Anthropic is taking another step to secure the physical infrastructure it needs for Claude. The artificial intelligence company has partnered with GIC and Macquarie Asset Management to create Theseus Infrastructure, a new platform that will develop, operate, and lease data centers built specifically for Anthropic’s needs, initially in the United States.
Key facts about Theseus Infrastructure in 20 seconds
- Anthropic, GIC, and Macquarie Asset Management have created Theseus Infrastructure.
- GIC and funds managed by Macquarie will own it and provide the majority of project capital.
- Anthropic will act as the primary client through long-term contracts.
- The initial centers will be developed in the United States.
- The deal expands Anthropic’s plan to mobilize tens of billions of dollars in AI infrastructure.
The model is especially interesting because Anthropic doesn’t need to become directly a real estate owner of all the data centers it uses. GIC and Macquarie contribute capital and infrastructure expertise, while the company behind Claude gains facilities tailored to its workloads through long-term contracts.
No locations, power capacity, number of campuses, construction schedule, or financial volume for Theseus have been announced yet. Nor should the creation of the platform be confused with the $50 billion investment in U.S. infrastructure that Anthropic announced in November 2025, though both initiatives are part of its rapid capacity expansion. Anthropic said it was working with Fluidstack on data centers in Texas and New York, along with future locations.
Anthropic aims to secure capacity before it’s needed
Creating Theseus helps explain how competition among major AI labs is changing.
In the early years of generative AI, much of the conversation focused on acquiring enough GPUs. Now, the challenge is much broader: it requires chips, but also megawatts, land, substations, cooling, fiber, funding, and high-density rack-capable centers.
Anthropic is already operating with a deliberately diversified infrastructure strategy. The company uses Nvidia GPUs, Google TPUs, and Amazon’s Trainium accelerators, rather than relying on just one architecture.
In April, it extended its agreement with Amazon to secure up to 5 GW of new capacity, with a commitment of over $100 billion over ten years in AWS technologies. Anthropic reports it already uses more than one million Trainium chips to train and run Claude.
That same month, it announced another deal with Google and Broadcom to access several gigawatts of next-generation TPU capacity starting in 2027. Most of this new infrastructure will be located in the U.S.
Additionally, the May agreement with SpaceX to use all capacity at their Colossus 1 facility adds over 300 MW and 220,000 Nvidia GPUs.
Theseus adds a different piece to this puzzle.
Instead of only contracting for existing computing capacity from a cloud provider or specialist, Anthropic can participate much earlier in the development of centers designed around its specific needs.
GIC and Macquarie bring something an AI company doesn’t
Building a large AI data center is a very different business from developing foundational models.
Projects can require billions of dollars before generating revenue, along with years of planning, permits, energy agreements, and construction. For Anthropic, it makes sense to transfer part of that risk and capital requirement to specialized investors.
That’s where GIC and Macquarie come in.
GIC is the sovereign wealth fund managing Singapore’s reserves, with investments in over 40 countries. Its relationship with Anthropic predates Theseus: in February 2026, it led, alongside Coatue, a $30 billion funding round valuing the company at $380 billion at the time.
Macquarie brings decades of experience in physical assets and digital infrastructure. A recent example is Aligned Data Centers. Macquarie Asset Management invested in the company from 2018, during which time the platform grew from two facilities and 85 MW to 51 campuses with over 6.4 GW of operational and planned capacity, before being sold in July 2026.
Theseus’s structure takes advantage of these capabilities.
GIC and funds managed by Macquarie will own the new platform and finance most of the capital needed for each project. Anthropic also provides something equally valuable from a financial standpoint: committed demand through long-term lease agreements.
The model is similar to the build-to-suit approach used for years in digital infrastructure, but scaled according to AI’s specific needs.
For investors, having an anchor tenant can offer greater visibility into future revenues. For Anthropic, the advantage is being able to reserve capacity designed specifically for its workloads without necessarily locking in all the capital required to own each campus directly.
Gigawatts are becoming a competitive advantage
The size of these deals might seem excessive until you look at Anthropic’s own forecasts.
The company estimates that training a single advanced model could require data centers of 2 GW in 2027 and 5 GW in 2028. It also calculates that the entire U.S. AI sector might need at least 50 GW of electrical capacity for training and inference. These are company estimates, not independent forecasts, but they help explain its moves.
The bottleneck is no longer just about manufacturing enough accelerators.
A GPU without sufficient power, cooling, or space doesn’t provide computational capacity. Similarly, buying thousands of accelerators is pointless if the electrical grid takes years to bring in the hundreds of megawatts required.
Macquarie notes that 92% of data center capacity currently under construction in North America is already pre-contracted, according to their investment analysis sources. The firm also expects data center electricity demand to grow at a compounded annual rate of 14.2% through 2035.
This makes early infrastructure planning a strategic issue.
Anthropic is pursuing multiple approaches simultaneously: AWS and Trainium, Google and its TPUs, Nvidia GPUs, partnerships with specialized operators, and now an institutional-backed platform to develop new facilities.
This reduces dependence on a single technology provider and broadens the avenues to secure computing power.
Another important element: Theseus partially separates data center funding from Anthropic’s accounts.
AI models require increasing investments before generating expected returns. Incorporating infrastructure funds distributes that effort among actors with different financial profiles. Anthropic can focus more resources on models, products, and chips, while specialized investors fund physical assets backed by long-term contracts.
It doesn’t eliminate risk. If future computing needs are lower than expected, capacity commitments could become a burden. There are also outstanding issues related to energy, land availability, grid connections, and construction timelines.
But the creation of Theseus demonstrates how the AI race has shifted from solely a model competition to a broader infrastructure race.
Claude needs software and algorithms, but behind each response is an ever-growing physical infrastructure. Anthropic is trying to secure that infrastructure years before it’s needed and is involving some of the world’s largest investors in physical assets in the process.
Frequently Asked Questions
What is Theseus Infrastructure?
Theseus Infrastructure is a new platform created through collaboration between Anthropic, GIC, and Macquarie Asset Management to develop and operate data centers for Anthropic’s computing capacity needs.
Who will own Theseus Infrastructure?
GIC and funds managed by Macquarie will own the platform and will contribute most of the project capital. Anthropic will serve as the anchor tenant through long-term contracts.
Where will Theseus data centers be located?
The initial phase will focus on the United States. Specific locations, number of facilities, and capacity details have not yet been announced.
Will Anthropic only build centers with Theseus?
No. Theseus is part of a broader strategy that includes capacity agreements with Amazon, Google, SpaceX, Fluidstack, and other providers and infrastructure initiatives.
