The round was led by Andreessen Horowitz and added Arm Holdings and M12, Microsoft’s venture fund, as new backers, chief executive officer Zain Asgar said. It closed just six months after Gimlet raised $80 million, a pace Asgar attributed to unsolicited term sheets from investors chasing exposure to heterogeneous AI compute.
From Software to Data Centres
Gimlet started with a narrow thesis: write software that carves up AI tasks and dispatches them to whichever chip handles them best. That work exposed a deeper problem. Most customers had no idea how to physically arrange data centre hardware for mixed-chip setups, where different processors need different cooling, power and temperature profiles.
“We figured out all this really cool tech around how to distribute workloads to different types of chips,” Asgar said. “But one of the challenges that we kind of realised is that nobody builds data centres in a heterogeneous manner. So we kind of started peeling the onion one layer at a time.”
The company now helps customers design those facilities and is building data centres of its own. Its customers include AI labs and financial services firms, though Asgar declined to name them. The focus has narrowed to one demanding task — running AI models as fast as possible — where combining multiple chip types is already producing compelling results.
Why Investors Are Rushing In
Gimlet’s rapid fundraising reflects a broader shift in how venture capital views AI infrastructure. Less than a year ago the company closed a $12 million seed backed by Factory, Intel CEO Lip-Bu Tan, Figma’s Dylan Field and Andreessen general partner Raghu Raghuram, who is also leading the firm’s Gimlet investment.
Raghuram argues the AI boom has so far run on a single technology stack anchored by Nvidia chips, an anomaly he expects to end. “The history of computing is never that way, especially a market that’s going to be trillions and trillions of dollars and so many different use cases,” he said. “Fundamentally, we think the world will be multi-silicon, and we think the world will have multiple different types of workloads with very, very different demands.”
The competitive landscape is forming fast. Last month rival Callosum raised $100 million in early financing from backers including the UK’s public AI fund. Arm’s direct investment in Gimlet, alongside a collaboration to make Gimlet’s software compatible with Arm’s chip technology, signals that incumbents beyond Nvidia want a foothold in multi-silicon orchestration.
What Happens Next
Watch for Gimlet to move from software vendor to infrastructure operator. Building its own data centres puts it in direct competition with hyperscalers and specialised AI cloud providers, and it will need to show that heterogeneous setups can beat single-vendor Nvidia clusters on both cost and performance. Arm’s involvement hints at a roadmap where mobile-derived chip architectures handle more inference work, pressuring margins across the AI hardware stack.
The bigger signal is timing. Three rounds in under a year, with valuations climbing from single-digit millions to $3 billion, suggests investors are pricing in a post-Nvidia world before it fully arrives. If multi-silicon AI compute becomes the default, Gimlet’s orchestration layer becomes critical plumbing. If Nvidia extends its dominance another cycle, that same bet looks early. Either way, the heterogeneous AI data centre is now a funded category, not a thesis.
— David Kim, technology desk, AXO News