for CY27 (not including any compute rental related revs). At $530, stock is trading at 15x so downside here appears limited (we’ll stick with the stock in our LT book) and bulls will say it can’t get worse than this, but very near-term stock like stays in chop until we hear more detail around compute leasing or better model release.

The #s:

Read more

MSFT +10%: Goldilocks print as Azure materially beat and raise; Co-pilot adoption inflected to 10M q/q adds vs buyside at 5M+, margins beat, and CY26 capex not increased as Satya and Amy did a good job talking up ROI. What’s not to like?

Key number here was Azure coming in at 43% for the q vs bogeys of 41-42% and guide at 45% vs. bogeys of 42% with demand continuing to exceed supply. Copilot added 10M seats and is now at 30M. To cut to the chase: we think this is an N&N inflection quarter (numbers & narraitve) and we think MSFT moves quickly to the top of the Megacap pecking order after being at the bottom and underperforming for a long time. We think Azure continues to accel in 2H of CY and could get close to an exit growth rate of 50% as Fairwater continue to perform well. We like the stock here and added last night.

We don’t mind buying a stock up on these inflection quarters - in fact, we prefer it. It sets an easy downside case of where the stock was before the print (-10% to the down). Buyside EPS likely moving to low 20s in FY27 and mid 20s in FY28. At 25x+ FY28 $25, that’s $625+ in our upside case with our base centered on 25x FY27 $22 = $550. So 30% up, 10% down in our base case. Yes we’re in a very skittish market environment for anything AI related, but we stick to our process: this sounds pretty good.

The other key number investors & AI semi investors are focused on is whether or not Capex was cut - it stayed flat. Details:

Read more

A little late getting back to my screens this afternoon and hopping on earnings calls now so we’ll just do some earnings recaps for now.

META -9.5%: Disappointing quarter with top line shy of what buyside was hoping for and modest opex increase another negative (vs expects for reit or slight downtick). Capex raise was limited to ~2% at the midpoint with the high end unchanged (buyside & AI semis bulls wanted a bigger raise of $145B at the midpoint). Op inc looks like s slight miss but backing out one time fine and severance costs ($3.6B total), actually a small beat and they reit FY26 op inc. Ad impressions down vs. Q1…not sure much zuck can say on the call to get this ship turned around tomorrow, especially in this tape

Q2 Revenue: $60.8B vs. Street $60.2B
Q2 Ad Revenue Growth: +27% FX-neutral
Q3 Revenue Guide: $61–64B vs. Street ~$63B
Q3 Implied Ad Growth: ~+24% FX-neutral
FY26 Expenses: $165–169B vs. prior $162–169B
FY26 Capex: $130–145B vs. prior $125–145B
FY26 Operating Income: Expected above 2025

MSFT +3%: Very nice Azure number (43% vs 39-40% guide and 41% buyside bogey) and co-pilot adds better to 30M seats (vs 25M expectations). All eyes on the guide on the call with 43% likely being the bogey now. Revenue and EPS also beat comfortably. Investors will now focus on whether the Azure guide sustains the 43% growth rate and what management says about capex. Future leases at $329B up from $197B.

Revenue: $90.0B vs. Street $87.62B
EPS: $4.74 vs. Street $4.24
Azure Growth: +43% constant currency vs. 39–40% guide and ~41% buyside bogey

FTNT +11%: Material beat-and-raise cleared a very high bar. Billings growth accelerated to 33%, product revenue grew 52%, and both the September-quarter guide and full-year billings raise were well above expectations. The pull-forward debate will persist, but the results strong enough here for move up

Revenue: $2.05B vs. Street $1.89B
Billings: $2.37B, +33% y/y vs. Street $2.13B, +20%
Product Revenue Growth: +52% vs. low-40% growth last quarter
Q3 Revenue Guide: $2.055B vs. Street $1.95B
Q3 Billings Guide: $2.30B vs. Street $2.11B
FY Billings Guide: $9.45B midpoint vs. prior $8.95B midpoint

LRCX +6%: Very strong guide meaningfully exceeded elevated expectations. The September-quarter revenue guide was ~14% above Street, with substantial upside across gross margin and EPS as well. The print should help ease concerns that large-cap semicap estimates lacked further upside, especially after KLAC

Read more

MSFT META LRCX ARM QCOM FTNT Bogeys below. Feedback welcome as always

Read more

Lots of earnings to go through. We’ll cover STX, Hynix, KLAC, NXPI, VRT, and CTSH then the usual below that. We’ll put our recap of BE in the slack since we’re running out of time here…

META and MSFT tonight among others. We’ll have some bogeys out shortly.

Let’s get straight to it…

EARNINGS

STX +4%: Big beat-and-raise, with pricing, mix and HAMR cost-downs driving a much larger-than-expected margin step-up, while demand visibility extended into CY28 and customer planning moved into CY29+

Overall very solid print with the highlight the implied mid 57s September GM guide. We know what kind of tape we’re in but numbers objectively good and mgmt did good job talking up visibility/pricing/incremental margins. Will be interesting to see how it goes today.

The #s:

• FQ4’26 revenue was $3.629B, +48.5% y/y (last q +44.1% y/y) vs Street $3.49B, +~42.8%; non-GAAP EPS was $5.71 vs Street $5.08-$5.10; GM was 52.7% vs ~50.3%; and OM was 44.6% vs ~42.0%.

