Gimlet Labs, a startup founded by Stanford adjunct professor Zain Asgar, has raised $80 million in a Series A funding round led by Menlo Ventures. The company is addressing a critical issue in the rapidly expanding AI landscape: the inefficiency of current hardware utilization for AI workloads. Gimlet Labs’ solution isn’t about building new chips but about making the most of the chips already in use.
The Problem: Wasted Compute Power
As AI models grow more complex, the demand for computing power skyrockets. Data centers are projected to spend nearly $7 trillion by 2030, yet current deployments only leverage between 15% and 30% of existing hardware capacity. This means hundreds of billions of dollars are effectively being wasted on idle resources. The bottleneck isn’t a lack of compute; it’s the inability to efficiently distribute AI tasks across diverse hardware architectures.
Gimlet Labs’ Solution: A “Multi-Silicon Inference Cloud”
Gimlet Labs has developed software that acts as an orchestrator, enabling AI workloads to run simultaneously across CPUs, GPUs, and high-memory systems. Different parts of an AI application – inference, decoding, and tool calls – each have unique hardware requirements. No single chip excels at all of them, but Gimlet’s platform dynamically assigns tasks to the most suitable hardware available.
“We basically run across whatever different hardware that’s available,” says Asgar.
This approach allows for significant performance gains—between 3x and 10x faster inference at the same cost and power. The software can even slice models to run across different architectures, using the optimal chip for each portion.
Key Partnerships and Early Traction
Gimlet Labs has already secured partnerships with major chip manufacturers including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix. The company launched publicly in October and immediately generated eight-figure revenue. Over the last four months, its customer base has doubled, now including a major AI model developer and a large cloud computing provider (names undisclosed).
From Observability to Optimization
The founding team previously collaborated at Pixie, an open-source observability tool for Kubernetes, acquired by New Relic in 2020. This prior experience in system-level optimization appears to have been crucial to Gimlet Labs’ rapid success. The company now employs 30 people.
Investor Confidence
The Series A round was oversubscribed, with participation from investors including Factory, Eclipse Ventures, Prosperity7, and Triatomic. Notable angel investors include Sequoia’s Bill Coughran, Stanford Professor Nick McKeown, former VMware CEO Raghu Raghuram, and Intel CEO Lip-Bu Tan.
Gimlet Labs’ approach highlights a growing trend: the focus on software-driven efficiency in hardware utilization. Rather than chasing the next silicon breakthrough, the company is maximizing the value of existing infrastructure. This is a pragmatic solution that could have significant implications for AI deployment at scale.
