Gimlet Labs Raises $80 Million to Scale Its Heterogeneous AI Inference Cloud

Gimlet Labs has closed an $80 million Series A funding round to address the growing compute challenges of the AI industry. The investment was led by Menlo Ventures and included participation from Eclipse, Factory, Prosperity7, and Triatomic. This capital injection brings the startup's total funding to $92 million just five months after its public launch, reflecting intense market interest in specialized AI infrastructure.

The company is building what it calls a multi-silicon inference cloud, specifically designed to run complex AI agents. Unlike traditional setups that rely solely on homogeneous GPU clusters, Gimlet's architecture orchestrates workloads across diverse hardware including CPUs and specialized SRAM accelerators. This approach has already produced 3-10X speedups on frontier models with over one trillion parameters while operating within the same power envelope.

Gimlet reports that its customer base has tripled in recent months, now including a top tier frontier AI lab and a major hyperscaler. These organizations are leveraging the platform to manage the high token throughput required by agentic workflows, such as coding assistants that ingest entire codebases. The platform's ability to run proprietary models on specialized hardware provides a significant performance edge in what the company calls the Inference Speed Wars.

Founded by the technical team behind Pixie Labs, Gimlet is also innovating at the physical layer of the data center. They are designing facilities to connect disparate accelerators over high-speed networks, solving complex thermal and plumbing issues inherent in mixed-architecture environments. The team, led by Zain Asgar, plans to use the new funding to expand their physical infrastructure and grow their engineering team to meet global demand.