Anthropic is paying xAI $1.25 billion per month through May 2029 for the entire Colossus 1 data center — $15 billion annually — according to SpaceX’s S-1 filing. The same week, reports emerged that Anthropic is in early talks with Microsoft to use custom Maia 200 chips. Both stories surface something frontier AI labs rarely disclose: compute supply is their most urgent operational constraint, and it flows directly into your contracts.

Key Takeaways

• Anthropic committed $1.25B/month to xAI/SpaceX through May 2029; either party can exit with 90 days’ notice.
• The deal covers all capacity at Colossus 1 — over 300 MW and 220,000 Nvidia GPUs — and immediately raised Claude rate limits.
• Microsoft Maia 200 chip talks are early-stage only; no deal exists yet.
• Compute obligations at frontier labs set the floor for enterprise API pricing and availability.

What Happened?

In early May, Anthropic and xAI announced a partnership in which Anthropic secured all compute capacity at xAI’s Colossus 1 data center in Memphis — over 300 megawatts and 220,000+ Nvidia GPUs. Anthropic immediately raised Claude Code rate limits and Claude Opus API ceilings for enterprise customers. SpaceX’s S-1 filing, submitted May 20, disclosed the contract terms: $1.25 billion per month with a 90-day mutual termination clause. xAI had the capacity to sell because Grok usage declined and xAI had shifted model training to its newer Colossus 2 facility.

Two days later, CNBC reported Anthropic is in early-stage talks to use Microsoft’s custom Maia 200 inference chip — a separate arrangement inside an existing relationship where Microsoft invested $5 billion in Anthropic and Anthropic committed to $30 billion in Azure spending. Those talks are not a deal.

Why Should Operators Care?

The $1.25 billion per month figure matters because compute obligations set the cost floor for every API price and enterprise contract downstream.

Supply-chain risk is live. The 90-day mutual termination clause means Anthropic could lose access to 220,000 GPUs within one contract-renewal cycle. Enterprise workflows running high-volume Claude API calls should have a contingency routing path.

Pricing pressure is structural. Frontier labs carrying heavy, long-dated compute contracts have less room to hold pricing steady. Tier restructuring or usage-cap adjustments are plausible before 2027.

Vendor entanglement is deep. Anthropic now carries financial or compute relationships with Amazon, Google, Microsoft, Nvidia, and xAI simultaneously. As we covered in Google’s $40B Anthropic investment, each adds capacity — and each creates potential conflicts around feature priority, model access, and platform routing.

For operators running Claude in production, the posture is ask sharper vendor questions before your next renewal:

• Does your SLA address rate-limit reductions caused by compute-supply changes, not just outages?
• What is your fallback routing if Claude Opus API capacity tightens?
• Who are your AI vendor’s compute suppliers, and what are their termination terms?

If the Microsoft Maia talks close, Anthropic gains more silicon diversity and potentially lower per-token costs — good for buyers. But talks are not a deal. The 90-day xAI clause is live today. Watch for API pricing or tier changes through Q3 2026, Maia deal confirmation or collapse, and infrastructure-dependency disclosures as OpenAI prepares its own IPO filing.

---

FAQ

Why is Anthropic buying compute from xAI if they are competitors? Compute scarcity overrides competitive logic. xAI had surplus Colossus 1 capacity available after Grok usage declined, and Anthropic needed more infrastructure immediately. Paying a competitor for idle infrastructure is unusual but practical when traditional cloud providers cannot fill the gap fast enough.

Will Anthropic’s $15B/year compute costs raise enterprise API prices? No pricing change has been announced. But the scale of these commitments creates a high cost floor. Enterprise buyers should monitor for tier restructuring or usage-cap changes at the next renewal — and ask vendors directly about pricing stability over a 12-to-24-month horizon.

What is the practical risk of the 90-day termination clause? If either party exits, Anthropic loses access to 220,000 GPUs within roughly one contract-renewal cycle. High-volume Claude API users should have alternative routing paths or contract language that covers compute-related service disruptions.