GPU cloud funding and acquisitions: who raised and who got bought (2023 to 2026)

Money is pouring into the layer between AI teams and the GPUs they run on. GPU clouds, compute marketplaces, orchestration tools and price indices have all raised large rounds, and several have been acquired. This post collects the main GPU cloud funding rounds and acquisitions from 2023 to 2026 in one place.

A note on numbers: valuations and deal values are as last reported in the press. Many were never confirmed by the companies involved.

Acquisitions

Company	What it did	Acquirer	Date	Reported value
OpenRouter	LLM API aggregator	Stripe (reported)	Aug 2026	$7B+
Run:ai	GPU orchestration	NVIDIA	Closed Dec 2024	~$700M
OctoAI	Inference platform	NVIDIA	Sep 2024	~$250M
Paperspace	GPU cloud	DigitalOcean	Jul 2023	$111M

Two patterns stand out. Chip and cloud companies are buying the software that sits on top of GPUs: NVIDIA bought both an orchestration layer (Run:ai) and an inference platform (OctoAI). And the biggest number on the list is not a GPU company at all. It is OpenRouter, which aggregates models rather than hardware.

OpenRouter: the aggregation layer, priced

OpenRouter put every major LLM behind one API. Its trajectory is the fastest on this list:

Date	Event	Reported value
Jun 2025	Series A	$547M
May 2026	Series B, led by CapitalG	$1.3B
Aug 2026	Reported acquisition by Stripe	$7B+

That is roughly half a billion to over seven billion dollars in about fourteen months. The lesson for the compute market: a neutral layer that sits across many suppliers and gives developers one API can be worth a great deal. We wrote more about that in OpenRouter for GPUs.

Funding rounds: GPU clouds and AI clouds

Company	What it does	Round	Date	Reported valuation
Together AI	AI cloud	$800M Series C	Jul 2026	$8.3B
RunPod	Developer GPU cloud	$100M	Jun 2026	$1B
Prime Intellect	Started as a compute aggregator	$130M Series A	Jul 2026	$1B
Mithril (Foundry)	Spot GPU auctions	$80M seed and Series A	Mar 2024	$350M
SF Compute	Tradable blocks of GPU time	$40M	Nov 2025	$300M

And the largest public company in the group: CoreWeave, the neocloud, went public in March 2025 at about $23 billion. If you're unsure what separates a neocloud from a hyperscaler, read what neoclouds are.

Funding rounds: the plumbing

The rounds above fund supply. A second wave is funding the plumbing around it: the tools that find, schedule, price and trade compute. Since June 2026 alone:

Company	What it does	Raise	Date
Hydra Host	GPU compute infrastructure	$100M	Jun 2026
Ornn	Price index, ICE futures and dedicated-capacity deals	$33M seed, led by a16z crypto	Jun 2026
SkyPilot	Orchestration across clouds	$20M seed	Jul 2026
Liquid Compute	Compute indices and futures	$15M	Sep 2026

Earlier, Silicon Data, whose GPU rental indices CME's new futures are built on, raised a $30.5M Series A with CME Ventures, DRW and Jump.

What the money says

Compute is being treated as a commodity. Index providers, futures and auctions are all getting funded, and CME lists H100 and B200 rental futures from 5 October 2026. You only build that machinery around something people expect to trade. Our explainer on GPU compute futures covers how it works.

Each company owns a slice. The funded companies split into clouds that sell their own capacity, tools that help you buy across clouds, and indices and exchanges that price compute. As we lay out in our comparison of GPU cloud aggregators and marketplaces, none of them yet combines a view of the whole market with routing that places the job.

The neutral layer is still open. OpenRouter showed what a neutral aggregation layer can be worth for models. For GPUs, buyers still compare prices by hand across a few of the 209 GPU clouds SemiAnalysis tracks.

Where Opengrid fits

Up front: Opengrid is pre-launch. The market view ships first and the routing API is in development.

Opengrid is building that neutral layer for compute: a live view of prices and availability across providers, then one API key that routes each job to whichever provider wins on price, availability, region and reliability.