Anthropic's $900B Funding Talk Is Really a Test of Compute Strain

Anthropic's latest $900 billion valuation rumor matters only because Claude's coding boom is starting to look like an infrastructure strain story, not just another private-market brag. CNBC reported on April 29 that Anthropic is in talks to raise money at that price, though no term sheet has been signed. The sharper pressure is lower in the stack: Anthropic is telling investors demand is exploding while also admitting Claude has grown less reliable at peak hours.

That makes this less a funding story than a test of whether coding demand can force real cloud and chip capacity into place fast enough to matter. Anthropic, the AI company behind Claude, said in an April 20 company post that it would commit more than $100 billion over 10 years to Amazon Web Services, Amazon's cloud division, secure up to 5 gigawatts of new capacity to train and run Claude, and bring nearly 1 gigawatt online by the end of 2026. In the same post, it said growth had already hurt reliability and performance for users during peak hours.

The timing matters because Anthropic had already spent the previous two weeks telling the market that paid demand was compounding unusually fast. In an April 6 company post, the company said its annualized revenue run rate had surpassed $30 billion, up from about $9 billion at the end of 2025. It also said more than 1,000 business customers were each spending over $1 million a year on Claude, double the figure it had shared in February.

Those are Anthropic's numbers, not audited filings, but they help explain why the compute language changed so abruptly this month. In its February Series G announcement, Anthropic said run-rate revenue was $14 billion and Claude Code had reached more than $2.5 billion. By early April it was claiming a run rate above $30 billion. That makes the valuation rumor read less like a clean verdict on company worth and more like a bet that coding demand can justify infrastructure commitments at hyperscale.

The Google side is large too, but it is narrower than some of the coverage made it sound. Anthropic said on April 6 that it had signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity expected to come online starting in 2027. TPU stands for tensor processing unit, Google's in-house AI chip. Reuters reported that Broadcom's part of that arrangement would give Anthropic access to about 3.5 gigawatts of AI computing capacity using Google's processors starting in 2027. That is not the same as saying all of that capacity is online now or fully locked in today.

The financing commitments around those buildouts are real too, but they need the same attribution discipline. Reuters reported on April 20 that Amazon would invest up to $25 billion in Anthropic as part of the AWS deal. Four days later, Reuters reported that Anthropic said Google had committed $10 billion now in cash at a $350 billion valuation and could invest another $30 billion if Anthropic met performance targets.

That sequence is the real story inside the latest headline. Anthropic first told the market that paid demand was accelerating hard, then admitted that the growth was already straining the product, then disclosed cloud and chip agreements big enough to pull Amazon, Google, and Broadcom deeper into its expansion path. The valuation chatter came after those operating disclosures, not before.

The skeptical case is still strong. The funding talks are unfinished. The cleanest growth figures come from Anthropic itself. Type0 has already covered the earlier Google and Amazon legs of this buildout, so a bigger paper valuation alone would not justify another story. What keeps this one alive is the combination of speed and strain: Anthropic is asking the market to believe that coding revenue can justify multi-gigawatt capacity plans before reliability problems start to damage the narrative.

That pressure does not stop with Anthropic. OpenAI now faces a rival that appears to be turning coding demand into cloud leverage unusually fast. Amazon, Google, and Broadcom face a harder test too, because these are no longer abstract AI capex claims. They are promises to deliver real chips, real power, and real datacenter throughput to a service whose operator has already said growth is stressing the system.

What to watch next is not whether investors float an even bigger number. It is whether Claude becomes more reliable as demand steepens, and whether the AWS and Google-linked capacity Anthropic has described arrives on a schedule that matters before users feel the strain more sharply.