# Anthropic Just Won the AI War.

> Anthropic courts a $900B valuation while the Musk v. Altman trial exposes OpenAI's governance and Meta executes 8,000 layoffs. Weekly executive intel.

- Canonical: https://siliconvalleyconfidential.com/en/dossier/anthropic-just-won-the-ai-war/
- Site: Silicon Valley Confidential (https://siliconvalleyconfidential.com) — weekly executive intelligence on Silicon Valley and global tech
- Author: Jose Luis Cases (https://es.linkedin.com/in/jose-luis-cases-lozano)
- Language: en
- Published: 2026-05-16 (original LinkedIn edition: https://www.linkedin.com/pulse/anthropic-acaba-de-ganar-la-guerra-ia-jose-luis-cases-1ej4e/)

---
MY TAKE
-------

Some weeks a lot happens and nothing really happens, and some weeks three things happen and you stop, think, and say "something's cooking here." This is one of those weeks.

It looks like the center of gravity in AI has shifted from OpenAI to Anthropic — and it shifted in public, with data and paperwork. Three data points tell the story.

Data point one.

On Tuesday, May 12, Bloomberg reported that Anthropic is in talks for a round of at least $30 billion at a valuation above $900 billion.

That's not a Twitter rumor — it comes straight from Bloomberg. In February, Anthropic closed its Series G at $380 billion. In other words: in three months, the company is worth 2.5 times more.

In the secondary market, where employee shares change hands ahead of an IPO, Anthropic is already implicitly trading at between $1 trillion and $1.15 trillion. Compare that with OpenAI, which in the same secondary market trades only 3% above its last primary. The divergence, as I see it, is enormous.

Data point two.

While Anthropic keeps climbing, OpenAI is stuck in courtrooms, and three facts together are not doing arch-villain Sammy any favors.

The first, covered last week: Mira Murati said Sam Altman lied to her. The second, this week: Satya Nadella took the stand on May 11 and admitted he feared Microsoft would become "the next IBM" in its relationship with OpenAI. And the third: Altman testified for two days while five witnesses were introduced, including Dario Amodei, all of whom questioned his honesty. The jury verdict lands next week, but the reputational damage is permanent.

Data point three.

On Wednesday, May 20, Meta will execute 8,000 layoffs. The most relevant piece this week is the one circulating on Blind and Tom's Hardware: Meta has also canceled 6,000 open roles, set severance at 16 weeks plus two per year of service, and confirmed that the internal justification remains the same — the aggregate AI capex for 2026 (between $125 billion and $145 billion) is coming out of the payroll line.

What do the three data points have in common?

The power transfer inside the sector isn't just financial — it's narrative. Anthropic wins because it sells compute to companies that have decided to replace people with GPUs. OpenAI loses not because its model is worse, but because the trial transcript has made it public that its leadership team is dysfunctional.

And Meta, which isn't in the model business but is in the infrastructure business, has decided the only way to finance its bet is to move capital from the payroll line to the compute-vendor line — and tell its employees exactly that, in those words.

My read is that we're watching the oligopoly consolidate in real time. Anthropic and OpenAI are the two poles, and the valuation premium is migrating toward the one with fewer management problems. The second-tier companies — Thinking Machines, Mistral, Cohere, the ones founded as "the alternative" — are watching Meta drain their teams at prices they can't match.

If I had to bet on the next 12 months, I'd bet Anthropic goes public in the fall, Meta picks off three or four more Thinking Machines founders, and OpenAI settles the case with a deal that doesn't break the for-profit conversion.

If you want more detail and have the time, I'll leave you with my minions.

THE BOMBSHELL OF THE WEEK
-------------------------

### Anthropic reprices to $900 billion while Microsoft confesses in court its structural fear of OpenAI

The story has two heads and one common conclusion. The financial head: on Tuesday, May 12, Bloomberg broke that Anthropic is in advanced talks to raise at least $30 billion at a valuation above $900 billion — 2.5 times the valuation closed in February ($380 billion in the Series G). No term sheet has been signed yet, but the round could close by the end of May. ([Bloomberg](https://www.bloomberg.com/news/articles/2026-05-12/anthropic-in-talks-to-raise-30-billion-at-900-billion-valuation)
)

The judicial head: on Monday, May 11, Satya Nadella took the stand in the Musk v. Altman trial and declared under oath two things that stay in the record forever.