• FQ1’27 guidance was $4.10B revenue, +56.0% y/y vs Street $3.78-$3.79B, and EPS of $7.30 vs Street $5.80-$5.85 and a roughly $7.00 buyside bogey. Implied September GM of ~57.3%-57.5%…The magnitude of the EPS upside came from the combination of another pricing increase, a higher-capacity product mix, HAMR cost-downs and opex remaining near $300M despite 13% sequential revenue growth

Key Takeaways:

• Pricing: Nearline pricing increased roughly 10% y/y, accelerating from approximately 6% last quarter, while Edge pricing increased roughly 35% y/y. Mgmt said the supply-demand gap has widened, and any incremental exabytes produced above contracted volumes are being sold at prices above the original contract price. The favorable pricing expiration for one early HAMR customer was only a small contributor to the September improvement, with broader demand and customers chasing incremental volume the more important drivers.

• GMs: GM reached 52.7%, +570bps q/q and nearly 15 points y/y, while the September guide implies ~57.3%-57.5%. Estimated incremental GM was roughly 87% in June and 93% in September, far above the 50% framework previously discussed. Pricing, cost per TB declining by roughly 16% y/y, higher-capacity drives and limited opex growth are translating revenue upside almost directly into earnings. Mgmt said GM should improve sequentially through FY27 and that incremental GM is already well above 60%, but explicitly declined to identify a ceiling.

• Visibility: Approximately 90% of exabytes now ship into data centers, the vast majority of nearline supply is allocated into CY28, and customers are discussing CY29 and beyond. Product configurations and pricing are already contracted through the entirety of CY27. Mgmt said cloud exabyte demand has increased sequentially for three years with “no evidence of a slowdown,” while enterprise nearline revenue increased for a fifth consecutive quarter.

• Demand: Mgmt argued that inference and agentic workloads create and retain more context, with tiered KV-cache architectures allowing HDDs to hold persistent data and reduce repeated GPU computation. Neoclouds, leading model developers and enterprises are also starting to adopt hyperscaler-style tiered architectures. However, mgmt repeatedly characterized agentic KV-cache and physical-AI demand as “very early days,” and said the ultimate HDD attach rate will vary significantly by application.

• HAMR: HAMR represented approximately 40% of nearline exabytes exiting FY26. Mozaic 3 is qualified across all major cloud customers, Mozaic 4 is ramping at the two largest global CSPs and should represent 50% of HAMR exabytes exiting CY26, while Mozaic 5 qualification shipments remain targeted for late CY27. Mozaic 4 was already a meaningful contributor in June and is expected to contribute more in September.

• Capital Return: FCF was $1.118B, a 31% margin and the best quarterly result in more than a decade. Gross debt declined to $3.6B, STX plans to retire another $1.2B during the September quarter, and buybacks are already running above the June-quarter level. Once the remaining targeted debt is addressed, mgmt expects most excess FCF to be returned through repurchases and dividends.

Bull vs. Bear Debate

The structural bull case is that STX is no longer simply a cyclical PC-HDD vendor. The business is becoming a concentrated provider of mass-capacity storage into a rational cloud-HDD market, where HDD remains dramatically less expensive than NAND for bulk storage. Approximately 90% of exabytes already ship into data centers, HDD represents a relatively small portion of total hyperscaler capex, and AI is increasing the amount of data that is created, retained, duplicated and reused. Bulls argue that this makes storage demand more durable than compute capex alone, particularly as inference, agentic AI, video and physical AI increase persistent context and unstructured-data requirements. Bulls like bisibility and supply discipline: the vast majority of nearline exabytes are allocated into CY28, pricing and product configurations are contracted through CY27, and customers are already discussing CY29+ with most checks showing $25-$30/TB already in CY27 which implies $70+ in EPS power. STX is keeping drive-unit output broadly stable and using areal-density gains to increase exabytes, which limits industry oversupply while allowing incremental production to be sold above contracted prices. Bulls therefore believe that the traditional HDD cycle has structurally changed: rather than price per TB falling every year, the industry can capture part of the value created by higher-capacity products while HAMR lowers cost per TB.

This quarter the September guide cleared the buyside bogey, Edge pricing showed the same positive inflection as nearline, incremental GM approached 90%, and Mozaic 4 is now contributing meaningfully. Opex remains near $300M even as revenue approaches $4.1B, creating unusually high earnings leverage. With debt rapidly declining, the next leg of FCF can increasingly fund buybacks, allowing EPS to grow faster than operating profit. Bulls will say this should trade at least mid teens multiple and with pricing reaching $30/TB by 2028, you can get $100+ in EPS power = a $1,500 stock.

The bear case is that the market may be capitalizing peak pricing and peak incremental margins as though they are permanent. Customer commitments reduce near-term miss risk, but they do not eliminate a CY28 or CY29 cloud-capex slowdown. If hyperscaler infrastructure spending moderates, open-source models reduce the capital intensity of AI, DC construction encounters power or financing constraints, or customers eventually discover they have accumulated too much storage capacity, pricing can flatten before unit demand visibly falls. The same customer concentration that provides visibility can create a sharp correction when LTAs eventually roll off.

