Oxmiq Labs builds licensable GPU architecture and AI infrastructure software for chipmakers and data center operators. The Campbell, California-based company has raised $35 million in Series A funding in the latest Oxmiq Labs fundinground, a deal that matters because AI demand is outrunning the supply of affordable compute. Founded in 2023 by GPU architect Raja Koduri, Oxmiq is trying to lower the cost of building custom AI silicon without forcing customers into a full chip program from scratch. It’s a hard pitch. But it’s a clear one.
Fundomo and Samsung Catalyst Fund co-led the round, with MediaTek, AM Intelligence Labs, Pegatron Venture Capital, CDIB-TEN, Darwin Ventures, and Morgan Creek Digital also joining. Total funding now stands at $60 million. Oxmiq will use the new capital to scale OxCore, its licensable GPU architecture platform for semiconductor companies and AI system builders.
Koduri framed the thesis in blunt terms: “Today, state-of-the-art AI reaches most people through a handful of channels, and the cost of the compute underneath is the reason. Bring that cost down, and you widen who gets to build with it.” That’s the wager.
What does Oxmiq Labs build?
Oxmiq isn’t selling a single chip. It’s building a stack that other companies can license, customize, and tape out under their own brands. The core piece is OxCore, a licensable GPU compute core that combines CPU-style scheduling and GPU tensor compute. It also includes CUDA-compatible programmability in one architecture. It’s meant to scale from a chiplet at the edge to a larger SoC and up to data center systems.
Under the hood, OxCore is split into distinct engines. There’s OxORC, a RISC-V orchestrator for async dispatch and hardware task scheduling. There’s OxTEN for tensor workloads. And there’s OxSIMT, the CUDA-compatible engine that supports native CUDA and PTX. That mix matters because Oxmiq isn’t asking customers to start over with a blank-sheet software stack. It’s trying to meet them where the AI world already is.
That software bridge is OxPython. In plain English, it lets code written for CUDA keep running even when the hardware underneath isn’t Nvidia. Oxmiq shows unmodified PyTorch and vLLM workflows running through its runtime. It also shows Hugging Face workflows, with verified targets including Tenstorrent systems and an FPGA implementation of OxCore. For anyone who’s watched alt-silicon startups die on the software hill, that’s probably the most interesting part.
Then there’s OxCapsule, which handles heterogeneous compute. A customer can use one command to authenticate and discover hardware. It can also spin up remote dev environments, stream workloads, and place models across mixed infrastructure. Oxmiq even shows a 397B-parameter model spread across 3 RTX A6000 systems as a single inference endpoint. Before that, teams would be doing manual sharding and driver wrangling. And a lot of swearing.
Who is Raja Koduri and how is Oxmiq Labs set up?
Founded in 2023 around one thesis
Oxmiq was founded in 2023 by Raja Koduri and is built around a simple idea: the GPU stack needs to be re-architected “from atoms to agents,” not merely tweaked at the margins. The company’s position is that customers should be able to own their compute with licensable GPU IP and modular chiplet platforms. It also wants a software layer that avoids dependence on proprietary GPU ecosystems. Oxmiq’s main offices are in Campbell, California, and Hyderabad, India.
Why Koduri has real market fit
Koduri isn’t a first-time founder with a fancy AI deck. He previously held senior roles at Intel, AMD, and Apple. He was Intel’s former executive vice president and chief architect for high-performance computing and GPUs, AMD’s former head of Radeon Technologies Group, and Apple’s former director of GPU architecture. He also worked on visual effects technology in India through Makuta VFX Studio, and he holds a master’s degree in electronics and communications from IIT Kharagpur.
That background is why investors are paying attention. At AMD, Koduri was associated with the architecture shift behind Polaris, Vega, and Navi. At Apple, he helped steer graphics hardware during the Retina era. You don’t have to buy every promise Oxmiq is making to see why people are willing to fund a serious look.
Early traction, team depth, and the round itself
Oxmiq hasn’t published revenue, and that’s fine — this is still a deep-tech startup. But it has shown a few early signals that matter. OxCore has been validated on FPGA hardware and, in simulation, has generated tokens on GPT-4-class models. OxCapsule entered public beta on November 4, 2025, and Oxmiq has 150 beta users across 15-plus organizations and access to 300-plus GPUs. The company has also announced AM Intelligence Labs as a customer.
The team story is stronger than the usual stealth-mode blur. Oxmiq says its staff spans the US and India and represents 500-plus years of combined experience. The advisory roster includes Jim Keller, now CEO of Tenstorrent, as a board member. That doesn’t guarantee execution. But it does tell you this isn’t a two-slide science project.
