The $35 Billion GPU Landlord — How CoreWeave Became AI's Infrastructure Backbone

Two Deals in Two Days — And the Numbers Are Staggering

On April 9, CoreWeave signed a $21 billion expansion deal with Meta for AI cloud capacity through 2032. That's on top of the $14.2 billion deal they struck in September 2025, bringing the total contracted value between the two companies to $35 billion.

The very next day, April 10, CoreWeave announced a major cloud supply agreement with Anthropic. Two of the biggest AI companies in the world, locked in as customers in consecutive days.

CoreWeave stock jumped roughly 12% across those two sessions.

This isn't just corporate deal-making. It's a signal that the power structure of AI infrastructure is shifting.

From Crypto Mining to AI Cloud King

CoreWeave's origin story reads like fiction. The company started life as Atlantic Crypto in 2017 — a cryptocurrency mining operation. They had warehouses full of GPUs doing what GPUs in 2017 did: mining Bitcoin and Ethereum.

Then the 2022 crypto winter hit, and someone at the company had a simple insight: "We have a lot of GPUs. AI companies need GPUs. What if we just rented them out?"

That pivot turned a crypto miner into what is now one of the most valuable AI infrastructure companies in the world.

Year	CoreWeave Milestone
2017	Founded as Atlantic Crypto (mining)
2022	Rebranded to CoreWeave, pivoted to AI cloud
2023	Nvidia-led $2.3B investment
2024	Microsoft $10B long-term contract
2025 Sep	Meta $14.2B first deal
2026 Apr 9	Meta $21B expansion (total $35B)
2026 Apr 10	Anthropic landmark deal

Unlike AWS or Azure, CoreWeave is laser-focused on GPU cloud for AI workloads. No general-purpose computing, no enterprise SaaS hosting. Just GPUs. That specialization gave them a privileged relationship with Nvidia and first-mover access to the latest silicon.

Inside the Deal

Why Meta Doesn't Just Build Its Own Data Centers

Meta's AI-related capital expenditures for 2026 are projected at $115–135 billion. That's nearly double last year. Even for a company that size, building all that infrastructure from scratch takes too long.

CoreWeave offers speed. They're already building data centers equipped with Nvidia's latest Rubin systems. For Meta, CoreWeave fills the gap until their own facilities come online.

Here's the interesting angle: this $21 billion deal focuses on inference, not training. Meta's Llama models are open-weight — training happens internally. What Meta needs at massive scale is inference infrastructure: the GPU power to serve Muse Spark and Meta AI to billions of users in real-time across WhatsApp, Instagram, Facebook, and Messenger.

The Anthropic Deal — Toward "The AWS of AI"

CoreWeave signing Anthropic is symbolically significant. Anthropic has deep ties to Amazon — they've taken $8 billion in investment from them. Yet Anthropic signed a separate major deal with CoreWeave. That signals the multi-cloud strategy is becoming standard among frontier AI companies. Nobody wants to be locked into a single provider.

For CoreWeave, customer diversification was critical. In 2024, 62% of revenue came from Microsoft alone. With Meta and Anthropic now on the books, no single customer accounts for more than 35% of total revenue.

The Junk Bond Play

Right after the Meta deal, CoreWeave tapped the high-yield debt market for $1.25 billion. The large contract backlog immediately translated into borrowing capacity. Data center expansion requires massive upfront capital, and the contracted revenue made lenders comfortable.

CoreWeave's trajectory makes one thing crystal clear: "renting GPUs" is becoming the default mode for AI infrastructure. Building your own is increasingly the exception.

The Bigger Picture — The GPU Cloud Wars

What's playing out in AI infrastructure is essentially a GPU arms race. Here's where the major players stand:

Provider	Position	Key AI Customers
CoreWeave	AI-only GPU cloud	Meta, Anthropic, Microsoft
AWS	General + AI cloud	Anthropic (investor), thousands of startups
Azure	General + AI cloud	OpenAI (exclusive)
GCP	Custom TPU + GPU	DeepMind, internal services
Oracle	Late entrant	xAI, startups

CoreWeave's bet on pure GPU specialization is paying off. While hyperscalers run millions of general-purpose workloads, CoreWeave optimizes everything for AI training and inference. That focus attracted Nvidia's direct investment and partnership, giving CoreWeave priority access to next-generation hardware.

What This Means for You

If you're building or operating AI services, the takeaway is clear: your infrastructure options are expanding rapidly.

The AWS-Azure duopoly is cracking. As specialized cloud providers like CoreWeave win contracts from the biggest AI companies, price competition intensifies and service quality improves. Particularly for inference workloads, CoreWeave is already gaining a reputation for better price-performance than hyperscalers.

And the old formula — "serious AI infrastructure means building your own data centers" — is breaking down. If Meta, with over $100 billion in annual AI spending capacity, chooses to rent GPUs from CoreWeave, there's even less reason for smaller companies to build their own.

$35 billion. That's the AI cloud contract value secured by a company that started mining cryptocurrency seven years ago. If you want a single number to illustrate how fast the AI infrastructure landscape is moving, that's it.

Sources

• Meta commits another $21 billion to CoreWeave — CNBC
• CoreWeave, Meta Strike Another $21 Billion Deal — Bloomberg
• CoreWeave Rockets 12% on Anthropic Deal — 24/7 Wall St.