Anthropic expects to pour at least $518 billion into cloud services, data centers and specialized equipment over the next decade. The figure, drawn from a confidential IPO prospectus reviewed by Reuters, stands among the largest capital commitments in artificial intelligence history. Roughly 80% of that total comes with strings attached. Most of the deals are non-cancelable or require payment regardless of actual usage.
The numbers tell a story of scarcity. Compute capacity, not customer demand, will limit how far and how fast AI models can advance. Anthropic made that assessment plain in its filing. Future systems will be “limited principally by the availability of compute.” So the company is securing access now. Even if it means writing enormous checks years before the hardware runs at full throttle.
Break it down. At least $111.1 billion heads to Google’s cloud between April 2026 and July 2033. Another $110 billion goes to Amazon over a similar stretch ending in April 2036. Microsoft gets $31.4 billion. Add $161.2 billion in largely non-cancelable equipment leases tied to Broadcom. The sums add up fast. And the contracts carry bite. “If our actual spend falls short, we must pay Google the difference,” the prospectus states. Similar language applies to the Amazon agreement.
Other partners fill out the roster. Arrangements with Elon Musk’s xAI could reach $84.5 billion for Nvidia-based capacity through 2029, though most of those can be canceled with 90 days’ notice. AMD supplies more than $20 billion in computing resources and agreed to buy up to $5 billion of Anthropic stock. Micron, Samsung and SK hynix round out memory partnerships needed to feed the accelerators.
Last year offered a preview. Anthropic generated nearly $4.6 billion in revenue in 2025, a twelvefold jump from the prior period. Yet operating losses topped $8 billion. The company spent $7.33 billion on compute and infrastructure alone, more than triple the amount from 2024. That spending represented well over half of total operating expenses. Put another way, Anthropic burned through $1.59 on compute for every dollar of revenue.
Such figures raise eyebrows. They also explain the urgency behind the long-term deals. Cloud providers and chip suppliers face their own constraints. Power availability, chip fabrication capacity and data center construction all move slower than AI ambition. By signing ironclad contracts, Anthropic gains priority. The hyperscalers gain demand certainty for their own massive buildouts.
Yet the arrangement creates a tangled web. Amazon and Google rank among Anthropic’s largest investors. They also serve as critical cloud suppliers and distribution channels. And they develop competing AI models. The prospectus flags the conflict. Reliance on these three firms, which control both infrastructure and rival AI efforts, carries strategic risk. Sales routed through their platforms accounted for 47% of Anthropic’s 2025 revenue, up sharply from earlier years.
Anthropic isn’t standing still. The company is shifting away from a pure cloud model. It now pursues dedicated data centers and directly leased chips. Deals with Fluidstack for facilities in Texas and New York, plus capacity at SpaceX’s Colossus 1 cluster in Memphis, signal the move. One recent report placed the total value of such specialized agreements even higher in some tallies. But the core $518 billion figure from the filing anchors the discussion.
Investors appear undeterred. Anthropic seeks a valuation near $2 trillion in its planned public debut. That price tag implies enormous expectations for Claude model adoption and future revenue. It also assumes the massive infrastructure spend will translate into defensible leadership. History offers mixed lessons. Previous technology waves rewarded those who controlled scarce resources early. This time the scarce resource is electricity and silicon running at unprecedented scale.
The bets extend beyond training. Inference, the process of running trained models for users, demands its own capacity. As Claude handles more complex tasks and reaches more customers, inference compute could eclipse training needs. The prospectus hints at that shift. Long-term obligations stretch into the 2030s precisely because demand forecasts point higher, not lower.
Amazon stands to benefit handsomely. The $110 billion commitment, part of an earlier announced deal exceeding $100 billion over ten years, provides AWS with visibility into years of revenue. It also covers access to as much as 5 gigawatts of capacity, including Amazon’s custom Trainium chips. For context, that single customer commitment represents a meaningful slice of AWS’s existing backlog. The Motley Fool noted that annualizing the spend at roughly $10 billion per year would equal about 7% of AWS’s recent trailing sales.
Google and Microsoft gain similar assurances. Their own AI infrastructure investments receive validation from a premier customer. Broadcom secures equipment orders that support its AI networking and custom silicon business. The entire supply chain, from memory makers to power utilities, feels the ripple.
Risks remain. Many of Anthropic’s largest customers lack long-term contracts and could curtail spending. Power shortages or construction delays could prevent full utilization of reserved capacity. And the non-cancelable nature of the deals means fixed costs even if adoption slows. The prospectus devotes considerable space to these warnings. It also acknowledges that advanced AI systems could pose existential risks to humanity, a consistent theme from Anthropic’s leadership.
Still, the company frames its approach as necessary realism. Compute scarcity isn’t a temporary bottleneck. It’s the defining feature of the next decade of AI progress. By locking in capacity today, Anthropic aims to avoid being squeezed out tomorrow. Competitors such as OpenAI pursue parallel strategies, with their own multibillion-dollar infrastructure pacts. The industry is engaged in a race not just for better models but for the raw resources to run them.
Recent coverage reinforces the scale. A Seeking Alpha analysis from September 30 argued that Anthropic’s AWS commitment, combined with OpenAI’s deals, alters the capex debate for Amazon by reducing utilization risk and supporting sustained heavy investment. Other reports detail additional partnerships pushing total exposure even higher in certain estimates, though the IPO filing’s $518 billion figure serves as the authoritative baseline.
What comes next? Anthropic’s public listing, whenever it arrives, will invite closer scrutiny of these obligations. Markets will debate whether the spending represents prudent hedging or an outsized wager on unchecked AI growth. For now the company has placed its chips. Hundreds of billions of dollars in binding agreements say it believes the table is still worth playing.
And the power meters keep spinning.
Anthropic’s $518 Billion Bet: Locking In Compute as AI Hits the Power Wall first appeared on Web and IT News.