macrostack

neocloud · vm · Layer 2.5

Thunder Compute

Individual developers and small teams who want a cheap single GPU for development and fine-tuning, billed by the minute.

medium portability Movable with effort. Expect to redo images, storage wiring and networking.

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 virtual machines. The familiar cloud model.
Accelerators
NVIDIA RTX A6000, L40, A100 80GB and H100 SXM instances; H100 and A100 sandboxes by access request.
Regions
North America, by its technical specifications page; no finer location is published.
Pricing model
Instances billed per minute while running, with extra vCPU, disk beyond the 100GB included per GPU, and snapshots priced separately, and no egress fee. Sandboxes billed per second; reserved capacity by quote.
Getting started
Self-serve sign-up on its own console. Sandboxes, which it says are never preempted, are by access request.
Capacity
Not published.
Ownership
Private, San Francisco. A $13M Series A led by Matrix Partners with Y Combinator and CEAS (VCA Online, 19 Aug 2026), which also reports more than 10,000 users.

Verified 2026-11-17. Fields reading “not published” are exactly that — we do not estimate a figure a vendor withholds.

The counterparty

A young, venture-backed company whose product is GPU virtualization: its funding release describes treating GPUs as network resources. For hourly development work that matters little; for a long commitment, weigh a small balance sheet.

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

The driver stack is theirs: its specifications page says not to reinstall CUDA. Work that needs full control of the machine and its drivers is a poor fit.

The economics

Priced at the low end on purpose: its own comparison table puts it cheapest of nine clouds for an H100, and it charges nothing for egress. That table is its own; the GPU index carries figures read at each company's source.

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

Sources

Macrostack does not benchmark clusters and does not pretend to. For hands-on performance ratings of GPU clouds, SemiAnalysis’s ClusterMAX tests the hardware directly. This page covers what their scorecard does not: who owns the company and how you get out.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.