In Techlandia, strength in semis continues although more bifurcated this morning with HDDs/Memory leading us higher (WDC/STX +8-10%) and SNDK/MU +2-3% while CPU names softer (ARM -2%, INTC -1% although AMD +50bps)

Asia mixed overnight: TPX -0.21%, NKY +0.13%, Hang Seng -1.4%, HSCEI -1.62%, SHCOMP -0.11%, Shenzhen +1.02%, Taiwan TAIEX +0.91%, Korea KOSPI +2.11%. Sk Hynix +4%; Softbank -50bps

Relatively quiet morning, but let’s get to it…

SPCX: SpaceX to buy AI coding startup Cursor for $60 billion in enterprise push

Reuters:

SpaceX had been eyeing Cursor ⁠for several months. The company said in April it had secured an option ​to either acquire San Francisco-based Cursor for $60 billion later this year, or pay $10 ​billion for their new partnership. It was not immediately ​clear if ⁠the deal would affect SpaceX’s agreements to rent out its data centers. The company has in recent weeks struck deals with Anthropic and Alphabet-owned (GOOGL.O), opens new tab Google to lease ⁠cloud computing ​capacity worth roughly $26 billion combined on an annual ​basis.

Both deals include 90-day termination clauses, meaning SpaceX could quickly reclaim computing capacity if needed.

Anthropic: Anthropic, Trump Officials Seek Deal on Restoring Powerful Model Access

WSJ:

Talks between Anthropic and Trump administration officials continued Monday without a deal to resolve the security concerns that pushed the White House to restrict access to the artificial-intelligence company’s latest model, increasing urgency on both sides to find a resolution.

“This follows numerous virtual meetings we have held with the government since the administration’s initial outreach. Both parties are working quickly to get this resolved,” an Anthropic spokesman said in the company’s first public comments since Friday. “This is part of our ongoing commitment to working alongside the administration toward our shared goal of protecting U.S. critical infrastructure and the U.S. lead in cyber defense.”

OpenAI: OpenAI spending hit $34bn last year ahead of planned IPO

FinancialTimes

Audited financial figures confirmed by people familiar with the matter show the company spent about $19bn on research and development in 2025 and nearly $6bn on sales and marketing, as well as other costs.

OpenAI booked about $13bn in revenue last year. By the end of 2025 it was generating $2bn in monthly revenue, up from $1bn a quarter at the end of 2024, making it one of the fastest-growing businesses in history. But heavy spending contributed to a nearly eightfold increase in the net loss attributable to OpenAI, which soared from $5bn in 2024 to around $39bn in 2025. Stripping out the charge and other non-cash expenses, such as stock-based compensation of staff and computing credits from Microsoft, OpenAI’s losses were $8bn

AMD/NAND/DRAM: Wells Fargo Says MEXT Could Make Flash Behave More Like DRAM for AI Workloads

Read more

Will be interesting to see what comes out of it, whether some legislation around foreign access or other, some USG stake, or maybe nothing at all! Either way, I hope to get Fable back soon!

In Tech, Semis +5% led the way higher with HDDs and Neoclouds leading the way while TMT Mo was +4.5%. Hyperscalers rallied while midcap internet also saw strength.

A little late on this wrap, so let’s get straight to the recap…

Read more

Anthropic is headed to the White House this morning to try to soothe things over regarding Fable/Mythos export controls.

Asia generally up: TPX +3.03%, NKY +4.99%, Hang Seng +0.5%, HSCEI +0.02%, SHCOMP +1.61%, Taiwan TAIEX +2.78%, Korea KOSPI +5.2%. Samsung +5%; SK Hynix +7%; Softbank +10%; Kioxia +12%

In The U.S., semis leading the way higher with NBIS +8%; Memory/HDDS +7-8%; semicap +5% while software and internet are up but lagging.

Hope everyone had a good weekend. Let’s dive straight in:

Anthropic: US issued export controls on Fable/Myhots post Friday. “They screwed us”: Personality clashes sent Anthropic’s models offline

Axios

On Thursday, Amazon CEO Andy Jassy called Treasury Secretary Scott Bessent expressing concerns that Anthropic's most powerful models, Mythos and Fable, could be jailbroken.

The administration official said Anthropic knew a jailbreak could happen and chose to distribute it anyway: “They came to every fork in the road and took the wrong fork.”

Anthropic failed to “honor” a recent cyber executive order, administration officials claim, and the company’s purported failure to take the matter seriously led to its most powerful products being scrubbed from the internet.

“Everybody said Anthropic was a bad actor. Some of us said it was time to give them a chance. Now those people are questioning that. They screwed us,” an administration official said.
“It’s like they just speak in different languages,” the source said, adding that the company has simply not figured out how to communicate with this administration.

ROKU: Fox to Buy ROKU for $160/share

DDOG: Truist Upgrades to Buy, Says AI Adoption Is Outrunning AI Optimization Concerns

Truist upgraded DDOG to Buy and raised its PT to $300, arguing investor concerns around AI optimization are being overwhelmed by the urgency of enterprise AI adoption. Recent customer and partner conversations suggest Datadog remains in the early innings of AI-driven growth, with non-AI-native enterprises increasingly deploying agents that require more telemetry, monitoring, and observability. The firm also highlighted improving confidence in the durability of DDOG’s relationships with frontier AI labs, reducing a key overhang on the stock. While valuation has expanded, Truist believes Street still underestimates the durability of consumption growth and the size of the opportunity as AI adoption broadens across the enterprise.

WDC: Morgan Stanley Raises PT to $650, Says HDD Demand Visibility Now Extends Into the Next Decade

Read more

From what we know, it seems like AMZN brought a jailbreak to the attention of the USG and Anthropic said the jailbreak used to justify the order was narrow, surfaced only known and relatively minor vulnerabilities, and offered nothing unavailable through other public models. Many unknowns as we sit here Sunday (including how long the export directive lasts). We’ll get more reporting over the next few days. We don’t know all the facts yet and there are plenty of takes on X exploring all sorts of different angles if you want to go down that rabbit hole, but that’s not what we want to spend time on today.

What we can be sure of is that it continues to bring the sovereign issue to light even more. It’s not a coincidence this directive comes on the back of increased National Security focus on frontier models. We had rumors of USG taking a stake in the frontier labs last week. We had a month of US agencies tightening up cyber security risks with Mythos. On June 2, Trump signed an executive order on AI and cybersecurity that, among other things, invites frontier developers to give the government up to 30 days of early access to new models before release. Days later, on June 9, reporting surfaced that China is preparing to spend roughly $295 billion over five years on a state-directed buildout of domestic AI data centers. While no one knows how this continues to evolve, I think the news this weekend is just part of an ongoing pattern of the U.S. government increasingly looking at frontier AI from a strategic asset perspective vs. a purely commercial one.

This likely accelerates the already ongoing trend of enterprises not having single-model dependency. Now it increasingly becomes table stakes. This also vindicates Europe’s sovereign-compute push (Chips Act 2.0, CADA, etc) and potentially pushes more sovereign countries to do the same. More questions to think about: does it also further increase the push for domestic chip production (INTC the obvious main beneficiary here)? Does this further slow the pre-release gating the Trump exec order defined or does the cadence stay the same but just get adjusted forward?

In our minds, it’s very hard to imagine the government actually wanting to slow frontier model progress given it’s becoming an increasing national security issue, but how model releases look and how they get accessed will be different in the future. For the hyperscalers, though, the open question is whether a government gate actually disrupts the stack at all, or simply thickens the compliance layer already wrapped around it.

These issues and questions aren’t going away. But assuming this particular directive gets resolved fairly quickly (and we are assuming it does according to initial reporting) —

— and doesn’t snowball into something unexpected (still possible), we don’t think this affects the AI semi trade in any significant way. It’s possible there might be a very small air pocket of token demand if this lasts more than just a few days however as I’ve heard several people ask: “why spend time coding with Opus 4.8/GPT 5.5, when I can just hold off and finish something better in a lot less time once Fable comes back online.”

Ok now onto the good stuff. Which stocks come out of this looking more attractive?

