It’s not every day a GPU architect steps off the trillion-dollar train at Intel and into the fire of startup land with nothing but his instincts, a blueprint in his head, and three decades of hard-earned scars. But Raja Koduri didn’t just leave the party, he brought his own sound system, rewired the lighting, and built a new stage while most people were still refreshing their LinkedIn. Two years later, Oxmiq Labs is out of stealth and already making noise, and with $20M in seed funding led by MediaTek, it’s not white noise. It’s signal.
Oxmiq isn’t chasing the GPU hype train, it’s laying new tracks entirely. While the world zigged into silicon arms races, Raja Koduri zagged into software-first GPU IP. The kind of tech that doesn’t just play catch-up, it bends categories. The OXCORE platform? It’s a modular GPU engine built to flex across scalar, vector, and tensor ops like it’s casually solving quantum mechanics before breakfast. The OXQUILT architecture? A chiplet-based orchestra where memory, compute, and interconnects play in perfect sync from edge devices to hyperscale data centers.
But it’s OXPython that should rattle some cages. A CUDA-compatible software stack that runs Python-based AI models on non-NVIDIA hardware without needing a rewrite? That’s not just compatibility, it’s liberation. It’s also how you land a partnership with Tenstorrent and start shipping revenue before your Series A even warms up.
The company’s IP isn’t just smart, it’s scrappy. Capital-efficient, no EDA baggage, no million-dollar tape-outs. A GPU IP stack designed by a squad with 500+ years of collective experience, from Apple to AMD to Intel, and a history of generating over $100B in value along the way. Volley Chen, Micah Villmow, Dmitry Babokin, Anurag Agrawal, Shawn Liu. Just to name a few, this isn’t a team, it’s a hit factory.
Lawrence Loh, SVP at MediaTek, called it “a bold vision and world-class team,” and yeah, that tracks. With MediaTek at the front of the investor table, and strategic angels in the mix, Oxmiq isn’t building for the present, they’re licensing the future. From mobile to automotive to AI at the edge, Oxmiq is going where the compute needs to be, not where the legacy roadmaps say it should.
Multimodal AI, India’s booming silicon ambitions, capital-light licensing, and a stack that actually respects developers? That’s not a startup pitch, it’s a masterclass in timing.
You don’t walk away from Intel unless you’ve got a real reason. Oxmiq Labs just gave us $20 million of them.