On funding, the facts are straightforward. The Oxmiq Labs funding round brought in $35 million in Series A capital, co-led by Fundomo and Samsung Catalyst Fund, and joined by MediaTek, AM Intelligence Labs, Pegatron Venture Capital, CDIB-TEN, Darwin Ventures, and Morgan Creek Digital. Total funding is now $60 million, and the company will use the new money to scale OxCore.
Where Oxmiq sits against incumbents
Oxmiq is entering a category that already has real players. Arm licenses IP broadly and offers access to pre-designed IP packages through its licensing programs. Imagination sells GPU and AI IP through its PowerVR family. VeriSilicon markets Vivante GPGPU IP ranging from embedded devices to server-class use cases. Those are the cleanest apples-to-apples comparisons on the IP side.
The legacy alternative is uglier: either license conventional GPU IP and adapt your system around it, or stay tied to Nvidia-centric hardware and software stacks. Oxmiq’s pitch is that customers should be able to tape out custom AI silicon under their own brand. They’d keep CUDA-style software compatibility and build across chiplets, SoCs, and mixed fleets without getting trapped by one vendor. That’s a sharper position than “we also do AI chips,” which, frankly, is where a lot of startups stop.
Why does Oxmiq Labs funding matter now?
Chip startups usually burn huge amounts of capital before they prove much of anything. Oxmiq is trying a less reckless route. Its model is IP-first, not full-SoC-first, and the company already generates revenue from customer engagements while preserving cash for the stack itself. That’s a lot more capital-efficient than pretending every ambitious architecture company should also become a product company. A systems company. And a manufacturing story all at once.
That’s also why Rajeev Surati’s comment from Fundomo matters. He said most compute IP forces customers to bend memory, packaging, and foundry choices around the chip, while Oxmiq is doing the reverse. Strip out the investor polish and the point is simple: if customers can start with their own constraints instead of the vendor’s, the economic case gets stronger.
The use of proceeds is targeted. Oxmiq isn’t saying it will build a moonshot factory or a giant proprietary cloud. It plans to scale OxCore. For buyers, that could mean faster access to custom AI silicon design paths. For investors, it’s a bet that licensable architecture and compatibility layers may be a smarter wedge than trying to outspend entrenched GPU vendors head-on.
How big is the market for custom AI GPU IP?
The timing isn’t random. Grand View Research estimates the global graphic processor market could reach $357.99 billion by 2030. It also sizes the data-center GPU chip market at $2.31 billion in 2024, with that segment projected to hit about $5.03 billion by 2030 at a 13.9% CAGR. Those numbers don’t map perfectly to Oxmiq’s addressable market, but they do show why investors are hunting for picks-and-shovels plays below the hyperscaler layer.
The structural trend is obvious. AI inference is spreading across edge boxes, enterprise clusters, sovereign infrastructure projects, and mixed fleets that don’t look anything like a neat rack of identical accelerators. Oxmiq’s own software messaging leans hard into that reality: heterogeneous hardware, CUDA gravity, and the need to make existing code run without rewrites. If that shift keeps accelerating, companies that license adaptable compute IP could get a real opening.
What to watch after Oxmiq Labs funding
The next thing to watch isn’t the headline round size. It’s whether Oxmiq can turn impressive architecture language into repeatable customer wins.
The Oxmiq Labs funding story is interesting because it sits between two extremes: old-school IP licensing and all-in AI hardware moonshots. If Koduri’s team can prove that custom GPU IP plus software compatibility really cuts time, cost, and lock-in, Oxmiq could matter a lot more than a typical Series A startup. If not, it’ll end up as another smart idea that ran straight into the brutality of the compute market.
Read how Together AI raised an $800M Series C led by Aramco Ventures to expand its open AI infrastructure platform, scale GPU compute capacity, and help developers build, fine-tune, and deploy open-source AI models at enterprise scale.
FAQ
- What is the latest Oxmiq Labs funding round? Oxmiq Labs has raised $35 million in a Series A round. Fundomo and Samsung Catalyst Fund co-led the deal, and the round brought the company’s total funding to $60 million. The money is earmarked to scale OxCore, the startup’s licensable GPU architecture platform.
- What does OxCore actually do? OxCore is licensable GPU IP that customers can customize and tape out as part of their own AI silicon programs. It combines a RISC-V orchestration layer and tensor acceleration. It also includes CUDA-compatible programmability, so developers can keep using familiar software flows instead of rebuilding everything from scratch.
- Who founded Oxmiq Labs? Raja Koduri founded Oxmiq Labs in 2023 and serves as CEO. He’s one of the better-known names in GPU architecture, with senior roles at Intel, AMD, and Apple before launching the company.
- Is Oxmiq Labs a GPU IP company or an AI infrastructure startup? It’s both, and that’s the point. Oxmiq sells licensable GPU IP through OxCore, but it also builds software and orchestration tools like OxPython and OxCapsule to make heterogeneous AI compute usable across different hardware environments.
