Koduri, who had a stint at AMD as senior vice president and chief GPU architect, reckons the world needs a new kind of GPU. Apparently unsatisfied with just tweaking what already exists, Oxmiq promises to rip the entire thing up and start again, leaning on what it claims is 500 years of collective experience and hundreds of patents.

With a suspiciously buzzword-heavy “Atoms to Agents” tagline, the company reckons it can rebuild from the transistor level up to intelligent AI systems. Given the growing mess of multimodal AI workloads chewing on everything from images to audio, Oxmiq thinks GPUs need to be smarter, faster, and more flexible than what AMD or Nvidia currently cough up.

The firm’s shiny new architecture is called OXCORE, which chucks together scalar, vector, and tensor units for whatever AI tasks you fancy. The design is allegedly scalable enough for tiny gadgets or sprawling data centres, powered by a chiplet design it calls OXQUILT. This setup supposedly lets punters mix compute, memory, and interconnect blocks like Lego, which in theory slashes R&D time and cost.

But Oxmiq’s main gamble is its “Software First” model. Its flagship platform, OXPython, promises to run existing Python-based CUDA apps on any hardware without needing code changes.

First in line to test that out is Tenstorrent, whose chief executive Jim Keller said: “We’re excited to partner with OXMIQ on their OXPython software stack. OXPython’s ability to bring Python workloads for CUDA to AI platforms like Wormhole and Blackhole is great for developer portability and ecosystem expansion. It aligns with our goal of letting developers open and own their entire AI stack.”

Oxmiq’s broader software play, called OXCapsule, claims to blur the lines between different platforms so developers don’t have to faff with low-level hardware differences.

The company has already wooed the likes of MediaTek into its corner. MediaTek senior vice president Lawrence Loh said: “OXMIQ has an impressive bold vision and world-class team. The company’s GPU IP and software innovations will drive a new era of compute flexibility across devices, from mobile to automotive to AI on the edge.”

Instead of selling its own graphics cards, the outfit plans to flog its designs through an IP licensing model to dodge costly chip production runs. It has already shaken down investors for $20 million in seed cash.

Koduri reckons Oxmiq is the first proper GPU startup to hit Silicon Valley in more than 25 years. We’ll believe it when we see the silicon.