← Back

AI Compute Crunch Drives Anthropic's Gigawatt Power Play

Sep 23, 2026
AI Compute Crunch Drives Anthropic's Gigawatt Power Play

Anthropic’s negotiations to lease up to 1 gigawatt of data center capacity from Stream Data Centers, an Apollo Global Management portfolio company, represent a pivotal move to secure its long-term compute foundation. This deal, potentially housing Google/Broadcom-designed TPUs, is a direct response to the intense compute scarcity that has defined the AI landscape for the past 24 months. By locking in this much power, Anthropic is not just leasing space but attempting to build a strategic moat against rivals, decoupling its scaling trajectory from the fiercely competitive GPU market dominated by Nvidia and creating an owned-and-operated infrastructure profile similar to its larger cloud-backed peers. This transaction fundamentally alters the AI infrastructure value chain by elevating private equity firms like Apollo into the role of kingmakers for well-funded AI labs. For Anthropic, securing a gigawatt of power provides an asymmetric advantage, allowing it to train larger, more complex models at a predictable cost, a direct challenge to the pay-as-you-go model offered by AWS, Google Cloud, and Microsoft Azure. The potential use of Google’s TPUs indicates a strategic diversification away from Nvidia’s hardware, exposing a vulnerability in the latter’s market dominance and forcing a competitive recalculation for all major AI players who now must secure their own long-term power and hardware pipelines. Looking forward, this deal sets a new precedent for AI infrastructure financing, suggesting a shift from venture capital-led funding rounds to asset-backed financing strategies. In the next 12 months, expect other well-capitalized AI labs like Cohere and Mistral AI to pursue similar large-scale data center leases, potentially sparking a land grab for power capacity. The critical variable will be whether these labs can secure the necessary hardware to fill this capacity. This trajectory suggests the AI arms race is moving beyond algorithms and into the physical world of power grids and real estate, making utility-scale infrastructure the next major competitive battleground.