Anthropic Data Center Expansion: Potential Monthly Revenue Exceeding $1.3 Billion
nashnova research
A single 330 MW Anthropic data center could generate over $1.3 billion in monthly revenue against roughly $200 million in chip-lease costs — as the AI company races to scale compute ahead of an imminent IPO, the market's central question is whether that unit economics holds at full build-out.
How much can one data center earn per month?
Anthropic's facility in St. Francisville, Louisiana starts at 330 MW, developed by Hut 8 and operated by Fluidstack, with Anthropic supplying Google AI server chips. Target go-live: Q2 2027, with a long-term expansion path to 1 GW.
Occam Edge analyst Christian Okoye estimates the site's monthly revenue potential at over $1.3 billion, based on publicly disclosed rates in recent long-term AI contracts. Anthropic's monthly chip-lease cost for the same capacity: roughly $200 million.
This means → gross margin on a single facility could exceed $1.1 billion per month — provided all capacity serves inference requests rather than model training.
In plain terms = for every $1 spent leasing chips, the facility could bring back about $6.50 in revenue. The spread is enormous.
From 1.4 GW to 15 GW — how fast is the ramp?
AI construction tracker Measured AI projects Anthropic's compute capacity at 1.4 GW in 2025 → 5 GW in 2026 → 9.5 GW in 2027 → over 15 GW by 2028.
For context, Microsoft Azure's entire cloud operation ran roughly 5 GW at the end of 2023 — before the current AI buildout wave. Anthropic plans to match that figure by 2026.
This means → a pre-IPO AI company is on track to rival or surpass the compute footprint a hyperscaler built over more than a decade.
Where does the money — and the power — come from?
Anthropic has signed lease agreements totaling $530 billion with AWS, Google, SpaceX's AI division, and over ten other cloud and data-center operators.
Several partners are former Bitcoin miners — Hut 8, TeraWulf, IREN, Nscale. Their key asset: grid-interconnection rights secured at decommissioned aluminum smelters, failed green-hydrogen sites, and renewable-energy land parcels.
In plain terms = remote power resources that hyperscalers once passed on are now scarce, high-value assets in the AI compute race — and former crypto miners happen to be sitting on them.
What undisclosed deals are in the pipeline?
Sources say Anthropic has signed an as-yet-undisclosed multi-year cloud-services agreement with listed company IREN.
Separately, Anthropic is in talks with Apollo-owned Stream Data Centers to lease up to 1 GW of future capacity. Stream recently acquired two failed hydrogen-project sites from Plug Power, gaining attractive grid-access terms.
This reflects a procurement strategy that goes well beyond hyperscaler partnerships — Anthropic is actively recruiting non-traditional players that have power but lack compute.
Why is "one month earlier" worth $1.3 billion?
With chip supply tight, go-live speed directly determines the revenue window — every month of delay is a month of inference revenue lost.
Vertiv CEO Giordano Albertazzi, speaking at the Hut 8 site event, said the company is pre-assembling power-distribution modules, batteries, switchboards, cooling, and fire-suppression units at the factory, then shipping them as complete packages. This can cut typical installation time in half.
In plain terms = the industry is shifting from building data centers brick by brick on-site to snapping together factory-finished modules — the same logic as prefab construction.
With the IPO imminent, what is the market watching?
Anthropic's IPO is expected to launch next month. Whether the economics of its compute expansion hold up in audited financials will be the central node for valuation.
This means → investors care less about how much compute Anthropic has and more about whether it can actually fill that capacity with paying inference workloads — turning "over $1.3 billion per month" from a paper estimate into real profit.
Put simply = no matter how large the footprint, if inference demand doesn't keep pace, the margin story collapses.
市场有风险，内容仅供研究参考，不构成投资建议。