Riot Platforms closed a $9.1 billion compute supply agreement with Anthropic, the foundation model developer backed by Amazon and Google. The contract runs through 2030 and represents the largest known private infrastructure commitment in the AI training market. Riot, which reported $336 million in revenue last year primarily from Bitcoin mining, will dedicate existing and planned data center capacity in Texas to Anthropic's model development pipeline.
The deal follows eighteen months of quiet repositioning. Riot began converting mining facilities in Rockdale, Texas in Q2 2023, adding high-density cooling and power distribution capable of supporting H100 and successor architectures. The company held 1,889 megawatts of contracted power capacity as of December 2024, with 487 MW already converted to AI-ready infrastructure. Anthropic's commitment covers the full build-out of an additional 850 MW across two phases, the first delivering in Q4 2025. Riot will finance the expansion through a $2.3 billion credit facility with KeyBanc and Morgan Stanley, secured against the Anthropic contract.
The market implication is structural, not speculative. Foundation model developers are locking long-term capacity at rates that assume sustained training intensity through the end of the decade. Anthropic's willingness to commit $1.5 billion annually for six years signals confidence that scaling laws remain intact and that proprietary infrastructure insulates against hyperscaler pricing volatility. The deal also validates the thesis that former mining operators hold an edge in power procurement and thermal management, competencies that matter more than legacy cloud expertise when deploying 400-watt accelerators at scale. Riot's all-in cost per kilowatt-hour in Texas runs $0.031, roughly half the rate available through AWS or Azure reserved instances in the same geography.
Allocators should note that this is the second $5 billion-plus private compute contract announced in thirty days. CoreWeave signed a $6.7 billion deal with Microsoft in late March. The pattern suggests that hyperscalers are de-risking training commitments by outsourcing fixed capacity to specialized operators, preserving balance sheet flexibility while securing inference and fine-tuning workloads in-house. If the trend holds, operators with contracted power above 200 MW and thermal infrastructure rated for next-generation chips will see valuation multiples converge with cloud peers by mid-2026.
Watch for Riot's Q2 earnings in August, which will disclose phase-one buildout progress and any early revenue recognition under the Anthropic contract. Separately, monitor whether Anthropic files for additional equity or debt financing before year-end. A $9.1 billion infrastructure commitment implies a training budget that exceeds the company's last disclosed capital raise by a factor of three.
The deal settles a question that mattered six months ago: whether the AI infrastructure build is speculative or contracted. Riot's pivot is now a $9.1 billion fact.