Timeline of the week:

*   Monday, May 11. Nadella testifies. He acknowledges that Musk never contacted him personally to object to Microsoft's investment in OpenAI. But the strategically most damaging moment comes when Musk's lawyers introduce as evidence an internal email in which Nadella said he feared Microsoft would become "the next IBM" — a reference to the company that handed its partner the platform and then watched from the sidelines. He called the OpenAI bet "a one-way door." A January 2023 memo from Brad Smith (Microsoft's president), also admitted into evidence, projected a cumulative return of $92 billion on the cumulative $13 billion investment. And in private emails, Nadella referred to OpenAI's board as "Amateur City." ([CNBC](https://www.cnbc.com/2026/05/11/microsoft-ceo-satya-nadella-musk-altman-trial.html)
    , [GeekWire](https://www.geekwire.com/2026/musk-v-altman-satya-nadella-was-worried-about-microsoft-being-the-next-ibm-in-openai-deal/)
    )
*   Tuesday, May 12. Altman testifies in the morning. His central line: "Musk didn't just leave — he tried to kill it." He rejects the "stolen charity" thesis. That afternoon, Bloomberg publishes the news of Anthropic's round at $900 billion. The calendar coincidence is hard to ignore. ([NPR](https://www.npr.org/2026/05/12/nx-s1-5811730/openai-sam-altman-testimony-elon-musk-trial)
    )
*   Wednesday, May 13. Altman returns to the stand for cross-examination. Musk's lawyer, Steven Molo, introduces a pattern: five witnesses who have questioned Altman's honesty over the course of his career, among them Dario Amodei (Anthropic), Tasha MacCauley and Sue Yoon (former OpenAI board). Mira Murati's deposition from the previous week is used directly to impeach credibility. ([CNBC](https://www.cnbc.com/2026/05/13/altman-musk-trial-testimony-takeaways.html)
    , [Axios](https://www.axios.com/2026/05/13/openai-trial-sam-altman-elon-musk-ai-safety)
    )
*   Thursday, May 14. Testimony phase closes. Closing arguments. Musk's close: Altman is structurally a liar, and five witnesses confirm it. Altman's close: Musk abandoned a charity and now wants credit for its success. The verdict is advisory — Judge Yvonne Gonzalez Rogers has the final word on liability. ([CNBC](https://www.cnbc.com/2026/05/14/closing-arguments-jury-openai-musk-altman.html)
    )
*   Friday, May 15. Jury deliberation begins Monday, May 18. Verdict expected within the next ten days. ([The Ringer](https://www.theringer.com/2026/05/15/tech/sam-altman-defense-testimony-elon-musk-trial)
    )

The signal: the transfer of valuation premium from OpenAI to Anthropic has stopped being a conjecture. It's now a quantitative fact. OpenAI trades in the secondary at 103% of its last primary ($880 billion). Anthropic trades at 250% ($1 trillion implied versus $380 billion in February). The differentiating factor isn't the product — Claude Code books $2.5 billion in ARR, ChatGPT books more. The differentiating factor is governance. This week's trial has turned OpenAI's dysfunctional governance into a measurable data point: five witnesses against the CEO, a memo from the majority partner calling the board "Amateur City," and a verdict pending. For a CFO renegotiating a Copilot contract, or for an institutional fund weighing OpenAI's IPO when it comes, these data points change the price.

* * *

POWER MOVES
-----------

### Thinking Machines Lab loses a third of its founding team (May 14 report)

As of May 14, roughly thirteen members of Thinking Machines Lab's original founding team have left the company. Three of the six co-founders are gone. The public timeline: in January, Barret Zoph and Luke Metz returned to OpenAI. In March, Devendra Chaplot signed with xAI and Joshua Gross with Meta Superintelligence Labs. In April, Andrew Tulloch (co-founder) joined Meta for a reported package of $1.5 billion over six years. This week's report puts the total exodus at a third of the founding team. ([American Bazaar Online](https://americanbazaaronline.com/2026/05/14/one-third-of-thinking-machines-lab-founding-team-exits-480782/)
, [Inc.](https://www.inc.com/ava-levinson/thinking-machine-labs-staff-exodus/91289207)
)

The context that connects the story: Meta tried to buy Thinking Machines Lab for roughly $1 billion — Meta officially denies it. Murati rejected the offer. Meta responded by signing the founders one by one, at amounts that in aggregate exceed the rejected acquisition price. Five original founders are now at Meta. Why it matters: Murati raised $12 billion in the largest seed round in history and still hasn't shipped a product. With a third of the founding team out before the first release, the $12 billion thesis is under real pressure. It's an anomaly: a company burning through its founding cohort before ever leaving port.

