Bloomberg reports that Anthropic, Macquarie Asset Management, and GIC have created a platform to develop, own, and lease US data centers — no size disclosed, no sites named, but the structure tells the real story.
According to Bloomberg, Anthropic, Macquarie Asset Management, and Singapore’s sovereign wealth fund GIC have formed a new entity called Theseus Infrastructure — a platform whose purpose is to identify, develop, operate, and lease data centers to Anthropic under long-term agreements. Macquarie’s infrastructure funds and GIC will own the platform and supply the majority of the equity for each project. Anthropic’s role is that of anchor tenant, not owner. No spending figure was disclosed for this arrangement; a $50 billion figure sometimes attached to Anthropic’s data-center ambitions is a separate 2025 statement and should not be read as the scope of Theseus.
The initial focus is the United States. No specific sites have been announced, and no projects have been named — Theseus is at this stage a platform and an intent: a mechanism for sourcing and building locations rather than a portfolio of built assets. The three parties will jointly identify new US sites, with Anthropic committing to occupying them as anchor tenant under long-term lease agreements that make each project financeable to outside capital.
Alongside the structural arrangement, Anthropic made two operational commitments: it will cover 100 percent of grid-upgrade costs associated with its data centers, and it will cover any consumer electricity price increases that its data-center demand causes. The scope and mechanism of the consumer-electricity pledge is not detailed in the announcement, but its presence in the deal framework is notable in its own right.
The key insight: Anthropic is not raising money to own data centers. It is arranging for infrastructure and sovereign capital to own them, and leasing the capacity back — keeping enormous capex off its own balance sheet while taking on long-term lease obligations that make the whole structure financeable. The mechanism, not any missing size figure, is the news.
ANTHROPIC’S COMPUTE-FINANCING ARC
2024–2025
Apollo and Blackstone structure a compute securitization vehicle — chips leased to Anthropic; capex owned off-balance-sheet by a special-purpose vehicle. Anthropic is the anchor counterparty.
2025
Anthropic articulates a broad US data-center ambition (the $50B figure cited in public reporting). That statement is separate from any specific structured vehicle.
August 10, 2026
Theseus Infrastructure announced — Macquarie + GIC own the platform; Anthropic is anchor tenant. Infrastructure and sovereign capital now own the buildings; Anthropic leases them. The compute-financing arc moves from chips to data centers.
OPEN
Sites unnamed. Projects unannounced. Equity per project funded by Macquarie funds + GIC. Platform exists; the buildout pace and scale remain to be established.
The Structural Read
The anchor-tenant-plus-infrastructure-capital model is not a new financial invention. It is how toll roads, airports, and utility pipelines have been financed for decades: patient capital owns the asset; an operator or offtaker with a long-term contractual obligation makes the asset financeable. The news is not that the structure is exotic. The news is that it is now being applied to AI compute — and that Anthropic is, across multiple sequential deals, the entity making it work.
The through-line is legible. First, securitized chips: Apollo and Blackstone structured a compute SPV that owned hardware leased to Anthropic, keeping that capex off Anthropic’s books. Now, leased data centers: Macquarie and GIC own the buildings, and Anthropic occupies them under long-term agreements. In both cases Anthropic is the anchor counterparty — the entity whose creditworthiness and long-term lease commitment is the thing that converts a speculative AI asset into something infrastructure capital will fund. The pattern is consistent: Anthropic’s capex steadily migrates off its own balance sheet and into structures owned by others.
That is rational for a lab that would rather spend its capital on models than on concrete, and it is rational for Macquarie — which funds long-duration contracted assets — and for GIC, a sovereign fund whose return profile favors stable, contracted cash flows. But the elegance of the structure should not obscure what it actually does. Moving data centers off Anthropic’s balance sheet is risk transfer, not risk elimination. Anthropic swaps ownership for long-term lease obligations — those leases are a liability, and the structure only works as long as those lease payments are money-good. The “safe,” bond-like return that attracts a sovereign wealth fund rests entirely on one frontier lab’s economics.
Business Engineer — Structural Framework
“Dressing AI-compute exposure as infrastructure does not remove the AI risk — it relocates it. The anchor-tenant model de-risks the builder and concentrates the counterparty risk on the lab. If the lab’s economics wobble, the infrastructure asset built around it is exposed to precisely that. The ‘safe’ asset and the frontier-AI bet are the same bet, layered.”
The grid and consumer-electricity pledge is the quiet tell that the binding constraint on AI infrastructure has become political as much as physical. The technical problem — finding land with power — is real, as the Nvidia-Lancium dynamic illustrates. But the political problem is adjacent and increasingly primary: AI data centers raise local electricity prices, and that generates backlash. Committing to cover those consumer cost increases is how you defuse the objection before it becomes a permitting obstacle. It is a structural concession built into the deal design, not an afterthought.
IMPLICATION 1 — AI DATA CENTERS ARE BECOMING AN INFRASTRUCTURE ASSET CLASS
When Macquarie — the firm that finances toll roads and airports — and GIC structure a platform to own AI compute facilities, it signals that this asset type has cleared the underwriting bar for long-duration infrastructure capital. That is a durable shift in how the AI buildout is financed, independent of any single deal’s size. The template, once proven, will be reused. Other labs and other infrastructure managers will watch this structure closely.
IMPLICATION 2 — ANCHOR-TENANT CONCENTRATION IS THE HIDDEN RISK
Every infrastructure vehicle that Anthropic anchors is also a concentrated bet on Anthropic’s long-term viability. The sovereign and infrastructure capital entering these structures is not diversified across multiple frontier labs — it is contracted to one. If the economics of that lab shift materially (through model commoditization, competitive displacement, or a demand correction), the “infrastructure” asset is exposed. This is the risk vector worth watching as AI financial structures multiply. The safe-looking asset and the AI bet are the same underlying exposure.
IMPLICATION 3 — THE POWER PLEDGE SIGNALS A NEW PERMITTING STRATEGY
Anthropic’s commitment to cover consumer electricity price increases tied to its data-center demand is not standard in infrastructure deals. It is a political instrument — a pre-emptive concession designed to remove a community-opposition argument before it reaches regulators and planning boards. As AI data centers compete for power in constrained US grids, the political economy of siting is becoming a real constraint. Expect this kind of grid-and-consumer pledge to become a standard term in future AI infrastructure deals, not an exceptional one.
Business Engineer Framework
The Map of AI — Where Theseus Sits in the Stack
Theseus Infrastructure operates at the physical infrastructure layer of the AI stack — the compute and power substrate that every layer above it depends on. The Map of AI traces how value and risk distribute across nine layers, from raw compute to frontier models to applications. Understanding which layer a deal like Theseus occupies — and how the financing structure at that layer affects labs, hyperscalers, and sovereign capital above and below it — is the analytical lens that makes the Anthropic-as-through-line pattern legible.
Theseus Infrastructure is, at this moment, a platform and an intent — no sites named, no projects announced, no size disclosed, and one sovereign fund plus one infrastructure manager making a single contracted, hedged bet. What survives those hedges is the direction: the AI buildout is being financed less and less on labs’ own balance sheets and more and more through infrastructure and sovereign capital that treats data centers as long-duration contracted assets, with the labs as anchor tenants. That is how a technology boom becomes an infrastructure cycle — and it is also how the risk of that boom gets distributed into places that are supposed to be safe. Anthropic is the entity that connects every link in that chain, which makes its long-term economics the load-bearing variable in structures that will be marketed as something more stable than a frontier AI bet
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
