Anthropic Negotiated Gigawatt-Scale Data Center Lease
The firm seeks to secure dedicated compute infrastructure independent of major existing cloud provider dependencies.
Updated on Sept. 23, 2026 in Data Centers
Live Poll
Do you believe massive investment in AI infrastructure will lead to better services for the public?
Anthropic has entered negotiations to lease one gigawatt of data center capacity from a developer backed by Apollo Global Management. This infrastructure move aims to reduce the startup's current reliance on Amazon Web Services, Google, and SpaceX for model training and inference.
Why it matters
By securing dedicated compute footprints, Anthropic intends to bypass the capacity constraints of third-party cloud landlords. The initiative represents a shift toward vertical integration of AI infrastructure, a necessity for firms training next-generation models.
The proposed site aims to host tensor processing units designed by Broadcom and Google, scaling significantly beyond current setups. Broadcom has signaled an appetite to finance up to 20 gigawatts of capacity alongside Apollo and Blackstone.
The players
Anthropic
An AI research and deployment company that develops large-scale foundational models and currently relies on external cloud providers for compute.
Apollo Global Management
An alternative asset manager that provides financing for large-scale data center infrastructure projects.
Broadcom
A semiconductor and infrastructure software company that designs custom tensor processing units for large-scale AI workloads.
Stream Data Centers
A developer of data center facilities currently in discussions with Anthropic regarding direct leasing agreements.
The details
The facility will be equipped with tensor processing units — specialized hardware accelerators designed to expedite matrix multiplication, the fundamental mathematical operation behind neural network training. To support these operations, Anthropic is discussing direct leases for sites developed by Stream Data Centers. This infrastructure strategy builds upon a US$35 billion arrangement already in place to lease TPU capacity through an Apollo and Blackstone-backed vehicle.
Timeline
September 23, 2026: News reports emerged detailing Anthropic's data center lease negotiations.
The Tech Race
This move marks a departure from reliance on general-purpose cloud providers toward dedicated, proprietary-aligned hardware stacks. It places Anthropic in a hardware-intensive race, where total industry chip spending is projected to reach US$700 billion to maintain training capacity.
The deal could take months to finalize, meaning no immediate change to AI model performance or availability for end users is expected. Long-term, this investment aims to secure the compute throughput necessary to prevent service bottlenecks during future model training cycles.
The takeaway
Anthropic is moving toward a self-contained compute model to ensure it has the hardware necessary for future scaling. Observers should monitor whether this model of dedicated infrastructure leasing becomes the standard for other top-tier AI labs as capital requirements approach the US$40 billion threshold.
Further reading
For more context on how infrastructure scaling impacts the industry, visit /tech/data-centers/.
Live Poll
Do you believe massive investment in AI infrastructure will lead to better services for the public?
