Anthropic has a problem. A good problem, but a problem nonetheless. Its revenue is growing faster than it can build infrastructure.
In late 2025, Anthropic's annualized run-rate revenue was about $90 billion. By April 2026, it had crossed $300 billion. The company had prepared for 10x annual growth. Actual demand in Q1 2026 grew at roughly 80x on an annualized basis.
Every new dollar of revenue requires more GPUs. And there are never enough GPUs. So Anthropic has been doing something extraordinary: it is signing compute deals with every company that owns chips.
The list is staggering. Amazon. Google. Microsoft. Broadcom. SpaceX. And now Meta. Anthropic is not choosing suppliers. It is eating all of them.
The scale is almost impossible to grasp. In less than a year, Anthropic has locked up more than 11 gigawatts of committed compute capacity. For context, a single gigawatt can power roughly 300,000 homes. But not all of it is ready today. Some of it won't arrive until 2028.
Anthropic's strategy is simple: secure the future, even if you have to pay for it now.
The urgency comes from a simple equation: demand is outpacing infrastructure. Claude Code alone now accounts for more than 7 percent of all GitHub commits. Enterprise customers with annual spend over $1 million doubled from 500 to 1,000 in two months.
The problem is so acute that Anthropic has been forced to cap usage of Fable 5, its flagship model. Subscribers have been told they can use Fable 5 for only half their weekly quota — if they can use it at all. The free access deadline has been pushed back multiple times. On July 20, Fable 5 will finally become permanent in Max plans — at 50 percent limits.
Compute is the new currency. And Anthropic is spending it faster than anyone.
For Anthropic's partners, the deals solve different problems. For SpaceX, the Colossus 1 data center in Memphis — 300 megawatts, 220,000 Nvidia GPUs — now runs Claude, not Grok. The deal is worth $45 billion over three years, or about $1.25 billion per month. For a company preparing a $2 trillion IPO, that is a revenue line investors can understand.
For Meta, the $10 billion lease deal with Anthropic is a test of a new business model. Meta's $145 billion AI infrastructure spend this year is more than double its 2025 capex. It needs to show returns. Renting compute to Anthropic at a premium would turn a cost center into a revenue line.
Amazon and Google have the same motivation. Both are Anthropic investors. Both count Anthropic's valuation gains as profit on their balance sheets. Both are also Anthropic's largest compute suppliers. Google's Q1 earnings report showed 40 percent of its future revenue commitments came from Anthropic.
Anthropic has become a financial node. Its success feeds into Google's earnings, Amazon's AWS backlog, SpaceX's IPO narrative, and now Meta's cloud ambitions. The company is not just an AI lab anymore. It is a systemic part of the AI infrastructure economy.
The risk is that the infrastructure cannot keep up. Anthropic signed a 20-year lease with TeraWulf for 401 megawatts — but the first capacity won't arrive until 2027, and the full site won't be ready until 2028. Planning for 2028 does not help with today's compute shortage.
The other risk is the cost. Anthropic is paying a premium for "immediacy." SpaceX charges more for compute that is available today versus waiting 1-3 years for a purpose-built campus. Anthropic's deal with Meta would be smaller than the SpaceX deal, but it would be the same logic: pay more to get it now.
Anthropic's compute hunger is not a strategy. It is a response to a market moving faster than any one company can build. And for now, that is the only strategy that works.
P.S. If you are an investor in Google, Amazon, or Meta, Anthropic is no longer just a competitor — it's a customer, a revenue line, and a balance sheet item. The question is whether the compute will arrive before the revenue outgrows the infrastructure again.