There are also technology and competitive risks. STX must pause or slow parts of its factories during transitions from 3TB to 4TB and eventually 5TB per disk, while yields, qualification timing and more complex head-and-media processing can temporarily constrain output or raise costs. Mgmt also acknowledged that areal-density transitions alone may not be sufficient to meet all longer-term demand. A slower Mozaic ramp would remove both the expected capacity benefit and part of the cost-down benefit. Beyond HAMR execution, a return to NAND deflation could improve SSD economics, WDC or Toshiba could become more aggressive, and weaker Edge or consumer markets could expose fixed-cost leverage. The margin debate is particularly important because the September guide has already moved GM close to 60%. If pricing normalizes while costs continue declining at only the historical low-to-mid-teens rate, incremental GM could fall sharply from the current 80%-90% level. Opex leverage would still be favorable, but the market would begin treating STX as a late-cycle hardware company again.

SKHY -2%: Hynix results ok as OP expectations had already come down in Asia give mgmt setting expectations lower, but the quarter was also modest quality miss as DRAM ASP and product mix lagged buyside expects; HBM4, LTAs and continued AI-memory tightness support a stronger 2H setup.

Revenue was ₩79.319tn, +257% y/y (last q +198% y/y) vs Street at₩79.964tn+~279% y/y, while OP was ₩60.543tn, +557% y/y vs Street at ₩64.986tn, +~605% y/y; OPM was 76.3% vs 77.1%. A lot of discussion where real bogey was - I couldn’t get a sense before the print, but remember that mgmt was out trying to massage expectations down a few weeks ago so many already expected the OP miss:

DRAM ASP of only +~30% q/q vs a higher-end +40%+ expectation. NAND was stronger, with bits +mid-teens and ASP +mid-50s; eSSD revenue doubled q/q and 30TB+ eSSD revenue more than tripled. For 3Q, mgmt guided DRAM bits +~10% q/q and NAND bits up low-single digits; no formal P&L guide was provided, while Street is at ₩105.0tn revenue, ₩83.0tn OP and 79.0% OPM.

Mixed feedback from investors on the call especially around Samsung production plans, and some pointing to fact they can’t accelerate buybacks due to ADR listing until next year. They did say 2H pricing should be better as Q2 impacted by some pushouts of higher value products. Overall, seems fine especially given stock down 40% from peak already.

Key Takeaways:

• DRAM: Blended DRAM ASP rose only ~30% q/q versus a higher-end expectation of +42.6%, despite conventional DRAM pricing remaining strong. The shortfall reflected higher exposure to lower-priced mobile DRAM, some fixed LTA pricing, legacy HBM3E mix and higher-value HBM4 shipments landing later in the quarter.

• 2H improves: HBM4 entered mass production in 2Q, with a full ramp planned for 2H, while HBM4 yield and quality are already nearing mature HBM3E levels. Mgmt expects HBM4, 1c conventional DRAM and a greater contribution from higher-value products to drive higher 2H bit growth, blended ASP and earnings.

• Demand: More efficient models and data-center leasing were characterized as signs of higher infrastructure utilization and faster monetization, not weaker investment. Mgmt expects CSP AI spending to remain solid beyond next year, although individual projects can move because of power availability and data-center construction constraints. Demand is broadening from HBM into server DRAM, LPDDR/SOCAMM and high-capacity NAND as agentic AI and inference expand.

• Major customers continue requesting more memory, and mgmt sees 2026 DRAM and NAND demand growing mid-20% and high-teens, respectively. PC and mobile are seeing temporary sales adjustments because customers cannot secure enough memory, rather than because of an underlying demand collapse. Advanced-node complexity and long fab lead times make a meaningful near-term supply-demand improvement unlikely.

• Capex: 2026 capex is expected in the high-₩40tn range, with M15X production pulled forward and Yongin Fab 1 capacity expansion accelerated. New packaging and NAND investments will be phased based on confirmed customer demand and investment efficiency. This supports the bull case on sustained shortages, but it also increases the bear debate around eventual 2027-29 supply normalization

• LTAs: SKHY has completed negotiations with around 10 customers, with additional discussions underway. Contracts are typically around five years and can include volume commitments, variable pricing mechanisms and deposits, improving demand and investment visibility

• NAND: eSSD revenue doubled q/q, with revenue from 30TB+ products rising more than threefold. Mgmt sees NAND becoming a core component of the AI memory hierarchy as inference, KV-cache offloading, near-GPU storage and AI data lakes require a mix of high-performance TLC, high-capacity QLC and SLC-mode products.

Bull vs Bear Debate

Read more

This is all happening with SPX down less than 3% from ATHs and AAPL hitting ATHs. Internet got the big bid today with mid cap names rallying (DASH +6%, CHWY +5%; CART +3.5% and even SPOT/NFLX +3% catching a bid) while hyperscalers languished again (META/AMZN flattish with GOOGL +2%). Software continued to be a beneficiary of the unwind & narrative change (open-source/lower model costs debate favors better margins for the SaaS players): HUBS +7%; WDAY +8%; MNDY +6% NOW +5% etc. PLTR -6% after Clev flagged below street commercial growth in Q2 and DOCN -6% getting looped with AI semis the big losers.

A quick snapshot of sentiment among investors right now:

Arun — who is always great at coming up with good mental models to frame outcomes/distributions — wrote a great post in TMTB Slack that’s worth a full read, which followed up on the thread we called out yesterday regarding what’s plaguing the AI trade. Great discussion in that thread from others as well. Here’s a brief summary of Arun’s points from Claude:

Read more

Software +1% benefiting from the unwind again early. Not much new to call out for the sell off with the familiar culprits being cited (more muddied AI semi narrative, momentum/leverage unwind, credit/equity issuance digestion, weaker price action earnings.) Big volume in SMH yesterday with puts outweighing calls by nearly 5 to 1. We had some thoughts on the current trading environment/narrative/setup in our EOD wrap yesterday in case you missed it.