You are reading a TMTB Weekly, part of TMTB Pro. For more info or a trial to TMTB Pro, please contact sales@tmtbreakout.com

Read more

#1 — SpaceX = “A Must Buy, Must Own” for Institutional Investors

Gerstner: “I think we’re all pretty AGI-pilled. And if you’re AGI-pilled, that means we got to build a lot more compute than the world thinks and that these models are going to be a lot more valuable than people think. You combine that with their core business — I don’t know another entrepreneur or another business that’s a better bet on the future than SpaceX. And so I think for most institutional investors, it’s a must buy, a must own. It’s set it and forget it, in order to have a real bet on both the space and the AI future.”

“When I look at the bull-bear case on the IPO, the bears are looking at last year’s revenue, say it was $18 billion, and the forecast from the banks of $160 billion three years from now, and they’re saying not many companies in the history of the world have basically 8x’d their revenue over three to four years. When you break it down as an analyst, first principles, part by part — Starlink looks totally doable. AI compute terrestrially looks totally doable. The model itself after the acquisition of Cursor looks to me like it could be an upside. Three years from now, there’s a decent chance everybody’s like, ‘Oh my god, that was super obvious.’”

“It was a hundred times trailing revenue. Well, after the deals they signed, I think it’s 39 times. They added $29 billion in a month. Have you ever seen that happen? Never.”

#2 — “Elon Web Services”: #4 Hyperscaler in 30 Days, Best-in-Class Monetization

Baker: “Clark’s analysis shows that XAI’s deal with Google for cloud computing generates more operating profit per gigawatt than Anthropic, than Meta, than Google, than OpenAI. Freda calculated a 55% IRR on Colossus 1. If you can borrow money at 6, 7, 8% and invest in something with a 55% IRR — I’m not the most sophisticated thinker, but that math maths.”

“Are they the number four or number five hyperscaler today? After the Google deal it will be number four. In 30 days, we went from not being an AI hyperscaler to being number four. And we passed a lot of companies, including Oracle.”

Fox: “The implied monetization rate on that [$160 billion] number is something like $14 billion per gigawatt per year for the AI business. They just signed Anthropic at 22 to 23. They just signed Google at 50. So I think you can invest behind the AI business terrestrially and still be excited about it.”

Gerstner: “Jensen said Elon is an N of one. What they achieved is singular — 100,000 GPUs, easily the fastest supercomputer on the planet. A supercomputer would normally take three years to plan and one year to get working. We’re talking about 19 days.”

#3 — Data Centers Are Not Commodities: First-Principles Design

Baker: “There is a belief that these data centers are commodities. I do not share that belief. In the same way that Elon was able to re-engineer a rocket from first principles and make it reusable, he engineered an electric car from first principles — I think he looked at data center design from first principles and designed something fundamentally different.”

“I did actually ask the team — I said, ‘Hey guys, maybe be a little less public about things that are very obvious to you about how to design a data center, but are revelations to other people. What you’re doing is maybe more differentiated than you perhaps realize.’ And that’s how he was able to do it in 122 days. Speed is literally cost — every day you’re paying electricians and plumbers.”

Tang: “There’s only maybe two or three players now that can actually reliably engineer behind-the-meter data centers. If you’re GE Vernova and you only have a certain number of gas combustion engines — you can sell them to XAI or one of these startup NeoClouds. Who are you going to sell them to? Everyone starts making more money when the GPUs get energized and sold faster. Speed is money for all of the suppliers.”

#4 — Orbital Compute Economics: $5B/GW vs. $20-25B Terrestrial

Fox: “When you back into the numbers, you get to something like five megawatts of capacity per Starship launch. The math you get to is about $5 billion per gigawatt of capex to put these in space. For comparison, terrestrially — the switch gears, the generators, the transformers, the shell, getting the power — that today is about $20 to 25 billion per gigawatt. So we’re talking about a 5x reduction in cost on half of your bill of materials.”

“Two-stage reusability — the cost per kg comes down significantly. We’re talking about going from $1,500 per kg on Falcon to $250, something lower. The more you can reuse the rocket, the more that price comes down, because you’re depreciating the cost of the launch, and eventually you asymptote to the cost of the fuel.”

Gerstner: “It costs $60 billion to put a gigawatt on the ground today — call it $35B the GPUs and silicon, $25B the land, shell, power, and cooling. I would hypothesize those elements are probably going to be inflationary. Space, power, cooling are effectively free in space. You’re talking about putting a gigawatt into space for $30 billion with lower operating costs, versus $60 billion that’s inflationary. As long as the reliability and maintenance is not dramatically lower, the math maths.”

Fox: “I don’t think orbital compute is necessary for the IPO valuation, but it’s certainly important.”

#5 — The Cursor Acquisition: The Most Underappreciated Lever

Baker: “My understanding is that Cursor and Anthropic have more tokens of proprietary coding data than anyone else — and each have more tokens of proprietary coding data than exist on the public internet. Cursor used Kimi K2.5, used their own private data, did some RL, some supervised fine-tuning, and got a really good model. Then they spent three weeks in the Colossus 2 cluster and got a model that 12 days ago was Pareto dominant with Composer 2.5. It suggests the Cursor data is very valuable for coding, and that XAI/SpaceX has a shot at being a real player in coding.”

“Right now the Grok 4.3 1.5 trillion parameter model is training. One would hypothesize based on scaling laws that will be a better base model, and then the Cursor data is being injected into the pre-training process, not just reinforcement learning. That is going to be a very important data point. And once you are at multiple places on that Pareto curve, if you have compute, you can scale really rapidly.”

Gerstner: “I think the thing that’s getting lost is they’ve dramatically advanced their capability when it comes to building a frontier model. Michael and the team at Cursor — this is an extraordinary team that he just downloaded right into SpaceX. If there’s an upside surprise, this is the place that’s getting the least amount of attention and could have the biggest upside surprise.”

Tang: “From a lot of my conversations, it looks like they’ve secured maybe up to 20% of Vera Rubin capacity — especially in the early days when these chips are very scarce.”

#6 — Fable 5 / Mythos and the Noam Brown Insight: “We Don’t Know How Smart These Models Are”

Baker: “Nobody has run Mythos for a year continuously, and we may never know how smart each generation of models actually is or was — because we don’t have time to appropriately evaluate their intelligence before the next model comes out. This is a profound statement.”

“Imagine Albert Einstein had just thought about fundamental physics 24 hours a day. He doesn’t have to eat, doesn’t have to sleep, never gets old, never has diminution of intelligence — and he thought for one year. We might already have solved a lot of these intractable problems. My takeaway was: however bullish I was on compute before, I’m just a lot more bullish.”

Tang: “We’ve just been locked at our desks hammering Claude, because it’s fascinating the things we can now do with Fable 5 that we just couldn’t do with Opus 4.8 a day before. It’s really, really good at multi-agent orchestration now. I threw in seven of our models and said, ‘Create a master view of my beliefs given all these assumptions — TSMC capacity — and produce me a report.’ The model reasoned through all of our assumptions: ‘Actually, if you believe this, what are the contradictions?’ I’ve also dumped all my notes into it and it reasoned across three years of notes — here are your ideas that were consistent, here are the sources that were highest signal to what actually played out. And we’ve just blown through our limit.”

Gerstner: “They gave examples in the release — a 50 million line Ruby codebase at Stripe refactored in a day versus many weeks with many people. If you believe this to be true about long-running agents, we’re going to produce and consume more tokens as far as the eye can see.”

#7 — Frontier Captures 90% of Value; Open Source Carries 80% of Tokens

Baker: “Two things can be true. The majority of economic value may continue to accrue to the frontier — and man, has it ever accrued to the frontier thus far. And the majority of tokens consumed in the world may be open source — and they are today. I think this current state is likely to persist.”

“Harvey used their own proprietary legal data to do reinforcement learning and supervised fine-tuning with Fireworks on an open-source model, then used a router, and they got better outcomes than Opus 4.7 or 4.8 at a lower cost. I think that is the future. They were still consuming a lot of Opus, but a majority of the tokens were probably their own open-source model.”

Gerstner: “The consensus going into this year was that open-source models, cheap tokens, were catching up on the frontier, that these models were beginning to asymptote, that people wouldn’t pay for premium tokens. The evidence six months in is just the opposite — frontier tokens are capturing the vast majority of all the revenues. What folks concluded — that frontier models won’t accrue most of the revenue — has been decisively wrong. It’s 90%, probably more.”

