Thunder Compute has raised $13 million in Series A funding to scale its GPU virtualization technology and help enterprises extract more computing capacity from underutilized AI infrastructure. The round was led by Matrix Partners, which also led Thunder Compute’s Seed round earlier this year, with participation from Y Combinator and CEAS Investments.
Thunder Compute is targeting what it describes as more than $200 billion of data center capacity currently sitting idle because GPUs remain substantially underutilized.
The company cited Cast AI’s 2026 State of Kubernetes Optimization Report, which found average enterprise GPU utilization of approximately 5%.
Thunder Compute argues that a major reason for the low utilization is the way GPUs are traditionally allocated. A GPU can remain assigned to a single workload even when that workload is not actively using it.
CPU, storage and memory infrastructure, by comparison, can be shared among multiple workloads through virtualization layers.
Thunder Compute is attempting to bring a similar abstraction to GPUs by treating them as network-accessible resources that can be dynamically assigned to workloads when needed.
Its software operates beneath the workload layer and changes how machine learning code interacts with physical GPUs.
The company says the virtualization process is transparent to developers, allowing the technology to be introduced into existing machine learning workflows without requiring applications to be rewritten around the infrastructure.
Thunder Compute has spent approximately four years developing its virtualization technology.
The company has also operated a self-service cloud offering that it uses to harden and validate the underlying technology.
More than 10,000 users have run workloads using Thunder Compute’s virtualized GPUs through that cloud environment.
The Series A financing will support a broader enterprise push as Thunder Compute works with organizations that want to increase capacity within GPU fleets they already own.
Rather than focusing primarily on optimizing individual AI workloads, Thunder Compute is addressing utilization at the infrastructure level.
The company ultimately wants to create an environment in which GPUs can be virtualized and dynamically shared across data centers in much the same way other computing resources are allocated today.
That model could become increasingly important as AI infrastructure spending accelerates and organizations struggle to obtain enough GPU capacity for model training and inference workloads.
KEY QUOTES:
“There is a massive efficiency problem with over $200 billion in data center capacity sitting idle because GPUs are the only hardware that aren’t virtualized. Four years ago, we saw this gap and set out to build out the VMware for GPUs. Since then, we have invented cutting-edge virtualization technology for GPUs and have hardened it through our self-serve cloud offering, where over 10,000 users have run their workloads on our virtualized GPUs. This funding will help us virtualize GPUs at scale by partnering with enterprises seeking to create more capacity within their existing GPU fleets.”
Carl Peterson, Co-Founder Of Thunder Compute
“Lots of startups focus on optimizing specific workloads but Thunder Compute stood out by figuring out how to optimize data centers in a generalized and transparent way. Carl, Brian and the rest of the team had a lot of foresight to approach the problem this way and we believe the timing is now perfect to deploy their solution toward addressing today’s GPU crunch.”
Ilya Sukhar, General Partner At Matrix Partners
