Gimlet Labs raises $80 million to optimize AI inference

Gimlet Labs has raised an $80 million Series A funding round led by Menlo Ventures to scale its software platform. The startup has built what it reports is the first and only multi-silicon inference cloud, which is software designed to run AI workloads simultaneously across diverse hardware architectures. In artificial intelligence, inference is the execution phase where a trained model processes data to generate outputs. Rather than locking workloads into a single chip type, the software allows tasks to run across traditional central processing units (CPUs), graphics processing units (GPUs), and high-memory systems. According to the company, this approach speeds up AI inference by 3x to 10x for the same cost and power.

The funding arrives as infrastructure demand grows, though existing hardware remains underutilized. McKinsey estimates data center spending will reach nearly $7 trillion by 2030. However, Gimlet Labs founder and CEO Zain Asgar states that applications currently use existing hardware somewhere between 15 to 30 percent of the time, leaving resources idle. Asgar explained the startup’s objective: “Our goal was basically to try to figure out how you can get AI workloads to be 10x more efficient than ever, today.”

Gimlet Labs was founded by Asgar alongside Michelle Nguyen, Omid Azizi, and Natalie Serrino. The cofounders previously worked together at Pixie, a startup that was acquired by New Relic in 2020, two months after launching with a $9 million Series A round led by Benchmark. Gimlet Labs publicly launched in October with at least $10 million in revenue. The company, which currently employs 30 people, has now raised a total of $92 million in funding. Its seed round was led by Factory, with participation from Eclipse Ventures, Prosperity7, and Triatomic, alongside angel investors including Bill Coughran, Nick McKeown, Raghu Raghuram, and Lip-Bu Tan.

To support its multi-hardware orchestration, the startup has established partnerships with chipmakers and hardware designers, including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix. Tim Tully, the lead investor at Menlo Ventures, noted in a blog post that different stages of AI agent workloads require different hardware, meaning the physical infrastructure is ready but has lacked the software layer to coordinate it. Gimlet Labs delivers its product either as software or via an API to its own cloud platform, targeting large AI model laboratories and data centers.

Why it matters

Gimlet Labs is attempting to solve the AI inference bottleneck by decoupling software from specific hardware, a critical move as data center spending is estimated to reach nearly $7 trillion by 2030. By allowing workloads to run dynamically across different chip architectures, the company aims to eliminate the inefficiencies of idle silicon.