“Think of JP Morgan — back of the house stuff, customer service, they may use an open-source model. But the really high-value stuff — coding — they don’t want to write second-tier code. You don’t need Albert Einstein to book you a trip. You don’t need Albert Einstein to do KYC.”

Baker: “Something very important on open source: there’s this belief that it’s bearish for AI. It’s actually — maybe bearish for the frontier models — but really bullish for compute and hardware. If the frontier models are capturing less of the margin, then you’re going to spend more on compute. The better open source does, the better it is for compute providers.”

#8 — Nvidia: “That’s a Cute ASIC” — The Open Source Counter-Move

Baker: “’Wow, that’s a cute ASIC you’ve built there. That is so cute. How would you like open source to join the frontier? How do you like them apples?’ I do think Nvidia is highly likely to be the world’s dominant provider of open source AI. Right now open source is whatever, six months behind the frontier — we might see it creep closer and closer.”

“If all of his customers are going to compete with him, then why not compete with his customers? He has his own models that are really, really good — Nemotron 3.1 was really cool from a compute-efficiency perspective, and he’s always careful to release small models so as to not tread on Anthropic, OpenAI, Google’s toes. But that is a choice he is making. If the economics change, I think Nvidia can join the frontier and become one of the world’s largest cloud computing companies much faster than people think.”

“You might not have the revenue or the margins to fund that ASIC [if open source closes the gap].”

Tang: “Everyone assumed Nvidia was going to lose share dramatically. Actually, if you look at the last few years, they’ve maintained their share very handsomely — and if you account for the fact that Anthropic was not really using Nvidia, they probably actually gained share in ‘25-’26. For ASICs, the argument now is more and more will look custom to the actual workload — MediaTek with their V8T versus Broadcom’s V8i for TPUs was a big topic. There’s a lot more nuance now versus one year ago when it was a Broadcom-or-Nvidia battle.”

Baker: “People are indexing to the OpenAI gigawatts — Nvidia has 10, Broadcom has 10, AMD has six with warrants, Cerebras has a gigawatt. I’ll be very surprised if that’s where they land. As long as we’re in a watt-constrained world, if you can get more tokens per watt — which is literally revenue — with Nvidia than alternatives, you may save some money building your factory with another chip, but you’re going to have less revenue. Meta and Microsoft have been probably disappointing [on ASICs]. You know who made a good ASIC? OpenAI — Jalapeño. They made a great chip. Unfortunately it needs to run at a much lower temperature than Nvidia GPUs, which means more money on cooling.”

#9 — The Capex Math: $1.5T Spend vs. $300B+ Inference Revenue — “The Math Maths”

Read more

Asia generally green overnight: TPX +1.35%, NKY +2.81%, Hang Seng +1.93%, HSCEI +1.91%, SHCOMP +1.12%, Shenzhen +0.99%, Taiwan TAIEX +2.36%, Korea KOSPI +4.63%. SK Hynix +1%; Samsung +6%; Kioxia +8% (largest market cap in Japan now, beats out Toyota).

In Tech, some hyperscalers strength early with AMZN, META and GOOGL +1-2%. Some strength in ALAB, NBIS, CRWV (all +5%) as they get added to Nasdaq 100. ADBE -8% on organic growth and strategy shift concerns which are just fueling the competition bear fears. We cover it all below, starting with ADBE.

Let’s get to it.

ADBE -8%: Beat and raised FQ2 revenue/EPS guidance, but narrative worsened as FY26 organic ARR was effectively cut by ~$480-500M as mgmt prioritized freemium MAU growth over near-term monetization; CFO leaves (on top of unresolved CEO search)

Not much else to say on this one than the above. Organic growth the main nit that will fuel AI disruption/competitive fears despite an otherwise decent set of numbers against low expectations going in. Mgmt is essentially rebalancing traffic away from direct-to-paid journeys toward freemium (Firefly, Express, Acrobat AI Assistant), explicitly trading short-term ARR for MAU growth and LTV. Mgmt said the ARR impact is roughly split: about half from deferring Creative Cloud line optimizations/pricing and half from going “full steam” on freemium. Plenty of bear fuel. More details below:

The #s:

• Revenue was $6.618B, +12.7% y/y (last q +12.0%) vs Street $6.454B, +~10%, and EPS was $5.96 vs Street $5.82.

• The guide was optically fine on revenue/EPS, with FQ3 revenue $6.67-6.72B vs Street $6.516B and FY26 revenue $26.50-26.60B vs Street $26.06B.

• Main negative ARR: total FY26 ARR growth stayed at 10.2%, but now includes ~$480M Semrush ARR, implying organic ARR growth was lowered by roughly ~190-200 bps.

• AI metrics good; AI-first ARR > $500M, ~3x y/y, Firefly ARR approaching $300M, and creative freemium MAUs >90M, +70% y/y, but investors remain focused on whether that converts to paid ARR fast enough

Key Takeaways:

• Mgmt is rebalancing traffic away from direct-to-paid journeys toward freemium (Firefly, Express, Acrobat AI Assistant), explicitly trading short-term ARR for MAU growth and lifetime value. Mgmt deferred previously planned 2H Creative Cloud "line optimizations" (pricing/packaging), removing a known growth lever for FY26. The combined hit is ~$480-500M (~2pp) to organic FY26 ARR, split roughly half pricing deferral, half freemium. On the callback, mgmt emphasized that the freemium shift is intended to access more demand, not simply defend against AI competition; creative traffic grew 50%, freemium MAUs accelerated to 70% y/y from 50% last q, and mgmt is more confident because usage is expanding beyond basic generation into editing, video and production workflows

• FY26 total ARR growth held at 10.2% but now includes ~$480M of Semrush, so organic ARR growth falls to ~8%, implying organic NNARR down ~20-25% y/y in 2H. ARR will also skew more to 4Q than the historical ~40/60 3Q/4Q split, since changes begin implementation in 3Q.

• FY26 non-GAAP OM ~45% maintained (Q3 ~44%, slightly below Street); mgmt was explicit it will not be short-term on spend (models, marketing, product). Per the callback, the freemium model is structurally more expensive and cost to acquire a dollar of ARR is rising.

• AI metrics were the bright spot: AI-first ARR >$500M, ~3x y/y (~2% of total ARR); Firefly ending ARR approaching $300M, ~+50% q/q via apps and credit packs; Acrobat AI Assistant paid MAU +150% y/y; CXO AI-first ARR 4x y/y; generated assets in Firefly Enterprise +4x y/y; credit consumption driven by video/audio.

• Engagement/funnel data is accelerating: adobe.com traffic +40% y/y (creative traffic +50%), Acrobat/Express MAU >850M (+20% y/y), creative freemium MAU 90M, +70% y/y (accelerating from +50% in 1Q). Mgmt argued LLM-driven, intent-based search is structurally changing user acquisition.

• Mgmt still sees Creative Cloud pricing/line optimization as available later, but the decision to defer planned 2H actions while expanding freemium will make bears argue that ADBE’s ability to push price in creative/prosumer/SMB is deteriorating.

• CFO Dan Durn is leaving (to become Marvell’s CFO); Steve Day (20-yr ADBE veteran) is interim CFO. No CEO successor named yet; goal is to have the next CEO in place to “put their stamp” on FY27 planning. CEO hire is prioritized ahead of the CFO hire.

• . The lower 2H organic ARR outlook was described as roughly 50/50 between less pricing contribution and a harder freemium push, with benefits expected to show up in FY27 and more detail expected later on MAU/free-to-paid conversion. Pricing remains part of the growth algorithm, but mgmt does not want pricing to be the message while trying to capture heightened AI/LLM-driven traffic; guidance methodology was described as unchanged despite CEO/CFO transition, with more ARR expected to land in FQ4 than usual.

