Oxmiq Labs, the chip-architecture startup founded by veteran GPU architect Raja Koduri, has raised a $35 million Series A to scale OxCore, a licensable architecture that fuses GPU, CPU and TPU functions into a single IP block. The pitch is deliberately Arm-shaped: Oxmiq does not build chips, it licenses the design and software so semiconductor firms and AI system builders can create custom AI silicon without funding a full chip program. Koduri put the ambition bluntly to Reuters: the company wants to be the Arm of this next era.

  • The $35M Series A was co-led by Fundomo and Samsung Catalyst Fund, bringing Oxmiq's total raised to $60M.
  • Its core product, OxCore, is a licensable architecture that combines GPU, CPU and TPU functions, sold as IP rather than finished chips.
  • Strategic backers span the supply chain: MediaTek reinvesting, Intel Capital as an IP partner, plus Pegatron and Morgan Creek Digital.
  • The timing rides a shift toward custom silicon: custom ASIC shipments are projected to grow 44.6% in 2026, nearly triple the 16.1% expected for GPUs.
Oxmiq's licensing model Oxmiq licenses OxCore IP to chipmakers who build custom silicon, instead of manufacturing finished chips itself. Oxmiq OxCore IP GPU + CPU + TPU + software stack Semiconductor firmsNeocloudsAI system builders CustomAI silicon No fab, no finished product. License the design; the customer tapes out the chip. genztech.blog
Fig 1 Oxmiq's model mirrors Arm: it licenses OxCore IP and software to chipmakers and system builders, who then build their own custom AI silicon without the cost of a full ground-up chip program.

What did Oxmiq raise, and from whom?

On July 1, 2026, Oxmiq closed a $35 million Series A, lifting total capital raised to $60 million. The round was co-led by Fundomo and Samsung Catalyst Fund, with a notably strategic investor list: MediaTek, a seed backer and one of the world's largest fabless chipmakers, reinvested; Intel Capital joined as a strategic IP partner; and Pegatron Venture Capital, Darwin Ventures and Morgan Creek Digital participated. Headquartered in Campbell, California, with a development site in Hyderabad, India, Oxmiq is stacking exactly the kind of supply-chain partners a licensing business needs to reach silicon.

RelatedTogether AI Raises $800M to Undercut Closed AI Models

What is OxCore?

OxCore is a licensable GPU architecture that folds GPU, CPU and TPU functions into one IP block, letting customers build custom AI chips without running a full, expensive chip program from scratch. It anchors a broader stack, OxCore, OxQuilt, OxPython and OxCapsule, meant to help semiconductor companies, neoclouds and AI system builders develop and deploy custom compute across the range of AI workloads. Crucially, Oxmiq does not manufacture anything. Like Arm, it sells the design and the software and lets licensees handle the tape-out.

Why does the timing work?

Because the market is tilting hard toward custom silicon. According to TrendForce, custom ASIC shipments from cloud providers are projected to grow 44.6% in 2026, nearly triple the 16.1% growth expected for GPU shipments. Nvidia still holds around 70% of the AI chip market, but that share is gradually eroding as hyperscalers and chipmakers build their own accelerators to control inference economics and their own compute roadmaps. A company selling the IP to build those accelerators is positioned directly in the path of that shift.

ApproachOxmiq (license IP)Buy merchant GPUsFull in-house chip
Upfront costLow (licensing)NoneVery high
CustomizationHighNoneTotal
Time to siliconFasterImmediateSlow
Roadmap controlShared / highVendor-setFull
Who runs itChipmakers, neocloudsAnyoneHyperscalers

Who is behind it?

Raja Koduri is the whole thesis. He has led GPU and architecture work at Intel, AMD and Apple, and Oxmiq says he has assembled a team with over 500 years of combined experience, hundreds of patents and a collective track record of generating more than $100 billion in revenue at prior companies. In a business where credibility with fabless chipmakers is the product, that pedigree is why MediaTek and Intel Capital are on the cap table.

RelatedBaseten Raised $1.5B, and AI Inference Is the Prize

What to watch · 2026-2027
  • First named licensees. The Arm comparison only holds if real chipmakers ship OxCore-based silicon. Watch for the first design win.
  • Software depth. Custom silicon lives or dies on its software stack. OxPython and the rest have to be genuinely usable, not slideware.
  • The ASIC wave. Prediction: as ASIC shipments outgrow GPUs, licensable AI architecture becomes a real category, and Oxmiq is early to it.

Our take

Oxmiq is a bet that the AI chip market rhymes with the mobile one, where Arm won by licensing an architecture instead of fighting to build the best single chip. The logic is sound: hyperscalers and chipmakers increasingly want custom accelerators they control, and almost none of them want to design a GPU architecture from nothing. A credible, licensable GPU-CPU-TPU core aimed squarely at that need is a genuinely good idea, and Koduri is one of the few people who could plausibly execute it. The risks are the usual ones for IP businesses: everything hinges on real design wins and a software stack customers actually adopt, and $35 million is seed-scale money for silicon ambitions. But the direction of the market is unmistakable, and Oxmiq is aiming at exactly the right gap.

Primary sources

Original analysis by GenZTech. Based on Oxmiq's Series A announcement, July 1, 2026.