Oxmiq raises $35M to build AI chip architecture that runs CUDA programs without Nvidia hardware

Oxmiq Labs, the AI chip startup founded by semiconductor veteran Raja Koduri, has closed a $35 million Series A funding round. The raise brings the company’s total capital to $60 million and puts it squarely in the crosshairs of one of tech’s most consequential battles: who gets to power the AI infrastructure boom besides Nvidia.

The round was co-led by Fundomo and Samsung Catalyst Fund, with MediaTek and Pegatron Venture Capital also participating.

What Oxmiq is building and why it matters

The core product here is OxCore, a licensable GPU IP architecture. Think of it as a blueprint that other companies can license to build their own custom AI chips, rather than designing silicon from scratch or simply buying whatever Nvidia is selling.

OxCore merges scalar, vector, and tensor engines into a single modular architecture. Each of those engine types handles different kinds of math that AI workloads require. Scalar engines process one data point at a time. Vector engines handle arrays of data. Tensor engines, the workhorses of modern deep learning, crunch through matrix multiplications at scale. Combining all three into one licensable package means chip designers can target everything from edge AI devices to large-scale data center deployments without starting from zero.

OxCore is designed to run CUDA-based programs on non-Nvidia hardware. CUDA is Nvidia’s proprietary software framework, and it’s arguably the real moat around Nvidia’s AI chip dominance. Developers have spent years building AI models and applications on CUDA, creating a massive switching cost. If OxCore genuinely delivers CUDA compatibility on alternative silicon, it could lower one of the tallest barriers keeping companies locked into Nvidia’s ecosystem.

The man behind the chip

Raja Koduri is a former chief architect at both Intel and AMD. At Intel, Koduri led the company’s push into discrete GPUs. At AMD, he helped architect the Radeon graphics lineup during a critical period of the company’s turnaround.

Oxmiq emerged from stealth mode in August 2025. Less than a year later, it has closed a Series A with strategic investors who have direct stakes in the semiconductor supply chain. Samsung manufactures chips. MediaTek designs them. Pegatron builds the devices they go into. OxCore is targeted for licensing availability in the first half of 2026.

What this means for the AI chip market

Oxmiq’s licensing model offers companies an alternative to buying Nvidia hardware at premium prices, designing custom silicon at a cost of billions over years, or settling for less capable alternatives. Companies could license the OxCore architecture and produce custom AI silicon tailored to their specific workloads.

ARM’s licensing model transformed the mobile processor market by enabling dozens of companies to build custom chips on a shared architecture. Qualcomm, Apple, Samsung, and MediaTek all license ARM designs and then customize them for their own products. Oxmiq appears to be pursuing a similar playbook for AI GPUs.

Startups like Groq, Cerebras, and SambaNova have all raised significant capital to challenge Nvidia with different architectural approaches. What distinguishes Oxmiq is the licensing model itself. Rather than competing as a chip vendor, Oxmiq is positioning itself as an enabler, selling the tools for others to compete.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.