What Corvex sells — and what it buys
GPU clusters — NVIDIA H200, B200 and GB200 NVL72, as Kubernetes, VMs, bare metal or SLURM, marketed from 1K toward 96K+ GPUs — with confidential computing as the differentiator: weights and data decrypted only inside the GPU's trusted execution environment, invisible even to Corvex as the host. Plus hybrid GPU nodes that join a customer's existing Amazon EKS cluster, Outpost-like on-prem deployments, and an upcoming hosted inference engine.
NVIDIA allocation — as an NVIDIA Cloud Partner with preferential GPU access — plus Tier III+ data-center capacity and power, Spectrum-X and 3.2Tbps InfiniBand networking, and liquid cooling for 100kW+ racks. The Movano merger added the other thing capacity buyers need: a public currency for capex.
Scale proof
The bet
That confidential computing stops being a feature and becomes a foundational requirement — that the industry's "cleartext gap," where model weights sit unencrypted in VRAM visible to any host, closes the way unencrypted HTTP did. If weights are the most valuable IP in AI, the cluster provider that can prove — with hardware attestation — that it cannot see them wins the customers with the most to lose: frontier labs, regulated enterprises, and sovereigns. Sovereignty by cryptography scales globally; sovereignty by geography has to be rebuilt in every country.
Natural counterparty
Model builders whose weights are the business — and regulated enterprises in healthcare, finance and defense that fine-tune on data they can't let leave a compliance boundary. Secondarily: EKS teams offloading GPU workloads to cut the AWS premium without leaving their console. On the buy side: NVIDIA and data-center and power providers — and unusually for this ladder, public-market investors, since the Nasdaq listing makes equity a capacity-funding instrument.
Buyer fit
You need dedicated NVIDIA clusters where weights and data stay encrypted in use — provably, even from the host — whether in Corvex's cloud, hybrid into your EKS cluster, or on your premises.
You want per-second serverless compute or pay-per-token APIs today — Corvex sells reserved clusters, and its hosted inference engine is still upcoming.
Five fields we track but don’t publish
Everything above is public — figures are Corvex-published unless otherwise sourced, and this card is a draft pending direct provider verification. The fields below change weekly and are verified directly with each provider — they live in a private supply-demand ledger, not on this page.