Bull vs. Bear Debate

Bulls start from the premise that ADBE remains one of the highest-quality franchises in software: ~97% subscription mix, ~90% gross margins, ~45% operating margins, $10B+ of annual FCF, and entrenched positions across creative tools, PDF, and the marketing/content supply chain. They argue this quarter actually delivered the AI proof points skeptics demanded: AI-first ARR tripled y/y to >$500M, Firefly ARR is approaching $300M with apps/credit packs up ~50% q/q on top of +75% last quarter, Acrobat AI Assistant paid MAU grew 150%+, and credit consumption is being driven by compute-heavy video and audio. The freemium pivot, in bulls’ view, is offense rather than defense: AI and LLMs are generating an explosion of high-intent search traffic (adobe.com traffic +40%, creative traffic +50%, freemium MAU +70% y/y and accelerating), and ADBE is replaying the Acrobat Reader playbook that built one of software’s most durable monetization funnels, this time with proven infrastructure for paywall placement and conversion. Bulls also point to the enterprise as a structurally underappreciated asset: AEP + apps growing 30%+, GenStudio 25%+, CXO AI-first ARR up 4x, 1,500+ agentic trials, and agency holding companies standardizing on ADBE. Add in commercially safe models (a differentiation sharpened by the Disney/Universal suit against Midjourney), distribution of Creative Agent inside ChatGPT, Claude, Copilot and Gemini, and Semrush extending ADBE into brand visibility/GEO at exactly the moment CMOs care about LLM placement, and bulls see a company repositioning from a position of strength.

Bears read this quarter as confirmation of the structural thesis: when AI natives (Canva, Midjourney, Runway, Synthesia, Gamma, HeyGen) and frontier labs are scaling rapidly in consumer/SMB creative, a company that simultaneously defers price increases and gives more product away for free is, by definition, losing pricing power. The freemium pivot may be framed as offense, but bears note the inconvenient math: total NNARR still declined y/y despite AI-first ARR tripling, meaning AI gains are not yet offsetting core erosion, and the ~10pp increase in AI's share of TTM NNARR is happening inside a shrinking pool. Organic FY26 ARR growth now sits around 8% with 2H organic NNARR implied down ~20-25% depending on the quarter, the 2H is more back-end loaded into 4Q than usual (raising execution risk), and freemium is structurally more expensive: mgmt itself acknowledged the cost of acquiring a new dollar of ARR is going up, while leading AI labs are moving toward paid monetization just as ADBE moves the other way. Bears also question why this freemium-to-paid motion reignites growth in an era of rising commoditization at the low end, and worry about cannibalization of paid Creative Cloud seats and eventual margin drift below 45% as inference costs and S&M intensity rise.

Layered on top is an execution/credibility problem: the CEO and CFO are both departing within a quarter of each other with no permanent successors named, guidance philosophy is unchanged but the people who built it are leaving, and any new CEO is likely to rebase FY27 expectations, meaning Street numbers (~8-9% revenue growth, ~$26.5-27 EPS) may still be too high. With limited near-term catalysts until a new team is in place and freemium conversion data arrives (mgmt says FY27), bears will say "show-me" story where the multiple is a value trap, not a floor.

TECH RESEARCH/NEWS

AMD: Citi Upgrades to Buy, Sees Agentic AI Driving a Larger CPU TAM and Meaningful GPU Upside

Citi upgraded AMD to Buy and raised its PT to $575, arguing investors are underestimating both the CPU opportunity created by agentic AI and AMD’s position in custom AI accelerators. The firm now forecasts the server CPU TAM reaching ~$137B by 2030, up from ~$132B previously, as agentic workloads increasingly shift orchestration, reasoning, memory management, and tool-use functions toward CPUs. Citi also believes Meta could become a significantly larger AMD customer than currently expected, with custom MI450 deployments potentially translating into ~$15B of revenue. Against this backdrop, Citi raised CY26/27 EPS estimates 12%/13%.

ALAB, CRWV, NBIS, RKLB and TER will be added to Nasdaq 100

SPCX: Wolfe Initiates Outperform, Says the Real Bet Is Starship, Not the 2035 Targets

Wolfe initiated SPCX at Outperform with a $175 PT, arguing investors do not need to believe the company’s long-term space datacenter ambitions to own the stock—they simply need to believe in Starship. The firm views SpaceX’s near-zero launch-cost advantage as one of the strongest competitive moats in any industry, enabling growth across Starlink, launch services, communications infrastructure, and eventually AI-related opportunities. Wolfe expects Starlink EBITDA to exceed $90B and EBITDA-less-capex to surpass $70B by 2030, driven by continued subscriber growth and significant capacity expansion. While the longer-term AI and orbital compute opportunity remains highly speculative, the firm believes Starship’s successful commercialization alone can support substantial upside, making future AI infrastructure optionality effectively a free call option for investors.

U/APP: Morgan Stanley Says Direct Payments Could Become a Meaningful Earnings Tailwind for Mobile Ad Platforms

Morgan Stanley argues the shift toward direct payments in mobile apps and games is increasingly bullish for both U and APP as developers look to reduce app store fees and reinvest a portion of those savings into user acquisition. The firm estimates every 5% reduction in payment processing fees could drive ~19% upside to U’s 2027 EBITDA and ~5% upside to APP’s, assuming roughly half of the savings are redirected into marketing spend. Beyond the advertising benefit, Morgan Stanley also highlighted U’s new Commerce product, which could allow Unity to participate directly in payment processing economics, with every 10% penetration of global in-game transactions potentially driving ~20% upside to 2027 EBITDA.

NOK: JPM Raises PT to €18/$21, Sees AI Optics and Hyperscaler Switching Wins Driving Upside

JPM says Nokia generated €1B of AI and cloud orders in 1Q, with JPM expecting new optical products including 1.6T coherent pluggables, 3.2T solutions, and advanced transponders to drive further growth. The firm also highlighted datacenter switching wins with Microsoft and at least one additional hyperscaler, with Google viewed as a likely customer. JPM believes Street is not fully reflecting the revenue and operating leverage from Nokia’s optical and datacenter networking businesses, leaving its 2026-28 EBIT estimates meaningfully above consensus.

Read more

Read more

In Tech, semis are leading the way higher with INTC (+5%) up on a dbl ug at BofA, Memory following Asia strength overnight (SNDK/MU +3-5%), and semicaps looking to be the first AI semi group to be regaining 52wk highs (ASML/AMAT +5%). ORCL -5% as print did nothing to change the bull/bear debate although bears had more fuel this morning with the IaaS miss and lack of FY27 raise. Apps missed as well which has software underperforming early.

In Asia, mixed px action: TPX -0.45%, NKY +0.06%, Hang Seng -0.65%, HSCEI -1.22%, SHCOMP -0.16%, Shenzhen -0.66%, Taiwan TAIEX -0.18%, Korea KOSPI +0.43% but Samsung +3%, Kixoia +4% SK Hynix +7%.

SpaceX (SPCX) starts trading tomorrow (we have some Starlink 3p data below).

Lots to get to so let’s get to it…

ORCL -8%: Solid AI-infra/RPO quarter with better revenue, EPS and OM, but IaaS missed buyside bogeys, and debate continues between accelerating OCI visibility and concerns on unchanged FY27 revenue guide, SaaS decel, & FY27 GM/capex and financing.

The #s:

• Revenue was $19.184B, +20.6% y/y / +20% CC (last q +21.7% / +18% CC) vs Street ~$19.10B, +~19.5-20%; EPS was $2.11 vs Street $1.96; non-GAAP OM was 44.8% vs Street ~43.3-43.5%.

• Cloud was $9.913B, +47% y/y / +46% CC, slightly below Street, with OCI/IaaS $5.787B, +93% y/y / +92% CC vs Street roughly low-90s growth and buyside in mid 90s, while SaaS/Cloud Apps slowed to $4.126B, +10% y/y / +9% CC vs Street closer to low-teens.

• RPO was the stock-moving standout at $638B, +363% y/y, up $85B q/q vs Street around $577B, with $67B of AI infra contracts signed and BYOH/prepaid commitments now $75B.

• F1Q27 guide was above Street: revenue +27-29% y/y, cloud +58-64% USD / +57-63% CC, EPS $1.72-$1.76 vs Street ~$1.68-$1.69;

• FY27 revenue guide stayed $90B, +34%, above Street but unchanged, and EPS guide moved to $8.05. FY27 capex of $90-95B was well above Street (~$62-85B range), though net cash capex of ~$70B (after $20-25B of customer prepays) was in line to slightly better than feared.

Key Takeaways:

• RPO was $638B, +363% y/y, up $85B q/q, and cRPO accelerated to +68% y/y from +65%. The RPO add was driven largely by AI infra contracts, including four customers signing $8B+ deals, while reports also noted a base of smaller $100-$300M deals, which helps push back on OpenAI-concentration concerns. 12% of RPO converts to revenue in the NTM (implying cRPO of $76.6B, +68% y/y), 34% in months 13-36, with both percentages expected to rise as capacity ramps.