### LinkedIn cuts 875 roles and never mentions AI

Microsoft confirmed on May 13 that LinkedIn is eliminating roughly 875 roles — 5% of headcount — across the Global Business Organization, marketing, engineering, and product. Revenue grew 12% year over year in the quarter ended March 31. The official justification: "reinventing how we work with agile teams." ([Reuters / Yahoo Finance](https://finance.yahoo.com/news/linkedin-layoffs-2026-company-cutting-150608852.html)
)

Why it matters: LinkedIn is the first of Microsoft's big subsidiaries — but not the last, according to internal sources — to go under the efficiency knife coming down from Redmond. The most significant thing isn't the number; it's the silence. While Cloudflare, Freshworks, Upwork, Meta, and others have made "it's because of AI" the official communications line, LinkedIn has deliberately avoided that narrative. Why? Because LinkedIn sells to professionals and HR teams — any admission that AI replaces marketing and product destroys its commercial pitch. Watching a Microsoft company refuse to use the frame that Zuckerberg himself is turning into doctrine is, in itself, a story.

MONEY TALKS
-----------

Anthropic in talks for $30B at a $900B valuation
------------------------------------------------

Covered in "The Bombshell" section. The additional financial element worth highlighting: the secondary market is pricing Anthropic at an implied $1–1.15 trillion. The divergence from the proposed primary valuation ($900 billion) suggests secondary buyers are willing to pay a 10–25% premium over the formal round — typically an indicator of pent-up institutional demand ahead of an expected IPO. Goldman Sachs, JPMorgan, and Morgan Stanley are listed as underwriters. Metaculus assigns a 66% probability to an S-1 filing before July 1. ([Bloomberg](https://www.bloomberg.com/news/articles/2026-05-12/anthropic-in-talks-to-raise-30-billion-at-900-billion-valuation)
, [Yahoo Finance Markets](https://finance.yahoo.com/markets/stocks/articles/anthropic-beats-openai-secondary-markets-213828157.html)
)

The signal: if the round closes at the announced range, it's the largest single private financing in history. More relevant: the speed. 2.5x in three months isn't a market correction — it's a fundamental repricing driven by ARR (from $87 million in January 2024 to a $30 billion run rate in April 2026) and by the transfer of narrative premium away from OpenAI.

### OpenAI DeployCo: $4 billion of PE for a professional-services entity (PYMNTS, May 11)

PYMNTS reported on May 11 the full financial detail of "DeployCo" — the independent company launched by OpenAI — backed by $4 billion contributed by TPG Capital, Bain Capital, Advent International, and Brookfield Corporation ($500 million each), alongside Goldman Sachs, SoftBank, and Warburg Pincus. The consulting partners are Bain & Company, Capgemini, and McKinsey. OpenAI keeps the majority. The vehicle deploys "Forward Deployed Engineers" embedded inside enterprise clients to redesign workflows around AI. ([PYMNTS](https://www.pymnts.com/news/artificial-intelligence/2026/openai-launches-4-billion-dollar-company-accelerate-enterprise-ai-adoption/)
)

The signal: OpenAI is copying Palantir's strategy — high-friction, high-margin deployment — and doing it with other people's money. This competes head-on with Accenture, IBM Consulting, and Deloitte's AI practice. More relevant: it explains why OpenAI isn't rushing to an immediate IPO — they're building monetization layers that justify the exit valuation. For the big integrators, the problem is structural. Are you going to hire Accenture to implement OpenAI's AI, or are you going to hire OpenAI's team directly?

### Cowboy Space: $275M for data centers in orbit

Cowboy Space (formerly Aetherflux), founded by Baiju Bhatt (co-founder of Robinhood), has closed a $275 million Series B led by Index Ventures, with IVP, Blossom, ASMSA, a16z, and Breakthrough Energy Ventures. Valuation: $2 billion. The thesis: terrestrial data center capacity is a structural bottleneck for AI; Cowboy plans to integrate one-megawatt data centers into rocket second stages, powered by solar energy in orbit. First proprietary launch: 2028. ([TechCrunch](https://techcrunch.com/2026/05/11/there-arent-enough-rockets-for-space-data-centers-cowboy-space-raised-275-million-to-build-them/)
)

The signal: a fund of Index Ventures' caliber leading a $275 million round into a technically extreme idea says more about the state of the AI capex market than any McKinsey report. There is institutional capital willing to bet that the physical constraint (land, water, power) becomes prohibitive within five years. If the hyperscalers are in orbital compute by 2030, this round will look conservative. If not, it's the archetypal bubble valuation.

