Fluidstack reliant on Anthropic and power

Fluidstack is no longer selling generic GPU capacity, it is effectively underwriting a small number of giant US AI buildouts whose economics depend on policy, power, and a few anchor buyers staying intact. After pulling back from its French megaproject, the companys pipeline became much more tied to US sites and to Anthropic, whose own compute strategy now spans AWS, Azure, Google, SpaceX, and Fluidstack.1 2 3 4

• The customer concentration is unusually high for a company of Fluidstacks size. Its estimated revenue was $180M at the end of 2024, while the Anthropic partnership alone is framed around a $50B US infrastructure buildout in Texas and New York, meaning one relationship can dominate future utilization.1 2 3
• Power and permitting matter because GPU clouds only make money when chips are installed, energized, and kept full. Fluidstack shifted from brokering spare GPUs to standing up dedicated clusters and data centers, so delays in grid hookups or site approvals now hit revenue timing directly, not just growth plans.1 3 5
• Anthropic also has real alternatives. It has committed massive spend across AWS, Azure, Google and Broadcom TPU capacity, plus a SpaceX leasing arrangement and the separate Fluidstack buildout. That makes Fluidstack valuable as extra supply, but also exposed if Anthropic reallocates demand toward larger partners or proprietary chips.2 4 6

The next phase is a race to turn provisional AI infrastructure demand into locked in, power secured capacity with more than one hyperscale caliber customer. The winners in GPU cloud will look less like brokers and more like utility scale operators with durable permits, financed sites, and diversified anchor tenants.1 2 3 5 6