Afton, Texas, United States
Coreweave Helios is an operational AI data center with AI chips owned by CoreWeave and possibly used by OpenAI. It hosts an estimated 160k H100-equivalents of compute, supported by 132 MW of IT power at a capital cost of $5.0 B. In Q1 2029, we project it to rise to 638k H100-eq compute, 528 MW IT power and a cost of $20.0 B. We estimate that it uses Nvidia B200 chips.
The timeline of IT power and compute capacity is based on satellite imagery, our cooling equipment power model, company disclosures, and regulatory filings.
Helios is retrofitted from a former crypto-mining site owned by Galaxy, which has an additional 2.7 GW of capacity under study, though not expected before 2029.
Satellite imagery
CURRENT COMPUTE
160k H100-eq
3.16e20 8-bit OP/s
PROJECTED
638k H100-eq
1.26e21 8-bit OP/s
CURRENT IT POWER
132 MW
PROJECTED IT POWER
528 MW
CURRENT COST
$5.0 B
PROJECTED COST
$20.0 B
The former crypto mining Helios site begins expanding, with land cleared for additional buildings.
Backup generator pads and steel framing for chillers are visible, similar to the 3D renderings.
Building 1 appears substantially complete. The main white-roofed data hall is visible with exterior mechanical/electrical equipment rows installed on both sides, and adjacent generator/equipment pads plus laydown yards are active. A large graded pad north of Building 1 suggests preparation for the next phase/building. No clear satellite evidence that the AI/HPC facility is operational yet.
Building 1 operational at full capacity. This corresponds to "Phase I", which is expected to be operational by the end of the first half of 2026.
Building 2 and 3 are operational at full capacity. This corresponds to "Phase II", which is expected to be delivered "throughout 2027". We assume it takes until the end of 2027 to reach full capacity.
Building 4 operational at full capacity. This corresponds to "Phase III", which is expected to be delivered "starting in 2028". We assume it takes until the end of 2028 to reach full capacity.
Read more about our methodology.