PRODUCT SECRETS
---------------

### Microsoft in talks to acquire Inception (diffusion-based LLMs)

Microsoft is in active talks to acquire Inception, a Stanford-founded startup that uses diffusion-based token generation (parallel, not autoregressive) instead of the traditional transformer architecture. Microsoft's M12 fund already participated in last year's $50 million seed. The founders are asking for a price above $1 billion. SpaceX appears as a competing bidder — an unusual data point worth tracking. ([WinBuzzer](https://winbuzzer.com/2026/05/15/microsoft-reportedly-seeking-to-acquire-ai-startup-xcxwbn/)
)

The strategic context matters: Microsoft considered buying Cursor earlier this year and walked away over regulatory exposure around GitHub Copilot. The Inception deal sidesteps that problem — it doesn't compete with Copilot, it complements it. If the diffusion architecture delivers the promised speed and cost improvements at scale, Microsoft would have an alternative to OpenAI inside its own stack. The "next IBM" narrative Nadella admitted to in court starts to show up here: Microsoft is building optionality.

### Runway Gen-4.5 beats Veo and Sora on the Video Arena leaderboard

Runway's Gen-4.5 tops the Artificial Analysis Video Arena with 1,247 Elo points, beating Google's Veo and OpenAI's Sora. The company is repositioning its narrative from "video AI" to "general world models" — arguing that intelligence built from video and physical simulation, not from text, is the most defensible moat. Current valuation: $5.3 billion, with $40 million in ARR added in Q2 2026 alone. ([TechCrunch](https://techcrunch.com/2026/05/15/runway-started-by-helping-filmmakers-now-it-wants-to-beat-google-at-ai/)
)

The signal: Runway is racing to claim the "world model" category before Google's Genie-3 or a better-funded competitor does. The narrative pivot — from creative tool to frontier competitor — is deliberate. For Hollywood and the creative agencies, this pivot may mean the tool is no longer built for them. For investors in defense, gaming, and robotics, world models are an entirely different category from video generation.

REAL NUMBERS
------------

Cisco Q3 FY2026 (published May 13). Revenue of $15.84 billion versus $15.56 billion expected (+12% year over year). EPS of 1.06 versus 1.00 expected. The number that moved the stock +15%: full-year AI infrastructure order guidance was raised from $5 billion to $9 billion. Cumulative YTD AI orders from hyperscalers: $5.3 billion. Q4 guidance: $16.7–16.9 billion, well above the $15.82 billion consensus. Implication: Cisco is the cleanest public proxy for AI networking demand, and the numbers say the hyperscaler buildout is not slowing down. ([CNBC](https://www.cnbc.com/2026/05/13/cisco-csco-q3-earnings-report-2026.html)
)

Applied Materials Q2 FY2026 (published May 14). Record revenue of $7.91 billion (+11% year over year). Non-GAAP gross margin of 50% (+80 bps year over year). Record GAAP EPS of 3.51. Dividend raised 15% to 0.53/share — ninth consecutive year. Combined with the $120 million purchase of NEXX from ASMPT (expected to close July 29), it confirms AMAT is positioning aggressively in advanced packaging for AI chips. ([GlobeNewswire](https://www.globenewswire.com/news-release/2026/05/14/3295316/0/en/Applied-Materials-Announces-Second-Quarter-2026-Results.html)
, [TrendForce](https://www.trendforce.com/news/2026/05/05/news-applied-materials-to-acquire-asmpt-nexx-for-120m-expanding-panel-level-advanced-packaging-portfolio/)
)

Anthropic ARR (run rate). $30 billion annualized as of April, versus $87 million in January 2024. Claude Code alone: $2.5 billion run rate. Year-over-year growth: 80x. ([VentureBeat](https://venturebeat.com/technology/anthropic-says-it-hit-a-30-billion-revenue-run-rate-after-crazy-80x-growth)
)

Tech layoffs YTD 2026: 128,270 people across 286 events, per TrueUp's tracker — an average of 1,002 people per day. The dominant funding source for all AI capex is the payroll line of high-margin software companies. ([TrueUp](https://www.trueup.io/layoffs)
)

Aggregate AI capex the Anthropic round is measured against. The confirmed public figures: Meta $125–145 billion, Google $40 billion in new data centers alone, Amazon $50 billion, OpenAI $100 billion in infrastructure with NVIDIA. The $30 billion round Anthropic is preparing represents, against these numbers, roughly six months of a single hyperscaler's capex.

