Anthropic has entered a $9 billion cloud computing agreement with Riot Platforms, marking a significant expansion in its infrastructure capabilities. The deal highlights the intensifying competition for high-performance computing resources as artificial intelligence models require increasingly powerful hardware to train and operate at scale.

The partnership positions Riot Platforms, traditionally known for cryptocurrency mining, as a key provider of cloud infrastructure for AI development. This shift reflects broader industry trends where data center operators are repurposing or upgrading facilities to meet the specific power and cooling demands of AI workloads.

Deal Details

The financial scale of the agreement indicates a long-term commitment from Anthropic to secure dedicated compute capacity. Such large-scale contracts are critical for AI companies aiming to maintain competitive advantages in model training and inference speeds.

What the Numbers Show

The $9 billion valuation of this single contract suggests that infrastructure costs are becoming a primary differentiator in the AI sector. By locking in capacity with Riot Platforms, Anthropic mitigates the risk of resource scarcity while enabling accelerated development cycles for its models.

Anthropic, Macquarie Asset Management, and GIC today announced a strategic partnership to establish Theseus Infrastructure, a new platform designed to develop, operate, and lease data center infrastructure at scale. This collaboration aims to meet the surging demand for Anthropic’s Claude AI model by creating purpose-built facilities that serve as long-term anchors for the company’s expanding capacity requirements. The initiative represents a significant commitment to scaling AI compute infrastructure, with an initial focus on identifying and developing new sites across the United States.

The structure of the partnership places Anthropic as the anchor tenant for each facility, securing its operational needs through long-term agreements. Funds managed by Macquarie Asset Management, in conjunction with GIC, will own the Theseus Infrastructure platform and provide the majority of the equity funding for each project. This arrangement allows Anthropic to secure critical infrastructure without bearing the full capital expenditure burden directly, while providing Macquarie and GIC with stable, long-term yield from a high-growth tenant.

Partnership Structure and Operational Focus

The collaboration draws on the distinct strengths of each partner: Macquarie Asset Management brings global expertise in developing, financing, and operating large-scale digital infrastructure, while GIC contributes deep experience in global infrastructure investing. The planned developments are expected to require significant capital investment and will generate thousands of construction jobs and permanent operational roles in the communities where the sites are located.

A notable component of this agreement involves energy cost management. Anthropic has committed to covering electricity price increases that consumers might otherwise face due to the power demands of these new data centers. This commitment aligns with broader pledges the company announced earlier this year regarding its environmental and community impact, addressing potential concerns about the strain large-scale AI infrastructure can place on local power grids and consumer rates.

What the Numbers Show

While specific financial figures for the initial equity injection or total project valuation were not disclosed in the announcement, the strategic alignment highlights a shift towards specialized, tenant-specific infrastructure models in the AI sector. By partnering with established asset managers like Macquarie and sovereign wealth fund GIC, Anthropic is effectively outsourcing the balance sheet risk associated with heavy capital expenditures while ensuring supply chain security for its compute resources. This model contrasts with traditional build-to-suit arrangements by creating a dedicated platform entity, Theseus Infrastructure, suggesting a long-term, scalable approach rather than ad-hoc facility acquisitions. The focus on the United States initially indicates where immediate capacity gaps are most acute, likely driven by regulatory environments and existing power grid capacities suitable for high-density computing loads.