neocloud · bare metal · Layer 2.5
TensorWave
Teams committed to AMD Instinct, or after the large-memory parts, and ready to sign a contract to get them.
high portability Portable. The workload is standard enough to move to another vendor without a rewrite.
What you are buying
- Business model
- Purpose-built GPU cloud. Owns or leases its own datacentre capacity and sells it directly.
- How you reach it
- You get the physical machine. Maximum control, and every layer above it is yours to run.
- Accelerators
- AMD Instinct only: MI300X, MI325X, MI355X and the new MI455X. Its bare-metal page quotes 288GB of HBM3E memory per GPU.
- Regions
- North America. Headquartered in Las Vegas, adding new North American MI355X regions (Magnetar, June 2026); no region list on its own site.
- Pricing model
- Not published. tensorwave.com/pricing returned a 404 on 17 Nov 2026, and every call to action leads to a booked call.
- Getting started
- Sales call. Bare metal with full-machine ownership, managed Kubernetes and managed Slurm are all reached through 'Book A Call'; there is no self-serve sign-up.
- Capacity
- 8,192 MI325X online and more than 2 GW of long-term capacity secured (Magnetar, June 2026). Its home page announces a multi-year, 457 MW partnership with OpenAI.
- Ownership
- Private. A $350M Series B at a $1.55B valuation on 10 Jun 2026, co-led by Magnetar and AMD Ventures, with Maverick Silicon, Nexus and Western Frontier (Magnetar).
Verified 2026-11-17. Fields reading “not published” are exactly that — we do not estimate a figure a vendor withholds.
The counterparty
AMD is its only chip supplier and, through AMD Ventures, an investor. Its home page announces OpenAI as a multi-year partner. A buyer signs with a company tied to one chip vendor's roadmap.
A multi-year GPU commitment is a credit decision wearing a cloud contract. This is the section no benchmark covers and the one that decides what happens to your workload in 2028.
The catch
AMD only, so code written for CUDA has to run on ROCm first. There is no published price and no credit-card route in: a call comes before anything else.
The economics
Prices are set per contract, so there is no rate to compare until you ask. Its scale is reserved capacity for large buyers, not on-demand hours.
No rate is quoted on this page on purpose. Published list prices at this layer move weekly and essentially nobody signing a real contract pays them. Dated, sourced figures live in the GPU cost calculator, where the cheapest and dearest published on-demand price for the identical H100 sit 5.6× apart.
The layers underneath this one
Whatever you rent here is a chip in a building that needs power. In 2026 megawatts, not silicon, are the binding constraint on the whole industry — a frontier rack draws 120–200 kW against a 2026 average near 27 kW, and the US interconnection queue exceeds 2,600 GW.
Layer 2 — Silicon · Layer 1 — Energy · The interconnection queue · Tokens per watt
Related providers
- Fluidstack — neocloud, high portability
- OVHcloud — neocloud, high portability
- Voltage Park — neocloud, high portability
- Civo — neocloud, high portability
- CoreWeave — neocloud, high portability
- GMI Cloud — neocloud, high portability