Anthropic is reportedly the frontier AI lab behind Riot Platforms’ $9.1 billion, 20-year lease for 191 MW of data-center capacity in Rockdale, Texas. The agreement shows how power, land and long-term infrastructure access are becoming decisive in the global AI compute race.
Riot Platforms disclosed on August 10 that it had executed one of the largest long-term AI infrastructure agreements announced this year: a $9.1 billion, 20-year data-center lease covering 191 megawatts of critical IT capacity at its Rockdale campus in Texas.
Riot identified the customer only as a “leading frontier AI lab.” Bloomberg subsequently reported, citing people familiar with the matter, that the customer is Anthropic, the company behind Claude. Neither Riot nor Anthropic publicly confirmed the identification. The initial term runs through June 2048, while two five-year extension options could increase the potential contract value to approximately $16.1 billion.
Riot’s shares jumped approximately 25 percent in late trading after the customer was reported to be Anthropic. For a company long known primarily as a Bitcoin miner, the reaction made sense. For the wider AI industry, the agreement reinforced something larger: frontier laboratories are no longer securing only chips and cloud access. They are locking in data-center capacity built around power, land, cooling and physical infrastructure—sometimes for decades.
What the Deal Actually Includes
According to Riot’s regulatory filing, the agreement covers 191 megawatts of critical IT capacity at a build-to-suit Tier 3 data center on the company’s Rockdale campus.
Delivery will take place in phases. The first 96 megawatts are expected to become available in December 2027, while the remaining 95 megawatts are scheduled for delivery by June 2028. The initial lease term runs through June 2048.
Riot expects the agreement to generate approximately $9.1 billion in contract revenue over the initial term. The tenant holds two five-year extension options. If both are exercised, Riot estimates that the total potential contract value would increase to approximately $16.1 billion.
The company has also secured a $573 million interim financing facility from Morgan Stanley to fund initial development and procurement while an investment-grade credit backstop is finalised.
This is not a small side agreement. The 191-megawatt figure refers specifically to critical IT load—the power available to computing equipment—rather than the facility’s total electricity consumption. Cooling systems, electrical conversion and other supporting infrastructure will require additional power.
In AI terms, it represents a substantial block of capacity around which a frontier laboratory can plan training, inference and product expansion for much of the next two decades.
Riot Platforms Is Not a Traditional Cloud Provider
Riot did not begin as an AI infrastructure company. It spent years known primarily for Bitcoin mining. Like several other mining operators, it eventually recognised that the large power connections, industrial sites and energy-management capabilities developed for cryptocurrency mining could also serve the rapidly growing demand for AI and high-performance computing.
Rockdale offered several of the foundations required for such a transition: an existing industrial campus, substantial power infrastructure and an approved grid interconnection that Riot says can support the planned delivery schedule.
That background matters. Traditional hyperscalers continue to dominate the cloud-computing market, but the AI infrastructure buildout has grown large enough to create opportunities for specialised data-center developers, former cryptocurrency miners, neocloud providers and private infrastructure platforms.
Anthropic’s reported selection of Riot suggests that frontier laboratories are willing to look beyond conventional cloud providers when an operator can deliver the required combination of power, infrastructure, financing and deployment speed.
The agreement also fits a broader pattern. Bloomberg has reported that Anthropic recently entered into a roughly $10 billion compute arrangement with Volta Infra Holdings and agreed in May to purchase nearly $45 billion of computing capacity from Elon Musk’s xAI. Taken together, these reported commitments indicate the extraordinary scale at which Anthropic is attempting to secure infrastructure.
They also point to the operational pressure created by Claude’s expanding adoption across enterprise applications, coding tools and general-purpose AI usage. Long-term capacity agreements provide one way to reduce uncertainty around future availability.
Why Anthropic Needs This Much Capacity
Training frontier AI models requires enormous computing clusters. Serving those models across enterprise, coding and consumer workloads creates another layer of sustained demand, particularly as applications make greater use of long contexts, advanced reasoning and agentic execution.
The computational requirements of individual models may change as architectures and hardware become more efficient. But aggregate demand can continue rising when improved models attract more users, process larger workloads and become embedded across more business processes.
For frontier laboratories, infrastructure planning is therefore becoming a strategic discipline. Access to advanced accelerators, available grid capacity and suitably equipped data-center sites must often be negotiated years before the resulting computing capacity becomes operational.
A 20-year agreement gives the reported tenant a level of capacity predictability that shorter arrangements cannot provide. It also indicates that the company is planning beyond the next model release or product cycle.
There is a competitive dimension as well. OpenAI, Google, Microsoft, Meta, xAI and other major AI companies are seeking access to many of the same constrained resources. The companies that secure power, land, financing and data-center capacity earlier will be better positioned to scale when the next wave of demand arrives.