• The majority of Q4 bookings were prepaid or bring-your-own-hardware; these now total $75B of RPO with "no degradation in margin," some at better gross margins and meaningfully higher ROIC (mgmt cites high-20s ROIC at steady state). Customer prepays cut FY27 net cash capex to ~$70B vs $90-95B reported.

• ORCL delivered >1.2GW in FY26 and expects F1Q27 delivery approaching 1GW, nearly the same as the prior four quarters combined. Abilene is 42% delivered with another 35% expected in the next 90 days, and several large sites are scheduled for 1H/2H CY27 delivery.

• ORCL plans ~$40B of FY27 debt/equity raises (including the $20B ATM), no incremental debt in CY26; credit metrics improved q/q (gross/net leverage 3.5x/2.7x) and bonds rallied ~10bp even as the stock fell.

• Margins were mixed: OM beat, GM pressure remains the FY27 overhang. F4Q non-GAAP OM beat on opex discipline, but gross margin compressed as OCI/AI infra mix rose. Mgmt guided FY27 gross margins lower due to data-center ramp timing and mix, while saying infra margin should improve rapidly as sites hit full contractual revenue levels.

• Apps grew +9% CC, down from +11% CC last q and below Street, which keeps the “SaaSpocalypse”/AI disruption debate alive. Mgmt pushed back by pointing to +16% SaaS deferred revenue, AI agents embedded in apps, Oracle Health/VA deployment momentum, and a new OPM Fusion HCM award not in Q4 bookings.

• Mgmt sees AI infra demand "massively higher" than supply with a trillions-per-year market; agentic coding demand shows "no slowdown." On SaaS, the "SaaSpocalypse" delayed decision cycle of two quarters ago has passed, with SaaS deferred revenue +16% y/y (faster than revenue) signaling re-acceleration; new token-bundle and outcome-based pricing launched to monetize agents.

Bull vs. Bear Debate

Bulls see ORCL as one of the clearest large-cap software beneficiaries of AI infra demand, with a unique combination of contracted backlog, database ownership, applications, multi-cloud distribution and a capacity ramp that is now becoming visible. The $638B RPO, up 363% y/y and up $85B sequentially, gives near-contractual visibility into a 30%+ revenue CAGR through FY30, and this quarter's $67B of bookings demonstrated both scale and improving diversification (four customers at $8B+ each, plus a long tail of $100M-$300M deals, with concentration declining vs a few quarters ago). Execution is matching the demand: 1.2GW delivered in FY26, ~1GW coming in F1Q27 alone, all five major sites on or ahead of schedule, capacity sold out for six months and 97.5% GPU utilization. Critically, the BYOH/prepaid model ($75B of RPO) answers the two biggest historical pushbacks, financing risk and returns, since these deals carry equal-or-better gross margins, high-20s steady-state ROIC (higher on prepaid structures), and cut FY27 net cash capex to ~$70B with only ~$40B of funding needed and no new debt in CY26. That gives bulls confidence that OCI growth can sustain or even exceed 100% as sites turn on, especially with Abilene and other named data centers moving toward delivery.

Bulls also argue the capex debate was partially de-risked. The reported FY27 capex number is large at $90-$95B, but net cash capex of ~$70B is lower than feared by some investors, and BYOH/prepaid structures move part of the capital burden to customers while preserving similar or better margins. The bull narrative is that ORCL is not just renting GPUs at poor returns, but is building a high-ROIC infra business with embedded database and application pull-through. Multi-cloud database revenue growth of 404% and bookings growth of 325% reinforce the idea that ORCL’s installed database base becomes more valuable in an AI world where proprietary enterprise data matters.

Bears argue the backlog is impressive but increasingly disconnected from the P&L: despite adding $85B of RPO in a single quarter, ORCL did not raise the $90B FY27 revenue guide, the Q4 revenue beat was just $81M (under 1%) vs $300M+ a year ago, and only 12% of RPO converts over the next twelve months. The structural concern is that ORCL is transforming from a 70%+ gross margin software franchise into a capital-intensive GPU landlord: gross margins have now compressed y/y for 20 consecutive quarters, FY27 GMs step down further, AI infrastructure tops out at 30-40% gross margins, and FY27 EPS grows only ~5-6%. Meanwhile, the cash demands are enormous: $90-95B of reported FY27 capex (2H26 capex roughly equaled net income), FCF negative through at least FY28-29, $40B of new financing including $20B of dilutive ATM equity, ~3.5x gross leverage, plus counterparty risk to AI labs, GPU obsolescence/depreciation risk, and rising competition from hyperscalers and neoclouds for the same capacity. Layer on the apps business decelerating to 9% cc (NetSuite SMB softness, tougher compares, lingering AI-disruption fears about seat-based SaaS), and bears see a company taking on hyperscaler-level risk without hyperscaler-level cash generation.

TECH RESEARCH/NEWS

INTC: BofA Double-Upgrades to Buy, Sees Much Larger CPU and Foundry Opportunity Emerging

BofA double-upgraded INTC to Buy and raised its PT to $135 from $96, arguing investors are materially underestimating both Intel’s long-term CPU opportunity and the value of its foundry business. The firm now models server CPU revenue exceeding $40B by 2030, driven by the expansion of the CPU TAM in an agentic AI world, while also seeing a $45B+ external foundry opportunity spanning Apple M-series production, MediaTek AI accelerators, custom CPUs, edge AI, and advanced packaging. BofA believes recent datapoints—including improving 18A execution, the CDNS 14A partnership, and Terafab-related engagements—support higher confidence in Intel’s IDM strategy. BofA thinks Intel’s earnings power and foundry optionality are increasingly visible, with BofA now modeling CY30 EPS potential above $6 and arguing ownership remains unusually low relative to the scale of the opportunity.

CPUs/AMD/ARM: BofA Raises Server CPU TAM to $170B, Says Agentic AI Fundamentally Expands the Opportunity

BofA argues agentic AI is creating a structural renaissance for CPUs, raising its CY30 server CPU TAM estimate to $170B from $125B as AI workloads increasingly require planning, orchestration, reasoning, memory management, tool use, and other sequential tasks that are better suited for CPUs than accelerators. The firm believes agentic AI shifts part of the value creation from GPUs/XPUs toward CPUs, creating a new standalone agentic CPU opportunity while also increasing demand for head and compute-node CPUs within AI clusters. Against this backdrop, BofA raised targets across CPU-exposed names, double-upgraded INTC to Buy, remains constructive on AMD and ARM, and continues to view NVDA as the top overall AI beneficiary.

OpenAI: OpenAI Considers Drastic Price Cuts, Anticipating War for Users With Anthropic

WSJ:

The company is weighing significant cuts to what it charges for tokens, the unit of measurement artificial-intelligence firms use to bill for their products, according to people familiar with the matter. The move would be in anticipation of similar cuts the company expects at Anthropic, the people said.

Business executives have begun to balk at the high prices for AI usage. OpenAI Chief Executive Sam Altman said at a recent event that costs had become “a huge issue.”

“I think we’ll have a lot of ways we can help people get more value for less spend,” he said.

SpaceX (SPCX) starts trading tomorrow:

3P check: According to Apptopia, Starlink growth is accelerating in Q2 and continually hitting monthly download highs; June is on pace to be another new high:

Read more

The main debate today in AI semis was around Token Optimization as the chart below as being passed around:

What exactly is the “Token Expenditure Index” and why does it matter? It’s basically a usage weighted average price per million tokens, which is supposed to capture the market’s marginal willingness to pay for frontier models. Here’s Claude explaining things:

It’s unclear to me - and others - how accurate this index is. But technicals and semis were weak today so lots of eyes on it today. I spent some time trying to figure out what moves this index and had Claude annotate it for me. This is a bit detailed, but the logic seems to hold up at first glance:

So conclusion is we should likely see a turn in the next couple weeks assuming demand holds up amidst the Fable/Mythos price increases. Still, I’m not convinced this index by itself can tell us anything useful without any more detail. It measures spend divided by tokens and Silicon Data publishes neither so tough to tell which side of the equation is driving things. This illustrative example from Claude shows us how the index can change with 3 drastically different interpretations:

The counterargument? This chart also going around today:

Ramp's spend-per-employee data through May 2026 is up and to the right across the entire distribution. If total wallet is rising while blended price per token falls (Ramp's data shows average cost per million tokens across major providers dropped from roughly $10 to $2.50 in a year), then token volume is arithmetically exploding.

