Anthropic Is Renting Compute From Everyone, Including Its Enemies
What happened
On May 5, Anthropic committed $200 billion to Google Cloud over five years starting in 2027, representing more than 40% of Alphabet's entire disclosed cloud revenue backlog. Five days later, Anthropic simultaneously rented SpaceX's Colossus 1 supercomputer in Tennessee, gaining access to 220,000 Nvidia GPUs and 300 megawatts of compute capacity despite Elon Musk having publicly called Anthropic 'evil' and 'misanthropic' three months earlier. The deal followed direct meetings between Anthropic leadership and Musk in April. The company has now secured compute from Google TPUs, AWS Trainium chips via Project Rainier, Nvidia GPUs via Colossus 1, CoreWeave's GB300 racks, and an Akamai deal worth $1.8 billion. Alphabet briefly surpassed Nvidia in market capitalization following the cloud commitment announcement.
Anthropic is not choosing compute partners based on strategic alignment or price. It is acquiring every GPU cluster it can access at any cost because training and inference demand is growing faster than its ability to procure infrastructure through normal channels.
Prediction Markets
Prices as of 2026-05-12 — the analysis was written against these odds
The Hidden Bet
The $200 billion Google Cloud commitment is a strategic partnership between equals
Anthropic's current annual revenue is under $1 billion. The commitment begins in 2027 and requires spending $40 billion per year with a single vendor. This is not a commercial contract in any normal sense -- it is a declaration that Anthropic's entire existential bet is on Google's TPU roadmap. If Google's TPUs underperform, or if Google's relationship with Anthropic becomes adversarial, Anthropic has no exit ramp at that scale.
Renting Musk's supercomputer shows Anthropic can maintain independence from any single provider
The Colossus 1 deal was driven by emergency. Hyperscaler procurement for equivalent GPU clusters takes 12-18 months. Anthropic went to Musk because it could not get compute fast enough through any other channel. Dario Amodei acknowledged this directly. Independence is not the story; desperation is the story, and it says something significant about how tight the compute market actually is.
Alphabet's 160% stock run is sustainable given the Anthropic relationship
Over 40% of Google Cloud's entire disclosed backlog is one customer whose annual revenue is a fraction of its annual spend commitments. Anthropic is a cash-burning company whose survival requires continued massive fundraising. If Anthropic's $50 billion fundraising round stalls, or if its model leadership slips (Polymarket gives Google only 13.5% odds of having the best AI model at end of May), the backlog commitment becomes a liability, not an asset.
The Real Disagreement
The fork is between two views of what Anthropic's compute strategy represents. Under the first view, Anthropic is the rare AI lab that understood infrastructure as a competitive moat before others did and locked in capacity at prices that will look cheap in retrospect. Under the second view, Anthropic is over-leveraged on compute commitments that assume demand growth far exceeding its current revenue trajectory, in an era where model performance is becoming commoditized and the pricing power to justify $40 billion annual cloud spend does not yet exist. The first view is what Anthropic's investors are telling themselves. The second view is what the $200B-vs-under-$1B-revenue ratio implies. Neither is provably wrong yet, but only one of them needs to be right.
What No One Is Saying
Musk called Anthropic evil in February and cut a deal with them in May. He did not change his mind about the company. He changed his mind about the money. SpaceX-xAI's Colossus 1 just became a revenue-generating asset because Musk owns the only cluster with fast deployment timelines. Anthropic needed compute so badly it handed Musk a business justification for every future attack on the company: 'we profit from them.' That changes the dynamic of every future Musk statement about Anthropic from ideological critique to competitive positioning.
Who Pays
Enterprise customers building on Claude
The dependency becomes visible in 2027 when the $200B commitment activates
They are running on Google TPU infrastructure without knowing it. If the Google-Anthropic relationship deteriorates, or if Anthropic's financial position forces renegotiation, the compute substrate under their applications can change. Vendor lock-in risk is hidden inside a product that markets itself as an independent alternative to Google AI
Independent AI labs without Anthropic's funding access
Ongoing; the CoreWeave, Colossus 1, and Akamai deals are already reducing available capacity
Anthropic's ability to lock in multi-year compute contracts across all major providers tightens the remaining supply for competitors. GPU scarcity was already the binding constraint; Anthropic's multi-provider hoarding makes it worse for labs that cannot match its fundraising
Alphabet shareholders
The reckoning is in 2027 when the first large tranches of the commitment are due
The concentration risk is real. If Anthropic stumbles -- and companies burning more than they earn always carry that risk -- Google's cloud backlog number deflates sharply. The stock was repriced on a $200B commitment from a company that doesn't have $200B
Scenarios
Anthropic wins the frontier race
Claude remains the enterprise AI leader. Revenue grows to match and eventually justify the compute commitments. The $200B Google deal looks prescient in hindsight. Alphabet's cloud dominance in AI becomes self-reinforcing as Claude and Google TPUs become inseparable in enterprise IT.
Signal Anthropic's $50B fundraising closes at or above the $1 trillion valuation target; Claude's enterprise market share holds above 30% through 2027
Compute arbitrage strategy collapses
Model performance converges across labs; Claude loses its premium. Anthropic cannot generate revenue sufficient to service $40B/year in cloud commitments. Renegotiation or default on the Google deal triggers a major revaluation of Alphabet's cloud backlog and a regulatory examination of whether a safety-focused AI company can survive without being de facto acquired by one of the hyperscalers.
Signal Anthropic's 2027 revenue projections are revised downward publicly; the next fundraising round prices below the current implied valuation
Musk complicates the picture
xAI-SpaceX's data on Anthropic's compute usage becomes a competitive intelligence asset. Musk resumes public attacks on Anthropic, this time with the credibility of a business partner who has seen inside the operation. Regulatory scrutiny on whether Anthropic's safety mission is compatible with being financially dependent on Musk infrastructure intensifies.
Signal Any public statement from Musk about Anthropic's internal operations or finances after a dispute
What Would Change This
If Google announces a new, genuinely competitive AI model before end of May -- Polymarket gives this 13.5% odds -- it changes the Anthropic relationship from 'Google's primary AI customer and strategic partner' to 'Google's customer and competitor.' The compute commitment would take on a different character if Google's own model is actively competing with Claude for the same enterprise contracts.
Related
The EU Is Forcing Google to Open Android to Rival AI. This Could Matter More Than Any AI Regulation.
powerGoogle Buys a Rival It Cannot Control
powerThe Government Is Now Testing AI Models Before They Launch. The Companies That Designed the Tests Are the Same Ones Being Tested.
powerThe Government Will Test Your AI Before You Get It. The Companies Volunteered.