On the earnings front, we get NXPI, STX, and KLAC today. Price action remains mixed in semis post prints with CLS volatile and now down 1% after being up high single digits following a big beat and raise while CDNS +3% with a beat against low expectations/sentiment, and GLW -17% after a mixed print.

We’ll hit earnings first (CDNS, FFIV, CLS, GLW) and then move onto the usual. Let’s get to it…

CDNS +3%: Modest Q2 beat but a much larger, broad-based guidance raise, with record backlog and accelerating AI/HPC demand lifting FY26 revenue growth to roughly 19%

Overall solid print against low expectations post Kimi with a call that went well talking up early agentic demand and INTC relationship and should help alleviate some of the concerns we’ve heard over the past 2 weeks around AI disintermediation.

The #s:

Q2 Revenue was $1.585B, +24.2% y/y (last q +18.7%) vs Street $1.577B, +~23.6%;
non-GAAP EPS was $2.11 vs Street $2.05;
non-GAAP OM was 45.5% vs Street 45.0%.

Q3 revenue guidance at the midpoint was $1.610B, +20.3% y/y vs Street $1.552B, +15.9%, with EPS of $2.04 vs Street $1.94.

FY26 revenue guidance increased $125M to $6.300B at the midpoint, +18.9% y/y vs Street $6.210B, +17.2%, while EPS rose to $8.10 vs Street $7.94. The raise was broad-based across core EDA, IP, hardware and SD&A, with record $8.1B backlog and early agentic AI traction strengthening the setup into FY27.

Key Takeaways:

• Revenue beat Street by less than 1%, but CDNS raised FY26 revenue by $125M, its largest single-quarter annual revenue raise. Roughly $7M came from the Q2 beat, about $58M from the stronger Q3 outlook and approximately $60M from higher implied Q4 expectations, suggesting the raise reflected forward pipeline strength rather than simply flowing through the q.

• Backlog reached a record $8.1B, up from $8.0B last quarter, while net bookings were roughly $1.7B for a second consecutive quarter. Notable bc 26 is a lower-renewal year in CDNS’s typical three-year contract cycle and the company normally draws down backlog during 1H

• Mgmt said the overall design environment is much better than one year ago, hyperscaler commitment to custom silicon has increased substantially, and analog, memory and consumer semiconductor customers are also improving. Startup activity has reaccelerated across AI, networking and CPUs

• Agentic AI is beginning to contribute, but the model still assumes no sudden revenue step-up: CDNS is seeing more evaluations, pilots, production deployments and early add-on business. Mgmt expects monetization through separate agent licenses and workflow products, plus increased invocation of the underlying EDA engines, but explicitly said guidance does not assume a sudden step function.

• IP revenue grew 43%, driven by interface, memory and foundation IP for AI/HPC, with mgmt saying most of the growth was organic. Better PPA, a tighter focus on leading-node “star IP,” and broader foundry relationships are producing wins CDNS would not have competed for two years ago. The offset is that IP revenue remains timing-dependent and could moderate after a very strong Q2

• Hardware posted another record quarter, added 12 new logos and remains constrained by CDNS’s ability to build systems rather than by customer demand. Inventory rose roughly 23% q/q as CDNS procured components ahead of demand.

• Mgmt called the new INTC engagement 100% incremental and said some benefit occurs in 2026, but most is still to come. Near-term monetization appears concentrated in IP, NRE and DTCO work, with potential incremental EDA and hardware opportunities as the relationship expands across Intel Foundry and product groups. Samsung positioning has also improved materially.

• On Margins, FY26 non-GAAP OM guidance increased 25 bps to 44.25% at the midpoint, but implied 2H OM is approximately 43.5% versus 45.1% in 1H. Mgmt attributed the step-down to Intel development and Hexagon/SD&A integration, called it targeted investment rather than model deterioration, and expects acquisition profitability and margins to improve in 2027.

Bull vs. Bear Case

The structural bull case is that CDNS is a mission-critical engineering R&D platform positioned at the intersection of semiconductor complexity, hyperscaler custom silicon, advanced packaging, memory bandwidth and system-level simulation. CDNS is not primarily dependent on semiconductor unit shipments or fab CapEx. Its revenue tracks the number and complexity of designs, the need to improve power, performance and area, and the amount customers spend on engineering productivity. As Moore’s Law slows, customers must rely more heavily on architectural innovation, advanced packaging, emulation and software optimization, increasing the value of CDNS’s full-stack portfolio. The recurring model, high switching costs and 40%+ non-GAAP margins reinforce the argument that CDNS deserves a premium valuation.

This quarter materially strengthened that case. CDNS delivered its largest annual guidance raise despite entering a lower-renewal year, backlog increased rather than declining, and the raise was supported by four or five growth engines rather than a single large hardware shipment. Core EDA grew 18%, IP grew 43%, hardware remained supply-constrained and SD&A grew 37%. The Intel relationship is fully incremental, positioning at Samsung has improved, and software add-ons are already benefiting from both customers designing more AI chips and customers deploying AI inside the design workflow. Agentic AI also looks more additive than cannibalistic because customers purchase agent licenses on top of underlying EDA software. Bulls will say CDNS can plausibly grow FY27 revenue 15%–17% to roughly $7.25B–$7.40B, vs the current Street expectation near $7.02B, if agentic add-ons, IP, Intel and SD&A remain strong. With margin re-expansion, that supports approximately $10.50-11+ of EPS in FY27

The bear case begins with valuation and uncertainty over who ultimately captures the agentic AI economics. CDNS trades at roughly 36x FY27 Street EPS, already pricing in durable double-digit growth. High level: sophisticated hyperscalers and semiconductor companies may build their own orchestration layers, and CDNS’s traditional time-based licenses do not automatically monetize every incremental engine invocation. CDNS must prove customers will pay separately for its agents and premium workflows, rather than simply consuming AI functionality inside existing license pools. The transition toward more cloud or consumption-based architectures could also require investment and changes to the commercial model.

