Oracle OCI Supercluster A100s
Oracle OCI Supercluster A100s is a large GPU cluster operated by Oracle, located in the United States, in the private sector. It came online around November 2023. At that moment it ranked number 4 among known clusters.
It is built on 32,768 NVIDIA A100. That amounts to about 10.3K H100e of AI compute. The cluster draws roughly 27.19 MW. Its hardware is estimated at $646M.
Epoch AI rates the certainty of this record as likely. Unclear exactly when this first became operational, if it was ever built. The announcement March 2023 states Oracle OCI can now scale to 32k A100s, but it's unclear whether this implies that a 32k A100 cluster has already been built, or just that they have the capacity to build one. It seems likely that a cluster of this size was built, but there's little direct evidence for this. I estimate Q4 2023 to be slightly conservative on the timing for when it was first operational.
- Owner
- Oracle
- Users
- Cloud
- Sector
- Private
- Status
- Existing
- Country
- United States
- Primary chip
- NVIDIA A100
- Primary chip count
- 32,768
- Total AI chips
- 32,768
- H100 equivalents
- 10.3K
- 8-bit throughput
- 2×10¹⁹ OP/s
- Power capacity
- 27.19 MW
- Hardware cost
- $646M
- First operational
- Nov 15, 2023
- Rank when it came online
- 4
- Certainty
- Likely