THE DRAMA
---------

### Nadella's "next IBM memo" and the problem Microsoft just made public

Satya Nadella's internal memo confessing his fear that Microsoft would become "the next IBM" vis-à-vis OpenAI is now evidence in a federal trial. That means Microsoft's internal narrative — that its strategic partner could eat its market — is public knowledge. Combined with Brad Smith's memo projecting $92 billion in returns on $13 billion invested, and the dismissive "Amateur City" reference to OpenAI's board, Microsoft has gone from collaborative partner to declared competitor in three days of testimony. ([The Next Web](https://thenextweb.com/news/nadella-feared-microsoft-would-become-the-next-ibm-the-trial-reveals-how-much-he-paid-to-make-sure-it-didnt)
)

The operational impact is immediate: any enterprise customer negotiating a Copilot contract now has leverage. They know that Microsoft, internally, considers OpenAI a structural risk and not a stable partner. Any pricing round can use the memo as an argument. And, internally, the news explains the chain of moves over the past week — the interest in Inception, the accelerated development of in-house models under the Phi family, the insistence on Mistral as a second source. Microsoft isn't "considering optionality." Microsoft is executing contingency.

### The Altman/Helion countdown starts hitting the trial this week

Although the House Oversight Committee letter with its May 22 deadline was sent the previous week, the public trigger that amplified the case landed this week: during Altman's cross-examination on May 12 and 13, Musk's lawyers formally reintroduced the Helion Energy matter — the nuclear fusion company where Altman holds significant personal equity — as part of the pattern of conflicts. GeekWire's May 15 coverage connects the two threads for the first time: the trial in Oakland and Congress in Washington are using exactly the same evidence. ([GeekWire](https://www.geekwire.com/2026/openai-ceo-sam-altmans-stake-in-helion-energy-draws-scrutiny-in-musk-trial-and-on-capitol-hill/)
)

The detail that hasn't gotten enough attention and came out in deposition this week: some OpenAI employees deliberately avoided the internal channel where the Helion investment was discussed, fearing their messages would become legal evidence. That chilling effect on internal communication at a company building systems adjacent to AGI is a governance signal that goes far beyond a single investment decision. What gets released (or denied) on May 22 will determine whether this escalates to a formal hearing or a subpoena.

THE WEEK AHEAD
--------------

Monday, May 18 - Jury deliberation begins in Musk v. Altman. The verdict is advisory, but its content will determine Judge Gonzalez Rogers's room to maneuver. A verdict for Musk on the "stolen charity" theory opens the door to structural remedies on OpenAI's for-profit conversion. A verdict for Altman closes the case but leaves the entire conflict-of-interest audit intact. Market pricing: any court order touching OpenAI's corporate structure moves the secondary price immediately.

Wednesday, May 20 - NVIDIA reports Q1 FY2027 (after market close). Consensus: $78.8 billion in revenue (management guidance: $78 billion +/-2%), EPS of ~1.77. The options market is pricing an 8% move. The number almost certainly beats. What matters: Q2 guidance and commentary on post-export-control demand in China. If Huang gives demand visibility through H2 2026, the capex cycle has another 12–18 months. If guidance is cautious, everything adjacent to infrastructure gets repriced. ([Motley Fool](https://www.fool.com/investing/2026/05/13/nvidia-reports-its-fiscal-2027-q1-earnings-may-20/)
, [247WallSt](https://247wallst.com/investing/2026/05/14/nvidia-is-up-20-in-a-month-could-the-may-20-earnings-report-knock-it-right-back-down/)
)

Wednesday, May 20 - Meta executes the 8,000 layoffs. Severance of 16 weeks base plus 2 weeks per year of service, 18 months of health coverage. Affected teams: Reality Labs, Facebook's social division, recruiting, sales, global operations. Watch for internal leaks in the following 72 hours — Zuckerberg's follow-up internal memos have always leaked.

Thursday, May 21 - Walmart Q1 FY2027 (premarket). The most relevant proxy for consumer resilience against tariff pressure. For tech, the data point to watch is e-commerce health and commentary on AI advertising spend. ([StockTitan](https://www.stocktitan.net/news/WMT/walmart-to-host-first-quarter-earnings-conference-call-may-21-ulnn8vj5sn1j.html)
)

Friday, May 22 - Oversight Committee deadline for OpenAI. Documents on Altman/Helion conflicts of interest. What gets delivered (or not) determines whether the story stays contained or escalates to a public hearing with a subpoena.

Monday 26 - Tuesday, May 27 - Palo Alto Networks and Snowflake. Palo Alto reports on the 26th, Snowflake on the 27th. PANW is the indicator for the adoption pace of AI-native security. SNOW is the direct test of the thesis that "the real bottleneck is data infrastructure, not the model."

End of May - Potential close of the Anthropic round. Bloomberg points to the end of the month. If it materializes at the $900 billion range, it's the financial event of the year.

Thanks for reading. Onward.

* * *

Silicon Valley Confidential is published weekly. Executive intelligence verified with primary sources. Compiled: May 16, 2026 | Period: May 9–16, 2026