Thunder Compute raises $13M to virtualise idle GPUs

What's the deal? Thunder Compute has raised $13 million to help cloud providers extract more work from graphics processing units (GPUs) that sit idle. The company builds software that virtualises GPUs, pooling them so capacity can be scheduled whenever it is available rather than reserved for a single workload.

Why now? Enterprise GPUs average just 5% to 20% utilisation, according to the CastAI 2026 State of Kubernetes Optimization Report. Thunder estimates that idle capacity amounts to almost $200 billion left on the table.

What's the endgame? Co-founder and chief executive officer Carl Peterson wants to build the "VMware for GPUs," making the underlying hardware disappear as it does for storage and central processing units. Thunder's software separates a workload's access to a GPU from the specific chip serving it, letting operators schedule fleets far more efficiently.

The product: Thunder's proprietary software treats GPUs as network resources, accessible to workloads across the data centre. It sits between the developer and cloud provider, abstracting the chip so it drops into existing workflows. "What's critical here is that the virtualization is invisible to the developer," Peterson said.

The economic benefit flows to the cloud provider or enterprise that bought the GPUs, Peterson said, not the developer. Higher utilisation lets operators squeeze more from hardware they already paid for and potentially pass savings on through lower prices.

To date, Thunder has supplied compute to more than 10,000 users on its own cloud of virtualised GPUs. Peterson declined to name customers but said two enterprises are piloting the software, with more to follow. Some have achieved gains of four times or more, though he cautioned results depend on the workload.

Why the funding matters: After four years of development, Thunder has largely been "selling this to ourselves," running its own cloud as a testbed. The round marks a shift from proving the technology in-house toward putting it in the hands of external cloud providers and enterprises.

The signal: At $13 million, the raise sits toward the smaller end of recent funding rounds. But it targets a real bottleneck as AI demand strains GPU supply: the question is not just buying more chips, but wasting fewer of the ones already deployed.

Read more: siliconangle.com

Image credit: Generated with Gemini