heterogeneous-computing
(4)
Behnam Bastani and the computer hiding in plain sight
From color science to Oculus Link, Behnam Bastani has kept asking where the work should happen. At OpenInfer, that question has become a business built around the computing power already within reach.
The Chip Was Only the Beginning
d-Matrix bet early that serving AI answers would become its own computing problem. Seven years later, the Santa Clara company is building the chips, network links, rack designs and software to solve it.
MIPS
MIPS is a San Jose-based semiconductor IP company with a 40-year heritage in RISC computing. After abandoning its own legacy architecture, MIPS reinvented itself around the open RISC-V instruction set and now builds compute subsystems for autonomous platforms across automotive, industrial, robotics, and edge AI markets. Its Atlas portfolio - spanning Sense, Think, Act, and Communicate building blocks - targets the low-latency, safety-critical demands of what the company calls Physical AI. In July 2025, GlobalFoundries announced a deal to acquire MIPS.
Zain Asgar
Zain Asgar is the Co-Founder and CEO of Gimlet Labs, a San Francisco-based AI infrastructure company building the world's first multi-silicon inference cloud. With a PhD from Stanford in electrical engineering focused on GPU energy modeling, Asgar previously led engineering at Google AI (where his work became Google Lens) and founded Pixie Labs, a Kubernetes-native observability platform acquired by New Relic in 2020. At Gimlet Labs, he is tackling one of AI's most pressing infrastructure challenges: making AI inference 3-10x more efficient by intelligently routing workloads across heterogeneous hardware including NVIDIA, AMD, Intel, ARM, and specialized accelerators like Cerebras.