OXMIQ RAISES $35MN TO LICENSE AI CHIP DESIGNS

Raja Koduri, the former Intel chief architect, is back with a startup that aims to do for AI chips what Arm did for mobile processors. Oxmiq Labs announced a $35 million Series A round on Wednesday, bringing total funding to $60 million, and the company already has a clear pitch: we license a block of GPU, tensor, and CPU compute that you integrate with your own design and skip the hundreds-of-millions-dollar design bill.

The round was co-led by Fundomo and Samsung Catalyst Fund, with participation from MediaTek, Pegatron Venture Capital, and several other financial and strategic investors. Jim Keller, the chip design veteran who now runs Tenstorrent, joins Oxmiq's board as part of the announcement. Keller said "Raja and this team are creating an open GPU architecture, a much-needed step toward removing the artificial boundaries around AI innovation,". The statement aligns with the company's broader thesis: that AI hardware should not require a full vertical chip program to be competitive.

you may like

ANTHROPIC TAPS BEN BERNANKE TO OVERSEA PUBLIC INTEREST MISSION

TRUMP SAYS APPLE AND INTEL REACH A CHIP DEAL. REALITY IS MESSIER

ANTHROPIC SIGNALS PARTNERSHIP WITH SAMSUNG TO BUILD OWN AI CHIPS

WHAT OXMIQ SELLS

The core product is OxCore, a scalable, licensable GPU design that bundles three compute engines into a single block: a GPU engine for graphics and general parallel work, a tensor processing engine for the matrix math that drives AI inference and training, and a central processor for control logic. That block is the kind of thing a company like MediaTek or a cloud provider could drop into its own system-on-chip without designing the GPU part from scratch.

Oxmiq also offers OxQuilt, a chiplet integration architecture that lets designers combine multiple chips and memory in a single package, and it is built to work across different foundries, memory types, and packaging options. The stack includes software as well. OxPython allows developers to run existing code written for Nvidia's CUDA platform on OxCore without changes. That is the killer detail: compatibility without a rewrite, which lowers the switching cost for any team that currently builds on Nvidia's ecosystem.

THE ARM COMPARISON IS THE WHOLE POINT

Koduri framed the ambition directly: "We would want to be the Arm of this next era,". The reference is precise. Arm succeeded not by selling finished chips but by licensing a processor architecture that everyone else could build around. Oxmiq wants the same model for AI accelerators. Instead of every AI company raising hundreds of millions to design its own GPU, they license OxCore, build their own custom die around it, and keep more control over the final product.

The comparison is aspirational but not empty. Building a state-of-the-art AI chip from scratch requires years of work and investments running into the hundreds of millions of dollars. Even large semiconductor companies hesitate at that cost. A licensable block that already works with existing AI software dramatically shortens the timeline and the risk profile. Oxmiq is targeting semiconductor companies, cloud providers, and AI system builders as potential licensees.

WHY THE TIMING MATTERS

Nvidia's dominance in AI chips has drawn a wave of challengers. Companies including Google and OpenAI have moved to design custom silicon, while others are betting on open licensing models and chiplet-based approaches. Oxmiq enters a field that is already crowded with startups, but the team's pedigree, Koduri's background at Intel and the board seat for Keller give it a credibility that most early-stage chip companies lack.

The primary use for the new funding is finishing the initial IP portfolio and bringing it to market, with plans to expand engineering headcount. The $35 million Series A is not a giant raise by AI chip standards, but it is enough to tape out a few test chips, refine the software stack, and sign early licensees. What matters is whether those licensees appear. If a major cloud provider or a second-tier smartphone chipmaker adopts OxCore, the model gains real traction.

THE SOFTWARE MOAT

OxPython is the part that should keep Nvidia's CUDA team paying attention. AI developers have spent years writing kernels, libraries, and entire pipelines that target CUDA. The cost of porting that work to a new architecture is a major barrier to switching. OxPython's ability to run existing CUDA code without changes removes that barrier. It does not guarantee performance parity on the first pass, but it makes the initial migration possible, which is more than most competitors offer.

Koduri and his team understand that Nvidia's real moat is not just the hardware but the entire software ecosystem built around CUDA. A drop-in compatibility layer is the most direct way to disrupt that cycle, and it is the reason Keller joined the board.

WHAT TO WATCH

The next twelve months will show whether Oxmiq can actually deliver a licensable GPU that performs competitively against Nvidia's current generation while keeping the software compatibility claim honest. The investors include MediaTek and Samsung Catalyst Fund, both of which could become licensees.

what to read next

APPLE TAKES TO OPENAI TO COURT FOR POACHING AND STEALING TRADE SECRETS

FIDJI SIMO DEPARTS OPENAI DUE TO CHRONIC ILLNESS

SALESFORCE BUYS AI STARTUP FOR $3.6BN

If Oxmiq signs a single marquee licensee before the end of 2026, the model works. If it signs none, it becomes another chip startup with a good story and an empty order book. For now, the $60 million in the bank and the two names on the board give the pitch weight. The rest depends on the silicon.