Skip to content

CoreWeave Signed Meta for $21 Billion. The Next Morning, Anthropic Called.

DS
LDS Team
Let's Data Science
10 min
Meta committed an additional 21 billion dollars through 2032, bringing its total CoreWeave bill to roughly 35 billion. Eighteen hours later, Anthropic signed its own multi-year cloud deal. CRWV closed Friday up 11.5%, and nine of the top ten AI labs now rent from the same landlord.
On Thursday morning, April 9, CoreWeave put out a two-page press release that almost no retail investor would have predicted a year ago. Meta Platforms had committed an additional $21 billion of cloud spend through December 2032. Stacked on top of the $14.2 billion Meta had already promised six months earlier, the Facebook parent now owed CoreWeave roughly $35 billion over the next six years.

The stock ticked up 3.49% that day, closing at roughly 92 dollars. A normal reaction to a very large deal.

Then Friday happened.

Before the New York open on April 10, CoreWeave's second press release of the week hit the wire. Anthropic, the Claude maker that had just signed a separate 3.5-gigawatt chip deal with Google and Broadcom two days earlier, had agreed to rent CoreWeave capacity for its production Claude workloads. The financial terms were undisclosed. Wall Street did not wait for them. CRWV opened higher, kept climbing through the morning, and closed at $102.58, an 11.5% single-day gain on volume of roughly 71 million shares, multiples above its three-month average.

In 48 hours, CoreWeave had collected commitments or contracts from the four most important AI model builders on the planet. With the addition of Anthropic, it now counts nine of the top ten leading AI model providers as customers, a list that runs through OpenAI, Microsoft, Meta, Mistral, Cohere, IBM, and Nvidia itself. The one frontier-scale holdout is Google, which builds its own Tensor Processing Units and rents compute mostly to itself.

The Meta Expansion Is the Backbone of the Deal Week

Meta's announcement on Thursday was the structural one. The new $21 billion sits on top of a $14.2 billion commitment that closed in the fall of 2025, and the combined backlog now runs through the end of 2032. The spending cadence is back-weighted: the bulk of the new dollars flow between 2027 and 2032, when the latest hardware is expected to be in wide deployment.

CoreWeave CEO Michael Intrator framed the Meta expansion in a single line:

"This is another example that leading companies are choosing CoreWeave's AI cloud to run their most demanding workloads." — Michael Intrator, Co-founder, CEO and Chairman, CoreWeave (CoreWeave press release, April 9, 2026)

The hardware detail in the Meta deal is the part ML engineers should not skim past. The expanded capacity will include some of the first commercial deployments of Nvidia's upcoming Vera Rubin platform, the next-generation AI supercomputing system that succeeds the Blackwell generation. Volume shipments of Rubin are expected in the second half of 2026. For the practitioners who will actually run jobs on this silicon, the takeaway is that Meta has just locked in priority Rubin slots on a non-Meta cloud, a hedge against its own AI infrastructure buildout running late.

The last time a single customer committed roughly 21 billion dollars of cloud capacity to a single provider in one announcement, it was Microsoft and Oracle. That deal took weeks of reporting. This one came in a morning press release.

Anthropic Was the Domino That Sent the Stock Vertical

If Meta was the structural deal, Anthropic was the one that made traders sit up. Here is why.

Until Friday, the conventional narrative around Anthropic's compute story had been clean: a company that had quietly built itself around Google's TPUs and was about to sign one of the largest custom silicon commitments in the industry. Two days earlier, on April 7, Anthropic had confirmed an expanded agreement with Google and Broadcom for 3.5 gigawatts of next-generation TPU capacity. That deal, read narrowly, suggested Anthropic was doubling down on specialized chips outside the Nvidia ecosystem.

The CoreWeave announcement complicated that picture. CoreWeave is almost entirely an Nvidia shop. Signing Anthropic meant adding meaningful Nvidia GPU capacity to the Claude infrastructure stack at the same time the company was ramping TPU deployments. Reuters had reported earlier in the week that Anthropic was also exploring custom chip designs of its own.

Put together, the picture that emerged on Friday morning was an Anthropic willing to run Claude on anything that shipped on time: TPUs from Google, GPUs from CoreWeave, and a custom chip program hedged against both. The strategic message was explicit: Anthropic does not intend to be capacity-constrained for a second time in 2026.

Intrator's language on the Anthropic deal was broader:

"AI is no longer just about infrastructure. It is about the platforms that turn models into real-world impact." — Michael Intrator, CoreWeave (CoreWeave press release, April 10, 2026)

The financial terms were not disclosed. The deployment was described as phased, with expansion options, and compute coming online later in 2026.

For context: LDS covered Anthropic's Google and Broadcom deal earlier this week in Anthropic Tripled Revenue in 90 Days. Then It Bought 3.5 Gigawatts.