The Token optimization debate has definitely heated up after many realizing tokenmaxxing without a focus on ROI wasn’t the smartest thing to do. UBS’ Karl Keirstead had a good note after doing a round of checks earlier this week on it, explaining how enterprises are thinking about:

Read more

Asia generally weaker overnight although AI names were down: TPX -1.25%, NKY -1.89%, Hang Seng -0.64%, HSCEI -0.07%, SHCOMP -0.42%, Shenzhen -1.97%, Taiwan TAIEX -3.31%, Korea KOSPI -4.52%. Kioxia -8%; Softbank/Samsung/SK Hynix -8-10%

Lots to get to so let’s get straight to it…

OpenAI: OpenAI in Talks to Lease 10 Gigawatt Ohio Data Center with Backing From Nvidia

TheInformation:

OpenAI is in advanced negotiations to lease a proposed 10 gigawatt data center campus on federal land in Ohio as part of a deal that could include financial backing from Nvidia, according to two people with direct knowledge of the discussions.

The campus under discussion would be among the biggest of its kind, with a total cost of at least $500 billion if fully built out based on today’s prices for chips, labor, power and other materials, the people said. OpenAI would control the equipment in the facility under a long-term lease and would be on the hook for payments once the project starts operations, with the first phase expected in 2028. The discussions haven’t been finalized and plans could change.

OpenAI: SoftBank Attempt to Get $6 Billion OpenAI Margin Loan Stalls

Bloomberg:

SoftBank Group Corp.’s talks with potential creditors to raise at least $6 billion from a margin loan backed by its OpenAI stake have stalled, people familiar with the matter said, just weeks after the Japanese conglomerate cut its initial target from $10 billion.

The company is considering various fundraising options, according to the people, who asked not to be identified discussing private matters. It could still move forward with the margin loan at a later stage, they added.

It’s unclear why the margin loan discussions stalled.

TSMC May 2026 revenue: NT$417.0B (+1.5% MoM, +30.1% YoY)

5 year median is 0.9%, so this is roughly in line with that.

QTD through two months: April NT$410.73B + May NT$416.98B = NT$827.7B, i.e. 66.9% of the low end and 65.9% of the midpoint. For reference, Apr+May has made up 67–72% of Q2 revenue in each of the last four years (67.1% in ‘22, 67.5% in ‘23, 69.1% in ‘24, 71.8% in ‘25).

For reference, June has been down MoM four straight years (−5.3%, −11.4%, −9.5%, −17.7% from ‘22 to ‘25; ~−11% average), which would imply they would need to outperform June to hit their guide.

However, given stronger TWD and assuming flattish June (March/April/May have plateaued at NT$410–417B, which looks like supply-constrained) gets you right at the midpt.

Optics: Morgan Stanley Says CPO Sentiment Has Reset, but the Long-Term Opportunity Remains Intact

Morgan Stanley addressed growing investor concerns around CPO and argued the recent pullback reflects a reset in expectations rather than a change in the long-term thesis. The firm now models 2027 optical engine shipments of roughly 6-7M units, well below some market expectations, citing downstream assembly yields of 20-50% and broader manufacturing bottlenecks as key constraints. While this pushes out the near-term CPO ramp, Morgan Stanley believes the industry remains at an inflection point, with CPO adoption beginning around 2028 and becoming increasingly important as scale-up and scale-out AI networks expand.

Power Infrastructure: Morgan Stanley Pushes Back on Rumors of an 800V AI Datacenter Delay

Morgan Stanley believes reports from Semianalysis that Nvidia may delay large-scale 800V datacenter deployments are being overstated and do not change its broader power infrastructure thesis. The firm notes 800V power racks are already being prepared for mass production and that development activity across hyperscalers remains on track, even if some standalone deployments slip modestly. Morgan Stanley argues the transition to 800V is being driven by fundamental increases in rack power requirements and remains necessary regardless of timing. The key takeaway is that any delay would likely reflect qualification, safety, and deployment timing considerations rather than a change in the industry’s long-term direction, with the broader 800V opportunity across power infrastructure and components remaining intact.

STM: BofA Upgrades to Buy, Sees Optical Interconnects, LEO Satellites, and Auto Recovery Driving a Multi-Year Earnings Inflection

Read more

Ok, we’ll cover more of what went on in AI semis below. Let’s get right to it…

AI/SEMIS

Couple key news items today, with the big one being the release of Anthropic Mythos/Fable 5 public release

Second one was Crusoe pausing the 1.8GW Wyoming Site. The tenant hasn’t been confirmed but likely OpenAI. Chatter is that they fired Crusoe for cost overruns and going self-build, but haven’t gotten confirmation on the speculation. CEO was on CNBC, although didn’t give much other than "site related issue"

Now onto the good stuff…

Read more

Speakers: David Goeckeler (CEO), Luis Felipe Vizsozo (CFO) | Host: VJ (Mizuho)

1. New Business Models / LTAs — Structure, Pricing Mechanics & Margins

CFO: “In general, there are fixed price components within these new business models, which I believe are important. There are parts of the agreement... with, let’s call it a floor and a ceiling of pricing... The reason we did that is because none of us wanted to be unhappy, right? If prices go up, we would be unhappy because we would not be capturing the upside. If prices go down, our customers would be uncompetitive... their peers would be paying lower prices... Importantly, as we said in our earnings call, even in the low prices, we like the margins. This is how we structure them. Margins will be consistent with the margins that we guided in for the fourth quarter, for our fiscal fourth quarter.”

CEO: “We’re not trading duration for price, right?... That’s not the value proposition. The value proposition is continuity of supply... Price is price. Price is whatever’s fair for both of us... You can assume we have somewhat unique insights into price being in the market... As our customers go through that equation... they’re coming to the conclusion that having longer term relationships with suppliers of NAND is a very good idea... It’s changing the trajectory of our business, I think in a very positive way... It’s a very powerful relationship between us and our customers, and it’s very much of a win-win conversation.”

2. LTA / NVM Pipeline — Five Deals Signed, More Coming

CFO: “When we closed the quarter, we said we had signed five deals... we’re very happy about that... It’s a win-win relationship with our customers... They’re coming back, and we’re in constant negotiations with them. As I also mentioned last quarter, there are several conversations going on, and they’re progressing well... We’re talking to customers across all segments, right? From data centers to edge customers. Obviously, our consumer business is more transactional... it doesn’t really apply there. But we’re talking to all other customers, and as long as we’re willing to operate in this new business model, we’re open for business.”

3. AI / Data Center Demand — Inference Is the NAND Unlock

CEO: “We see a market with sustained mid to high teens growth... Clearly, there’s a lot of AI demand that’s helping that equation. Data center has now become or is quickly becoming the largest market in NAND. We’ve had many revisions of data center CapEx going up... it’s 14 revisions now... all going up. A lot of that is as AI moves into inference... [Our customers] are working on what is the architecture for inference? That’s when NAND becomes a big part of that equation... We’ve seen it with... KV cache or RAG. NAND has always been the most scalable semiconductor technology. As you start to scale any architecture, I think it’s naturally going to come towards NAND... if you need storage... Figuring out what’s the right concentration of processing power, what’s the right concentration of DRAM, HBM, what’s the right concentration of NAND. It’s very use case dependent... I think the big-picture answer to that question is you need a lot more NAND. Because NAND is the most scalable semiconductor technology. And if you’re going to scale something economically, you’re going to want to use as much of the most scalable technology that you possibly can... NAND has been something where it’s been almost always focused on density... Now it’s focused more on high performance... It’s a much more multidimensional equation... There’s been a lot said about AI, but [it’s an] extremely fundamental technology shift that we’re clearly in the very early innings [of]... A lot has changed in the last year, but I really do think we’re just getting started.”

4. Enterprise SSD Ramp — 7x YoY, ~25% of Revenue, Second Leg Just Starting

Host framing: enterprise SSD is “almost 25% of the revenues now... might be mid-single digit last year. It’s grown 7x year-over-year.”

