Summary
- Anthropic, Macquarie Asset Management, and GIC have established Theseus Infrastructure.
- Macquarie-managed funds and GIC will own the platform and provide most of the equity for individual projects.
- Anthropic will act as anchor tenant under long-term agreements, with initial site development focused on the US.
Anthropic, Macquarie Asset Management, and Singapore sovereign wealth fund GIC have established Theseus Infrastructure, a dedicated platform designed to develop, operate, and lease purpose-built data centres to the AI company under long-term agreements.
The platform will identify and develop sites where Anthropic is expected to act as anchor tenant. Funds managed by Macquarie Asset Management, together with GIC, will own Theseus and provide the majority of the equity required for each project.
The initial development focus is the United States, so Theseus is not currently a European capacity announcement. Its relevance lies instead in the ownership and financing structure being created around large-scale AI infrastructure.
Rather than relying solely on conventional cloud procurement or leasing individual facilities from unrelated operators, Anthropic is pairing long-duration compute demand with an infrastructure platform backed by institutional capital.
The arrangement gives developers and investors greater visibility over the eventual customer before construction starts. An anchor tenant prepared to enter long-term agreements can support project financing because lenders and equity investors have a clearer revenue counterparty.
Macquarie’s announcement also says Anthropic will cover increases in consumer electricity prices attributable to the sites, adding an explicit power-cost commitment to the infrastructure model.
AI tenants move closer to asset ownership
The Theseus structure is another sign that the boundary between AI company, data centre tenant, infrastructure investor, and property developer is becoming less distinct.
Traditional hyperscale capacity can be delivered through a mixture of owned facilities, wholesale colocation, and cloud infrastructure. The growth of very large AI clusters adds an incentive to secure dedicated capacity further in advance because power, cooling, and land may take years to deliver.
An infrastructure platform built around one anchor tenant can improve certainty, but it also concentrates risk. The economic case for each facility depends heavily on the tenant’s long-term requirement for compute and its ability to honour commitments over the financing period.
That makes technical flexibility important. Data centre buildings and power systems usually have longer useful lives than a generation of AI hardware. Facilities commissioned for one computing architecture need enough electrical, cooling, and structural adaptability to remain useful as rack densities and accelerator designs change.
The Macquarie and GIC involvement adds infrastructure-finance discipline to that equation. Their return depends not merely on Anthropic continuing to buy GPUs, but on sites moving through planning, power procurement, construction, and commissioning without cost or schedule overruns that undermine the economics of long-term leases.
The initial US focus reflects the sheer scale of AI infrastructure being developed there, but the financing model has obvious relevance for Europe. European data centre projects increasingly face a shortage of sites with credible grid access, while institutional investors are seeking long-duration digital-infrastructure assets.
If similar anchor-tenant platforms are deployed in Europe, they could provide developers with stronger commercial backing when competing for scarce power and land. They would not remove grid or planning constraints, but they could reduce uncertainty around who will occupy expensive capacity once it is built.
Theseus therefore represents a structural development more than a conventional site announcement. Anthropic is placing institutional capital closer to its physical compute requirements, while Macquarie and GIC are taking a direct role in translating long-term AI demand into buildings, substations, cooling systems, and operating data centres.