Anthropic’s reported agreement with Riot should be understood as one part of that broader contest.
The Market Reaction and What It Reveals
Riot’s stock-market reaction was immediate. Its shares jumped approximately 25 percent in late trading after Bloomberg identified the customer as Anthropic.
Investors appeared to view the agreement as potentially transformative for Riot’s business mix. Moving from predominantly Bitcoin-linked revenue toward long-term contracted data-center revenue could change both the company’s growth prospects and its risk profile—provided Riot completes construction on schedule and the agreement performs as expected.
Riot estimates that the lease could generate cumulative net operating income of between $7.3 billion and $8.2 billion over the initial term. Its investor presentation projects an NOI margin of approximately 80 to 90 percent. These are company forecasts rather than realised results, but they help explain the scale of the market response.
For the wider market, the agreement is another data point in a much larger infrastructure expansion. Capital expenditure by major technology companies is now measured in hundreds of billions of dollars annually, with a substantial share flowing into computing hardware, power systems, cooling infrastructure, networking and data-center construction.
Each agreement of this size reinforces the view that the AI infrastructure cycle continues to expand, even as questions about financing, utilisation and eventual returns remain unresolved.
What the Deal Says About AI Infrastructure
Look beyond the headline numbers and several broader trends become visible.
First, power has become a first-class strategic asset. AI companies can no longer assume that electricity will simply be available wherever they want to deploy computing capacity. Large blocks of power must increasingly be secured years in advance through cloud providers, data-center developers, site owners and energy partners.
Second, the supplier base is widening. Hyperscale cloud providers remain central to the industry, but they are no longer the only route to large-scale capacity. Specialised operators, former miners and newer infrastructure platforms are finding positions within the AI value chain.
That diversification may eventually expand supply. In the near term, however, it primarily demonstrates how urgently large AI companies are searching for deployable capacity.
Third, infrastructure-planning horizons are stretching. The Riot agreement shows that AI development is adopting timeframes more commonly associated with energy, utilities and commercial real estate than with conventional software-product cycles.
The capital intensity, grid requirements and construction timelines involved in frontier computing demand longer planning windows. Companies that can credibly commit capital across those windows may gain access to capacity that less well-funded competitors cannot secure.
There is also a geographic dimension. Texas continues to attract large data-center projects because of its industrial land, energy market, development ecosystem and availability of large-scale infrastructure sites. Rockdale is another example of that pattern, although the growing concentration of data centers also raises questions about grid capacity, local infrastructure and community impact.
Risks and Open Questions
No agreement of this size is risk-free.
Construction delays, supply-chain disruptions and technical challenges remain possible. Power-market conditions can change, financing arrangements may evolve and scrutiny of large data-center developments is increasing in several regions.
Riot must deliver the facilities according to schedule and provide the reliability required by a frontier AI laboratory. The tenant, reportedly Anthropic, must deploy enough productive computing equipment and workloads to justify a commitment extending through 2048.
There is also a broader question about utilisation. The industry is directing enormous amounts of capital into infrastructure based on expectations of continued AI demand growth. If that growth slows—or if improvements in chips, models and software significantly reduce the amount of computing required—some long-term commitments could appear expensive in hindsight.
The leading laboratories are currently betting that increased adoption will outweigh efficiency gains and that overall demand will continue rising fast enough to justify the investment.
Another open question concerns smaller AI companies, research organisations and open-source projects. When the largest laboratories secure power, accelerators and data-center capacity years in advance, the remaining supply can become harder and more expensive for smaller participants to access.
That dynamic has been visible for some time. Agreements of this magnitude make the emerging divide more concrete.
Looking Ahead
Riot’s agreement with an unnamed frontier AI laboratory—and Bloomberg’s identification of that laboratory as Anthropic—is not an isolated development. It represents another move in a high-stakes contest over the physical foundations of artificial intelligence.
Chips matter. Models and algorithms matter. But none of them can operate at scale without electricity, cooling, networking and physical space delivered on a predictable schedule.
The $9.1 billion figure attracts attention. The timeline may ultimately prove more consequential. Frontier AI companies are no longer planning infrastructure only around product cycles measured in months. They are making commitments that extend across decades.
That shift will influence how capital is allocated, where data centers are built, which infrastructure operators gain strategic importance and how competitive advantage is established.
On current reporting, Anthropic appears to have secured a substantial block of future computing capacity in Texas. Riot has gained a flagship frontier-lab tenant capable of transforming its business story. The rest of the industry has received another reminder that the race for compute is still accelerating—even as the financial and physical cost of competing continues to climb.