This quarter did not fully resolve those concerns. The Q2 revenue beat itself was small, with a disproportionate contribution from IP and upfront revenue, while recurring revenue missed one cited estimate and first-half recurring mix was below the approximately 80% target. Mgmt still did not quantify agentic AI revenue and explicitly said there was no sudden step function embedded in guidance. IP is timing-dependent, hardware can be lumpy, China’s growth benefited from unusually easy comparisons, and China exposure has risen to around 15% of revenue. Meanwhile, implied 2H margins step down despite higher revenue because of Intel and Hexagon investments, and the outlook assumes export controls do not tighten.

FFIV +1%: A clean beat-and-raise led by Systems and gross margins, but the debate shifts from whether the refresh is working (it is) to how much growth survives into FY27

Overall solid print with systems much better, expansion rates improving, and direct-AI customer growth accelerating as bulls look to FY27 for sw accel. Beras will focus on how GMs look going fwd and how long the refresh lasts.

The #s:

Q2 Revenue was $865.1M, +10.9% y/y (last q +11.0%) vs Street $834.6M, +~6.9%,
non-GAAP EPS was $4.73 vs $4.00, although a 12.4% tax rate contributed roughly $0.30-$0.40 of upside.
Systems was the standout at $240M, +32% y/y (last q +26%) vs Street expectations of roughly +15%;
Software was $223M, +7% y/y (last q +17%), roughly in line.
Non-GAAP gross margin was 84.2% vs Street 82.8%,

FQ4 revenue guidance of $870-$890M was above Street $856.9M.

FY26 revenue growth was raised to 9%-10% vs Street 7.7%, with EPS raised to $17.21-$17.33 vs Street $16.49.

Key Takeaways:

• Customers are adding capacity at the time of refresh and making capacity purchases outside the refresh, driven by AI inference traffic, data-center modernization and sovereignty. More than half of the hardware installed base reportedly remains on iSeries ahead of the Q2 FY27 end-of-support date, supporting continued activity over the next several quarters.

• Mgmt explicitly acknowledged that part of the current 30%+ hardware growth is cyclical and is not guiding anything close to that level beyond the refresh. However, mgmt believes sovereignty, on-premises AI traffic, competitive takeouts and platform consolidation should create a shallower post-cycle slowdown than prior product cycles. Mgmt did not update the formal FY27 revenue framework on the call.

• Direct-AI customer count rose 50% q/q, AI-security customers doubled, and deal sizes range from hundreds of thousands of dollars to eight figures. Mgmt did not update the previously disclosed direct-AI revenue level of more than $50M and cautioned that customer growth should not be assumed to equal revenue growth. The larger current benefit may be indirect, as enterprise inference traffic drives higher-capacity ADC purchases and expansion.

• Mgmt said advanced financial-services and safety-sensitive customers are specifically budgeting for AI-model, agent, governance and discovery security. AI-powered WAF has already reached 15% of the Distributed Cloud WAF base, with 75% of adopters operating in blocking mode, while AI-security customer count doubled in the quarter

• Software growth decelerated to 7% from 17% last quarter, reflecting the weak FY23 renewal cohort. Mgmt said the legacy SaaS and managed-services transition is effectively complete, Distributed Cloud trends are strong and the substantially larger FY24 renewal cohort should support double-digit Software growth in FY27, with more of the acceleration expected in the second half.

• The 84.2% gross margin and 83%-84% FQ4 guide benefited from higher-performance Systems mix, lower-than-expected memory and SSD costs, supplier diversification and improved discount discipline. Mgmt clarified that it did not formally reiterate FY27 gross margin of 80%-82%; it simply declined to update the framework because memory and broader BOM signals remain mixed. Component-cost pass-through pricing should contribute more meaningfully in FY27.

• Mgmt described underlying demand as healthy across all regions and said APAC’s 11% reported decline reflected shipment timing rather than weaker orders. Tariffs were not specifically addressed; the explicit cost issue was memory, SSD and broader component pricing

• EMEA grew 27% y/y, and mgmt said the trend of moving workloads from global hyperscalers to local or private infrastructure has expanded from Europe into the Middle East and Asia. FFIV also cited takeouts of SaaS-only, cloud-native and incumbent hardware solutions, supported by better performance, throughput, capacity and power efficiency

Bull vs. Bear Case

Read more

Read more

Big week for earnings this week. Here are some bogeys & set ups for those reporting Mon/Tues…

Read more

Big week for Tech on the earnings front with META/MSFT/AAPL/AMZN among others on deck (we’ll have some bogeys out shortly). FOMC is on Wednesday. Going to be a fun one.