The 48-Hour Timeline

MONDAY, APRIL 6, 2026
Anthropic expands Google and Broadcom deal
Anthropic confirms 3.5 gigawatts of next-generation TPU capacity with Google, an agreement built around Broadcom-manufactured silicon. Reuters reports the same week that Anthropic is studying its own custom chip program.
THURSDAY, APRIL 9, 2026
Meta expands CoreWeave deal to roughly $35 billion total
CoreWeave announces an additional $21 billion commitment from Meta, running through December 2032 and including first deployments of Nvidia Vera Rubin systems. CRWV closes at $92, up 3.49%. Meta shares rise 2.6%.
FRIDAY, APRIL 10, 2026 · PRE-MARKET
CoreWeave announces multi-year Anthropic deal
The press release states that nine of the leading ten AI model providers now use CoreWeave. Financial terms are not disclosed. The deployment is phased with expansion options and starts later in 2026.
FRIDAY, APRIL 10, 2026 · CLOSE
CRWV closes up 11.5% on heavy volume
The stock closes at $102.58, a gain of about $10.58 on the day. Volume runs around 71 million shares, multiples above its three-month average. Year to date, CRWV is now up by double digits.

The Customer List Is Now a Monopoly of Influence

A table makes the scale of what CoreWeave just accomplished clearer:

AI Model ProviderCoreWeave Customer?Notes
OpenAIYesExisting multi-year capacity agreements
AnthropicYes (new, April 10, 2026)Phased deployment starting later in 2026
MetaYesRoughly $35 billion total through December 2032
MicrosoftYesLong-standing capacity sub-leasing arrangement
MistralYesNamed in CoreWeave customer disclosures
CohereYesNamed in CoreWeave customer disclosures
IBMYesEnterprise AI workloads
NvidiaYesUses CoreWeave for internal research
GoogleNoBuilds its own TPUs, rents compute mostly to itself

Nine of ten. For a company that only completed its Nasdaq listing in March 2025, this is an extraordinary concentration of the industry's demand on a single provider's roadmap. Every frontier lab that buys Nvidia silicon, with the lone exception of Google, is now financially tied to the execution of a single Livingston, New Jersey provider that was still operating as a crypto mining outfit less than a decade ago.

The Other Side: A Debt-Funded Landlord Is Still Carrying the Risk

The bull case for CoreWeave is cleanly summarized by Friday's price action. The bear case is sitting in the company's own filings.

CoreWeave is in the middle of a capital-intensive data center buildout, financed in large part through high-yield debt offerings and equity raises. Bloomberg reported Friday that a new CoreWeave junk bond, issued around the Meta announcement, had jumped in secondary trading on the back of the Anthropic news, which is a useful signal of credit appetite but also a reminder of how the buildout is being paid for. Critics inside and outside the investor base point to one structural concern: CoreWeave's revenue is enormous and contracted, but its capex bill is even larger in the near term, and the company has to keep raising money until the backlog starts converting to cash in volume.

There is a second risk that practitioners should care about more than the stock price. The entire deal structure assumes Nvidia's Vera Rubin roadmap lands on time. The Meta agreement explicitly names Rubin deployments. The Anthropic agreement was described in terms of next-generation GPU capacity. If Rubin ships late, or if yields disappoint, the 2027 and 2028 portions of these contracts land in a harder environment than the press releases imply.

There is also the concentration question. Nine of ten top AI labs on one provider is extraordinary commercial validation. It is also a single point of failure for an industry that has spent the last year talking about resilience. The CoreWeave outage risk is now effectively the industry's outage risk. A prolonged incident affecting a major cluster would land across Claude, ChatGPT, Llama training runs, and enterprise AI workloads simultaneously. It would do so at precisely the moment the hyperscalers are laying off tens of thousands of engineers to fund this same infrastructure buildout.

CoreWeave has not published a public commentary on this concentration risk. Its customers, individually, continue to diversify. Meta runs its own data centers. Anthropic uses Google TPUs. OpenAI has Microsoft Azure and the new Stargate commitments. But the shared dependency on CoreWeave capacity is now a fact of the market, and any future regulatory review of AI infrastructure concentration will have to reckon with it.

The Bottom Line

Two press releases in 48 hours turned CoreWeave from a fast-growing AI cloud provider into the indispensable landlord of the 2026 AI buildout. Meta's total commitment is now roughly $35 billion through 2032. Anthropic is on board for a multi-year deal whose size was not disclosed but whose symbolic weight was immediate. Nine of the top ten AI model providers are now paying CoreWeave rent.

For ML engineers and data scientists, the immediate implication is practical: the hardware that will train and serve the next wave of frontier models is going to live inside CoreWeave data centers more often than not, running Nvidia Rubin as it becomes available. The strategic implication is harder. CoreWeave's execution is now the industry's execution. Its debt schedule is now the industry's debt schedule. Its Rubin delivery dates are now the industry's delivery dates.

As Intrator put it in a single line buried in Friday's press release: the industry is no longer just about infrastructure. It is about the platforms that turn models into real-world impact. The question the market answered on Friday, with an 11.5% close, is whether one platform can hold that much of the industry on its shoulders. The answer it offered was: for now, yes.

Sources

Practice with real Ad Tech data

90 SQL & Python problems · 15 industry datasets

250 free problems · No credit card

See all Ad Tech problems
Free Career Roadmaps8 PATHS

Step-by-step roadmaps from zero to job-ready — curated courses, salary data, and the exact learning order that gets you hired.

Explore all career paths