CEO: “It’s been no secret, we’ve been investing heavily in enterprise SSD. It’s a part of the market where we’ve been under-penetrated, and it’s about building the right portfolio to increase our mix in that part of the market... There’s really two major categories of products there. There’s a performance-based TLC NAND product... used a lot for KV cache and a number of things. That’s been the driver of the portfolio over the last year. The second part of the portfolio is the storage-based [capacity] product... This is the first quarter we’ll recognize revenue on that product. You’re right, we’re just getting started on that leg of... the portfolio... Now we’ve got the second leg of that growth coming. Where’s that going to lead to over the next year? I think you’re going to see the mix of data center go higher... I’m not going to put a specific number on it... We believe that a robust portfolio where we can cover as much of the market as possible is the best for long-term profitability of our franchise... We have a global consumer business that already gives us a great starting point. We have a great client business, and now we’re building out that great enterprise SSD business.”

5. HBF (High-Bandwidth Flash) — Roadmap, System-Level Design, Latency Pushback

Read more

Some good stuff in here…thematic quotes first, then more stock specific read-throughs at the end…

#1 — The Enterprise AI “L Curve”: Less Than 1% Penetrated

Sacerdote: “The enterprise AI or enterprise application AI market is less than 1% penetrated, and we’ve never seen — we talk about S-curves — we call this an L curve, just straight up.”

“Even though 800 million people are using AI, they’re just using AI 1.0, which is like a search engine on steroids. But now with these new primitives — where you have Claude on your computer, linking it in, then you build skills, companies are going to build skills, then true AI bots, then big corporations are going to build much larger. Sundar said it’s 10 bips of the knowledge workers of the world. It’s classic S-curve — these are the tinkerers, then early adopters, then early mainstream. You’re going to go from 10 bips to 1% to 2-3% to 5% to 15% in the next four years.”

“A light switch this year went off in the enterprise where everybody realizes they need to do this now and do it fast. It’s like internet 1.0 — you knew you needed a website in 1998, but it’s hard to build that website.”

“We’re at 10 basis points of people really using AI and we’re already sold out — there’s not enough compute in the world. Anthropic has half of what they need right now, and that’s before this huge take-up. Marc Andreessen said in the next four years, one thing he’s sure of is there’s not going to be enough compute.”

#2 — Code as the True Unlock of AI: The Karpathy Flip

Read more

We get NET investor day at 1pm, DDOG DASH conference at 10ET, ZS ZenithLive at 4pm ET, and NBIS Inflection at 5ET.

Lots to get to this morning, so let’s get right to it:

Anthropic: Mythos believed to go public today - Sources

OpenAI: Files for IPO

China AI: China Preps $295 Billion Plan to Fund Nationwide AI Buildout

Bloomberg:

China is preparing to spend around 2 trillion yuan ($295 billion) over the next five years on building data centers across the country, fueling Beijing’s ambition to propel the domestic AI sector and surpass the US in a potentially game-changing technology.

Key government agencies including the National Development and Reform Commission are drafting a blueprint to erect a network of inter-connected computing hubs across the country, people familiar with the matter said. State firms such as China Mobile Ltd. and China Telecom Corp. will operate the bulk of the data centers and ensure they’re connected, one of the people said. The idea is to rely on local suppliers including Huawei Technologies Co. for at least 80% of technology such as AI chips, effectively squeezing out Nvidia Corp. and Advanced Micro Devices Inc., the people said.

China AI: Taiwan Mulls Curbs on AI Chip Exports to China to Align With US

Bloomberg:

Taiwan authorities are considering much stricter export controls on AI chip sales to China to further align with US measures, according to people familiar with the matter, an effort to address semiconductor smuggling that risks drawing a rebuke from Beijing.

The idea is to give authorities more legal tools to address diversion of advanced hardware, like AI servers with Nvidia Corp. chips, from Taiwan to China. Such sales are already banned under US regulations unless companies get Washington’s permission, per curbs the US first imposed in 2022 to prevent Beijing from using advanced Nvidia processors to gain a military edge.

INTC: Citi and JPM with some commentary on TheInformation article yesterday

Read more

Let’s get straight to the good stuff…

AI/SEMIS

• MRVL +10% after announcement to join SP500. FLEX -1% lagged as only moving from SP400 to SP500. Digitimes had an interview withMRVL’s COO Chris Koopmans here where he said that MRVL has secured enough production capacity to meet its 50% growth target this year, 55% in 2027. He said MRVL’s SerDes now at 200G on the 2nm process and 500G on A14. Some other key quotes re: optical:

“Most people believe the limit is around 200G or 400G. Today, it is already very difficult to use copper transmission to interconnect 72 compute chips.”

“I believe NPO and CPO will be adopted gradually over the next five years... the scale-up field will basically move gradually toward optical transmission... it will move toward a CPO format.”

• Semicap outperformed today. ASML +6% as Elon called it the greatest company in Europe and co. invite him to speak about Terafab. Other semicaps strong with AMAT +9%/LRCX +8%. Nothing to dislike in this group here as bulls continue to see a big disconnect between valuation and where earnings growth is going. The narrative is simple: if you believe the memory/AI supply story, the world is structurally under-supplied on the exact capacity that’s hardest to add (especially with HBM eating far more wafer capacity per bit than commodity DRAM), and equipment is the levered, lagging way to play it. If WFE goes to $250B in 2028, then implies AMAT multiple near 15x. Bulls think bottleneck noise is overplayed as higher-throughput litho platforms mean ASML won’t cap the ramp, so total WFE can keep climbing to 250B and beyond. Higher-throughput litho also means value acrruse to etch/depo, which is why some bulls prefer AMAT/LRCX vs. ASML.

• INTC +11% after TheInformation confirmed GOOGL as a customer saying that GOOGL “recently placed an order with Intel to manufacture more than three million Tensor Processing Units in 2028, after months of testing Intel’s advanced packaging technology.” Some pointing that the word “manufacturing” is interesting, although this is very much likely for EMIB. GOOGL as a customer has been speculated by Asia press and others for a couple months now but the number of TPUs on order was new. TheInformation also confirming Nvidia running early trials on 18A.

  • GF Securities’ Jeff Pu was also out early morning with positive note on INTC saying they now expect Intel 3 capacity to increase ~80% and 18A capacity to double between 2026 and 2028, supported by external foundry customer commitments and improving manufacturing execution and highlighted Clearwater Forest as a meaningful server CPU catalyst, with production expected to begin in 3Q26.

• Memory bouncing back nicely with MU +9% and SNDK +5% as street was out defending what was obvious late last week given the SoCAMM reductions at NVDA. Citi anod others said they expect no meaningful reduction in overall SoCAMM demand as memory suppliers remain capacity constrained and Nvidia prioritizes shipment volume, saying move actually helps NAND demand which we agree. We went into the rationale for better NAND demand in our weekly:

Read more

Crude +1.4%. Yields flattish. Asia mainly down off the U.S. weakness Friday: TPX -2.45%, NKY -3.85%, Hang Seng -1.22%, HSCEI -1.13%, SHCOMP -1.7%, Shenzhen -3.14%, Taiwan TAIEX -3.48%, Korea KOSPI -8.29%

AAPL WWDC scheduled to kick off at 1pm ET where investors want to see the progress they’ve made on AI platform and agentic vision.

Lots to get to this morning, so let’s get right to it.

MRVL / FLEX: Set to join SP500

ROKU: Set to join S&P Midcap400

OpenAI: OpenAI plots biggest ChatGPT overhaul since launch

FinancialTimes:

The company intends to transform the chatbot into a “superapp” that combines coding tools and AI agents, adding products that executives believe will generate more revenue. The changes are part of a broader reorganisation at OpenAI as the San Francisco-based company shifts resources into trying to win lucrative business customers and compete more fiercely with rival Anthropic, according to more than a dozen current and former employees.
The overhaul, which is set to begin rolling out in coming weeks, will initially appear as changes to ChatGPT’s website and mobile apps, encouraging customers towards using coding, image-generation and apps from external partners.

“Chat is dead,” said one senior OpenAI employee.

OpenAI: Trump administration, OpenAI discussing possible government stake in the AI startup (Out Friday)

Semafor/CNBC:

OpenAI CEO Sam Altman and the White House are in ongoing talks about a possible government stake in the artificial intelligence company, CNBC confirmed on Friday.