Asia generally up overnight: TPX +1.37%, NKY +0.5%, Hang Seng +0.98%, HSCEI +1.14%, SHCOMP +1.15%, Shenzhen +2.93%, Taiwan TAIEX -0.05%, Korea KOSPI +0.97%. SK Hynix +2.4%, Samsung +1% but Kioxia -2.6%…CXMT (Chinese DRAM maker) +466% overnight on debut in Shanghai, close to a $500B market cap.

In the U.S., Semis +2.2% leading the way higher with Internet/Software +1.3%

Good to be back. Let’s get right to do it…

Nvidia in Talks With OpenAI to Guarantee $250 Billion Financing for Data Center - WSJ

WSJ reports Nvidia is considering a roughly $250 billion guarantee to support OpenAI’s lease of a 10-gigawatt Ohio data-center campus being developed by SoftBank’s SB Energy, with the full project potentially exceeding $500 billion including chips. Nvidia may separately finance as much as $350 billion of GPU purchases, underscoring the increasingly circular funding structures supporting AI infrastructure. The project would give OpenAI more direct control over compute capacity, but also raises concerns about financial risk if AI growth or investor appetite weakens.

STX/WDC: Morgan Stanley Sees Pricing-Driven Beats and Reignited HDD Bull Case

Read more

Read more

Good afternoon. QQQs -1.6% with plenty of red on the screens across the board. Semis -4% led the way lower as the unwind/risk off continues, with Kimi 3 benchmarks near Opus 4.8 levels adding some fuel to the fire.

In TMTB Slack, we were finally able to communicate a more cohesive narrative picture of what has been going on in semis recently, something we’ve tried but struggled to clearly express over the last couple of weeks:

It's also the forward looking fears why names trading down on good news. What settles the model competition/open source debate? Right now, it's very open-ended with many questions around how the landscape evolves, particularly in respect to the labs viability, and how that flows into semi spending. No one has a crystal ball here -- we could debate endlessly, just look at X -- plenty of intelligent arguments arguing several paths (hiccup in spend, training spend slowing, opensource demand picking up the slack, everyone becomes a neocloud, etc.) So we're in an environment where good #s & datapoints (capex, token growth, & lab ARRs included) being brushed aside because no one is particularly confident on the broader questions. Just a tricky environment overall. Today was the first day where I felt the vibe from some bulls as acceptance we weren't going back to the May-early June days and that we might be entering new paradigm for the time being (not necessarily negative, just different which requires a different approach to AI semi investing like we've been writing about in our wraps/weekly recently)….part of the confusion is bulls want to brush aside the broader arguments by saying this is all mo unwind / leverage / positioning, but I think Tech investors are grappling with some very real questions that have arisen over the last month re: landscape

Arun added on in his own inimitable mental framework:

Read more

Good morning. QQQs -1% as the Momentum/AI semis unwind continues with Semis -3% and Software +70bps as China AI in conversations again today (Kimi K3 rumored to surpass Opus 4.8). Hyperscalers outperforming early as well with AMZN/GOOGL/MSFT all in the green. Korea -6% overnight with memory names down big: Kioxia-15%, Hynix-10%, Samsung-7.6%, Nanya-8% and US names following early: SNDK -7%; MU -5%; STX -6%.

TSM rev #s, capex, & AI commentary were solid although GMs missed bogeys and stock is down 4-5% early (more sell the news reactions following other AI semi prints like Samsung, ASML, and DOCN).

Lots to get to, so let’s get to it…

TSM: Capex and FY26 guide both above bogeys, but GMs misses bogeys for both Q2 and Q3

Bulls will like the better than expected FY26 guide of “slightly above 40%” vs bogeys in mid to high 30s and implied 48-49% for Q4 growth given the Q3 guide and positive AI commentary from mgmt; while bears pointing to weaker GMs and the below comment on weaker consumer markets

Read more

Good afternoon. QQQs -27bps as Megacap +3% & Internet/Software +1.5%+ led the way higher while semis -2% lagged.

Today’s action and conversation have us going back to what we wrote over the weekend:

Sometimes the market vibe is clear ( the early “HOT AI SUMMER”) and it’s time to press the accelerator down, sometimes it’s not. We think it’s the latter right now. The big picture remains the same (yes, we’re still LT AI infra bulls and AI datapoints remain strong), but it’s hard for us to come up with a coherent and broad near/medium-term narrative/direction around how AI semis are going to trade (more on that below).

It’s not only factor whiplash investors are suffering from, but narrative whiplash (which usually follows the former). Just today, the script flipped several times. ASML’s print & positive Codex vibes got the AI semi bulls excited in the early morning. Then memory sold off and hypers rallied and I heard variations of the following thought expressed, which was also fueled by Buffett saying he initiated the GOOGL investment, giving credence to the Hyperscaler ROI argument:

(We generally find this narrative/argument compelling and the charts are looking good, particularly META and AMZN’s). Then semis bounced off the lows as Trump defended the datacenter buildout and hypers flat-lined and the narrative started to shift. “Wait, no” some said, maybe it was all about Kimi 3 and that’s what drove the AI sell-off this morning:

The whole back and forth today was just a microcosm of what has been going on in some manner for a few weeks now.

It’s not that the narratives don’t make sense — most are bullish, and many are nuanced and intelligent when not. I find myself nodding to many of the arguments made on X around open source, cheaper intelligence, hyperscaler ROI and other debates on both sides. The outright baffling bearish AI takes have largely disappeared (maybe LLMs have made everyone sound more intelligent or maybe I’ve gotten better at weeding out the noise). Readers know we love our narratives here at TMTB, but narratives need to meet at the interplay of vibes, price (both level and price action), and conviction to drive confidence in a specific direction, and investors are struggling with all those 4 aspects at the moment, partly catalyzed by the pace of new AI developments and shifts that have investors in a bit of a “wait-and-see pattern” before earnings; partly because of factor whiplash; and partly catalyzed by what I call price narrative dissonance (eg, why is ORCL hitting 52wk lows as Codex/OAI vibes ramp?; why was AAPL +4% and DELL -10% today and both being attributed to memory prices peaking? why is NVDA/AVGO at 15x? what’s with the hyperscaler ROI whiplash? etc etc). That’s not to mention the generally broad weak AI semis price action despite seemingly positive developments.

Things won’t always stay like this — investors as a group generally like to be very dramatic. A choppy environment is not the right one to be over-reacting either way. It was only a month ago when the narratives and price action were clean. Things will keep shifting, and our job as investors is to adjust accordingly. We should enjoy the HOT AI SUMMER vibes when they’re here. And adjust our mental models, positioning, attention, and risk when they’re not. Maybe earnings will clear some things up for us.

Let’s get to it…

Read more

Good morning. QQQs +30bps as some turbulence on the Iran front as US carried out another round of strikes near Strait of Hormuz overnight and Trump said US may hit power plants and bridges next week unless Iran makes a deal. Oil is holding fairly steady. In Tech, semis +1% while software/internet are flat. Memory names lower with SNDK -2.5% and MU -1%.

Overnight, Asia mixed as Korea and memory names were strong following Hynix’s big move up in U.S (although underperformed KOSPI).: TPX +.22%, NKY +1.49%, Hang Seng +1.4%, HSCEI +1%, SHCOMP -0.29%, Shenzhen -0.68%, Taiwan TAIEX +2%, Korea KOSPI +6.24%. Samsung +2%. SK Hynix +4%.

Lots to get to, so let’s get to it…

ASML +3%: Very solid beat and raise above street and buyside. Q3/’26 guide and ‘28 commentary better while co points to large capacity increases. Only nit is ‘27 EUV slightly below bogeys

Revenue came in at €9.3B vs. Street expectations of €8.8B, gross margins were 54% vs. 52% expected, helped by higher sales volumes and business mix. Q3 GMs guided to 55-57%, well ahead of street in low 50s. ASML raised full-year guidance to €43-45B of revenue from €36-40B previously. Gross margin guidance also moved materially higher to 54-56% from 51-53%

The co. has indicated that they will raise both EUV and immersion DUV capacity in both 2027 and 2028 by 30% which would result in approx. 85/110 EUV and 169/210 Immersion DUV tools in 2027/2028, which is higher than the high on the street for ‘28 and would imply more than €65 in EPS for 2028, potentially even more given strong Installed Base Management revenue that ASML is already seeing. Only nit I’m hearing negative this morning is that the co. does not achieve 90 EUV tool capacity in 2027 (co guided 30% on 65 = 85), which is slightly below 90 bulls were hoping for.

Mgmt said demand remains exceptionally strong across logic and memory. AI-related investment continues to accelerate as customers pull forward capacity additions to meet robust demand for advanced logic and memory chips. In memory, ASML said current DDR and HBM pricing clearly signals the need for more supply, prompting customers to accelerate expansion plans. Company talked up very strong intake continuing across 1H and customers looking to accelerate capacity expansion. The company is seeing greater High NA readiness and points to Intel use at 18A earlier than expected. Also announces CMD for June 10th 2026.

AEHR +30%: Record Bookings and Backlog Drive FY27 Outlook Well Above Street

AEHR reported F4Q revenue of $18.8M, +34% y/y and roughly in line with Street with key record bookings of $60.7M lifted effective backlog to $100.6M, supported by expanding AI accelerator, silicon photonics and SiC programs. FY27 revenue guidance of $130M–$150M implies roughly 160%–200% growth and was about 50% above Street at the midpoint, with non-GAAP EPS guided to $1.80–$2.20 versus Street near breakeven.

PYPL +20%: Stripe, Advent offer to buy PayPal for more than $53 billion at $60.50 a share sources say - Reuters

Reuters:

The proposal follows an initial approach made in early April, the sources ​said. Stripe and Advent have not received a response from PayPal and are seeking to advance discussions in the coming weeks, the sources added.

Under the proposal, ​Stripe and Advent would jointly own PayPal, with each holding an equal stake, rather than breaking up the company, the people ⁠said. There is no certainty the approach will result in a transaction, they added.

PYPL: Bernstein Says Stripe Bid Could Reshape Global Payments but Faces Financing and Regulatory Hurdles

Bernstein sees strategic logic in Stripe/Advent’s reported $60.50-per-share bid for PayPal, given PYPL’s valuable Braintree, Venmo and branded-checkout assets, but believes a higher offer is unlikely because the proposal already values the company near 15x FCF versus a pre-deal multiple below 5x. Financing would be substantial—roughly $50B of committed debt plus rollover equity—and could push leverage toward 5x–6x, while integration complexity and regulatory scrutiny would also be significant. Bernstein sees a greater chance of selective asset sales or a broader bidding process than a clean full-company transaction at a materially higher price.