The discussions have been in progress for more than a year, as Altman first shared the idea with the Trump administration in 2025, according to a source familiar with the matter who asked not to be named because the details are confidential.

The talks continued this week as Altman met with a range of lawmakers and officials in Washington about regulation and the latest developments in AI.

As part of the potential agreement, OpenAI could donate equity to the U.S. government to seed something like the “Public Wealth Fund” that the company outlined in its April policy proposal, the person said.

NOTUS first reported Altman’s pitch, which the president told reporters Friday he might discuss with “all the big” AI firms as soon as this week: “The American people can benefit from the success of AI, and by doing that, they’re going to like it better,” Trump said.

INTC: GF Securities Raises PT to $135, Sees Significant Foundry Capacity Upside and Improving Yield Trends

GF Securities raised its PT on INTC to $135 from $94, citing improving 18A yields, stronger server CPU demand, and growing confidence in Intel Foundry’s capacity expansion plans. The firm now expects Intel 3 capacity to increase ~80% and 18A capacity to double between 2026 and 2028, supported by external foundry customer commitments and improving manufacturing execution. GF also highlighted Clearwater Forest as a meaningful server CPU catalyst, with production expected to begin in 3Q26 and drive stronger DCAI growth through 2027. The analyst views potential collaboration with NVDA on x86-based PCs, server CPUs, and foundry services as incremental upside rather than a core part of the thesis.

VRSN: Data continues to accelerate

AI Infrastructure: Bernstein Estimates Vera Rubin NVL72 Costs ~$9.1M per Rack and ~$47B per GW

Bernstein updated its Vera Rubin NVL72 cost model and now estimates a fully configured rack costs roughly $9.1M, above widely cited ~$8M figures, driven primarily by higher memory and storage costs. The firm estimates GPUs remain the largest cost component at roughly $4M per rack, while networking contributes another ~$1.2M and memory/storage approximately $3.2M. On a datacenter basis, Bernstein estimates all-in AI infrastructure capex of roughly $47B per GW, including both IT and physical infrastructure. Despite higher costs, the firm argues Rubin continues to deliver improving performance-per-dollar, with major increases in compute density, memory content, networking, power, and cooling requirements.

MEMORY

Memory: Citi Says Rubin SoCAMM Reduction Reflects Supply Constraints, Not Demand Weakness and positive for NAND demand

Citi believes Nvidia’s decision to reduce Vera Rubin SoCAMM2 capacity by moving from 192GB to 96GB modules is being driven by industry-wide DRAM shortages rather than weaker demand. While the change cuts DRAM content per rack, the firm expects no meaningful reduction in overall SoCAMM demand as memory suppliers remain capacity constrained and Nvidia prioritizes shipment volume. Citi also argues the move reinforces the need for larger KV-cache memory pools, increasing the importance of Context Memory Extension (CMX) architectures. As a result, the firm sees the adjustment as neutral for DRAM demand and potentially positive for NAND demand over time as CMX adoption expands.

Memory: Jefferies Says Rubin DRAM Reduction Reflects Supply Constraints, Not Weakening Demand

Read more

We were lucky to be traveling Friday and not in front of our screens, which allowed us to avoid the hyperbole that typically accompanies a decline of such magnitude.

Why the sell off and what comes next?

We’ll tackle the former question first. We think a mix of multiple things. First the macro: fed expects shifting in a more hawkish direction over the past couple weeks exacerbated by the hot NFP print on Friday (market now expecting 27bps worth of hikes this year); yields across the curve moving up; continued concerns around Iran war and oil inventories running out shortly and what that means for oil and headline inflation; USD ripping which creates EPS headwinds for multinationals.

Second, positioning, sentiment & leverage: we were extremely extended with the SOX up 95% since the end of March and RSI at 80 by Wednesday. Everyone has been piling into semis. No need to mention the froth …

Read more

#1 — AI Compute Revenue Already Materializing: $15B ARR Anthropic Deal

Johnson: “We decided to start demonstrating what that model will look like even before orbital on the terrestrial side. You saw the announcement a month ago with Anthropic — a deal that from an annual run rate perspective is $15 billion of additional revenue. We’re talking to a number of other customers right now for the same type of service. What we had talked about doing as far as hosting others in space — why are we waiting? Why don’t we move our model to the cutting-edge compute that we’re able to put up from both a training and inference perspective? As we move off other nodes of processors, we’re able to host others and monetize that compute.”

“We’re very different because we’re builders, because we’re vertically integrated infrastructure players — we’re able to have multiple paths to monetization of AI. It’s the infrastructure in terrestrial before even orbital compute, which has an unlock on the cost even from terrestrial. It’s also the fact that we’ve got over 100 million people already using our consumer Grok tool today. Multiple paths of monetization for this compute investment — really differentiating for us.”

Shotwell: “We were not looking to leverage our excess compute — that was not a revenue source we planned. We actually thought about it for AI in space and racks in space. We were approached by a number of companies, Anthropic being the most aggressive and the quickest to sign deals. People are coming to us. They saw that there might be some capacity of chips being brought online.”

#2 — $28 Trillion TAM: Most Ambitious in IPO History

Shotwell: “If you think about AI only at the highest levels, from an outside looking in perspective, that TAM does look quite high. However, desk jobs are about 40% of GDP. If you could enhance all those employees doing desk-type jobs with basically a digital human, which AI brings, think about the gains your company can make. Think about how that employee who has all this extra capacity and knowledge at their fingertips — think about how much more fun their job could be.”

“We’re not talking about $28 trillion being $28 trillion of the existing economy. Artificial intelligence is going to increase the pie manifold — that’s really where you get that TAM. And look at the play it has with connectivity. Humanoid robots, cars — all AI equipped. Every one of those machines is going to need connectivity to get back to mother brain. So I think our connectivity business is going to go bananas with AI, even outside of the AI segment.”

Johnson: “The first $5.7 trillion is not enterprise AI, and is in segments where we are either leading today already or have a right to win. Spacing and connectivity alone is $2 trillion a year when you add those together. We’ve already demonstrated with broadband — 12 million customers in 160 markets around the world, an industry-leading position with a network that’s never existed before.”

#3 — Starship: 99% Cost Reduction, 20x Starlink Multiplier, Full Reusability

Shotwell: “Starship is the ship that will take humans to the moon, to Mars. The first version, V3, will take 100 metric tons to orbit. After propellant refueling, 100 metric tons to the moon or Mars. With the V3 satellite combination, one launch of Starlink V3 satellites is about 20x the capability and capacity of a Falcon launch full of Starlink satellites.”

“It’s fully reusable — the first vehicle where both the first stage and the second stage are fully and rapidly reusable. Right now, Falcon’s first stage is reusable, but not rapidly — we’ve got it down to about a week. Imagine what travel would be like if you needed a week between aircraft flights. With full and rapid reusability, we’re finally at a place where regular people can participate in the space economy.”

Johnson: “In the next year or so, as we land that second stage and start to rapidly reuse it, that’s going to drive a 10x improvement in cost per kilogram to space from what we’re already doing with Falcon, which is industry leading by a lot. When you talk about 99% improvement from where the industry had been for decades, that’s transformative.”

#4 — Orbital AI Compute: Clean Energy, No Land/Water/Power Constraints

Read more

Asia generally red: TPX -1.11%, NKY -1.36%, Hang Seng -1.48%, HSCEI -1.1%, SHCOMP -0.64%, Shenzhen -0.41%, Taiwan TAIEX -1.68%, Korea KOSPI -1.84%. SK Hynix -5%. Samsung -5%. Softbank -11%

Lots to get to, so let’s get to it. AVGO/CRWD first then the usual.

Read more

Post-close:

AVGO -12% numbers mixed at best. Ai semi sales were $10.8 vs street at $10.7B and bogeys closer to $11.5B. Q3 AI semis Guide also weaker at $16B vs street at $17.2B and buyside at $18.5B. For F26, Ai semis will come in at about $56B (street closer to $57.5B and buyside closer to $60B6) and on the call they reit F27 outlook of >$100B. I still need to dig into Anthropic timing and how that is affecting F26 - still a bit confused on that front. Software was the bright spot accelerating to 8% y/y.

Call did not go well as Hock putting a very sub-par performance, especially vs. his usual stellar performance, only reit F27, and admitted GOOGL will diversify:

Read more