AMD: Rosenblatt names top long idea going into earnings and raises target from $490 to $665

We are expecting AMD server CPU, EPYC, to deliver over 70% y/y revenue growth at above corporate average gross margin,” and “we see AMD’s Venice server CPU as dominating high-end market as Intel’s Diamond Rapids server CPU has been delayed,” writes analyst Kevin Cassidy

AMD: UBS Raises PT to $700 as AI GPU Estimates Move Higher Ahead of AI Day

Read more

Lots to get to today so we’ll dive straight in…

AI/SEMIS

• NVDA +4% acting better since the middle of last week. Keybanc was out today saying their Asia checks show NVDA’s CoWoS supply in 2027 to 1.1M We wrote a little about why stock has been acting better over the last few days in our weekly Sunday:

First, Jensen had a couple NDRs where he sounded especially bullish. One line that stuck out from the MS note: Notably, one of the more prominent frontier models was developed mostly on an ASIC, so NVIDIA had minimal exposure, but that has risen to close to 50%. Implication here of course is Anthropic. The other change is on the narrative front. We’ll quote ISI’s K. Rippey again as he puts it better than we could have: “Nvidia needs 2 things to work: 1) a large number of viable frontier lab customers; 2) lock-in among customers without their own dedicated ASIC programs… Anthropic’s rise in 1H26 largely challenged both of those premises. What we’ve seen from META + xAI over the past 2 weeks has acted to rebuff that bear case.” While NVDA is a staple in most LO and HF portfolios, it has also become a funding short over the past several months for more high-beta semi trades like memory while the NVDA narrative has become more muddied. That dynamic seems to be reversing in a somewhat choppy semis environment. Talking to investors and feeling the vibes, we are finally getting that sense of resignation that NVDA will just “fail its breakout” and “remain choppy.” In other words, put into words what people are feeling: “Why add back to NVDA if I’m just going to be disappointed again?” (“The seductress")…which all means: maybe the pain trade is now up rather than chop?

The way I’d frame the narrative is that we’ve had a lot of “bad’ news over the past 6-9 months (margin/TPU/moat fears, Kyber/V Rubin push out, etc.) and now we are finally getting some incremental positive tailwinds to the narrative. What I’d add to the above quote re: narrative is increasing realization that NVDA could actually be one of the biggest winners from the main narrative gripe investors are grappling with right now: opensource.

For example, this Clark Tang (from Alitmeter) tweet has gotten passed around a ton the past couple of days, where the implication is NVDA hardware is the biggest winner from Opensource:

Despite NVDA being flat for 9 months, we’ve had some decent PnL playing some short-term narrative shifts along the way. Still: it’s NVDA, and as I told a buddy of mine this afternoon — the opportunity cost of putting that $ into NVDA over the past year instead of something like SNDK means the trades were likely net negative.

Chart looking more interesting if we can get above $212/$213

Hard to imagine NVDA having a nice run as semis chop around, but we’ll see…the seductress strikes again…

Disclosure: We’re long some ST calls in addition to our LT long.

• SKHY +27% the big winner today. Lot of questions on why the big move — didn’t hear anything specific other than Hynix overseas putting in a hammer candle. ADR Premium now 45%+ although that likely reverses somewhat once Asia opens. Some good color in this Acadian post around premiums and ADR’s using TSM as a precedent:

• AVGO +1.3% as MS was out defending TPU share saying they expect AVGO to retain roughly 80% of Google’s TPU share over time vs bears fearing 50%, with MediaTek likely gaining a modest, additive position rather than disrupting Broadcom’s scaled platform.

Read more

In the last few weeks of June, we saw purchases to secure supply-constrained infrastructure ahead of expected price increases. While we anticipated some supply chain related impact in our expectations, we did not anticipate the magnitude of the capex reprioritization. In addition, clients were distracted with rapidly-evolving, industry-wide cybersecurity concerns in the quarter.

The #s:

Revs $17.2B, vs street at $17.9B
Software Revs +5% y/y vs street at double digits
Consulting Revs flat vs street at +LSD
Infra Revs -7% vs Street at -1%
RHT accelerated to 11% the one bright spot
EPS: $2.93 vs street at $2.97

Lots of software names down high single digits, particularly those in the “AI loser bucket”: WDAY -9%, HUBS -9%, IT Services -8%, INTU -6%, etc…

IBM wasn’t the only one calling out higher component costs: ERIC -9% also lower after warning elevated component costs could pressure margins going forward…

In Asia, things were generally green: TPX +0.79%, NKY +0.74%, Hang Seng +0.52%, HSCEI +0.46%, SHCOMP +1.36%, Shenzhen +2.26%, Taiwan TAIEX -1.42%, Korea KOSPI +0.73%. Memory names higher with Samsung +4% and SK Hynix +6%. Softbank +3%.

Lots to get to so let’s get to it…

AAPL: KeyBanc Downgrades to Underweight as Hardware Growth Slows and Valuation Stretches

KeyBanc downgraded AAPL to Underweight from Sector Weight with a $250 price target, citing June hardware spending down 2% m/m and continued below-trend growth despite easier comparisons. The firm sees risks from slowing iPhone builds, weaker U.S. upgrades as carriers pull back subsidies, and softer Mac, iPad and Wearables expectations, while Services growth is also expected to decelerate to roughly 7% in FY27 versus Street near 12%. With AAPL trading around 24.5x FY27 EV/EBITDA and 35x P/E, KeyBanc believes the stock’s premium is difficult to justify absent a stronger hardware cycle or Services reacceleration.

AVGO: Morgan Stanley Defends TPU Share as MediaTek Displacement Fears Look Premature

Read more