Oxmiq Labs raises $35M to license custom AI chip architecture
Oxmiq Labs has raised $35 million in a Series A funding round to develop a licensable architecture aimed at reducing the cost and complexity of designing custom AI silicon. Led by former Intel executive Raja Koduri, the startup plans to offer an integrated IP block featuring GPU, CPU, and tensor components to help systems builders create custom chips more efficiently.
Key Takeaways
- Series A round was co-led by Fundomo and Samsung Catalyst Fund, bringing total financing to $60 million.
- The OxCore architecture combines GPU, CPU, and a tensor engine into a single licensable IP block to minimize data movement.
- OxPython software enables existing CUDA and PyTorch code to run on OxCore without modifications, lowering software barriers.
- OxQuilt computing fabric allows designers to mix heterogeneous chiplets and memory types across different foundries and interconnect standards.
- Tenstorrent CEO Jim Keller has joined Oxmiq’s board of directors as the company scales its engineering team.
Why It Matters
Oxmiq’s licensing model directly challenges the high-cost, multi-year barrier of custom silicon design, currently dominated by Broadcom and Marvell. By bundling compute and software compatibility in a single core, the startup provides a faster path for neoclouds and robotics firms to build proprietary chips. For the streaming ecosystem, this could lower the CapEx required for high-density video encoding and AI-driven personalization, shifting the market toward heterogeneous edge and data center architectures. Watch for the first commercial tape-outs using the OxCore IP to validate claims of lower non-recurring engineering costs.
Additional Context
The push for custom silicon comes as the total cost of ownership for merchant GPUs reaches unsustainable levels for large-scale operators. Per Tom’s Hardware in May 2026, custom ASIC shipments are projected to grow 44.6% year-over-year, nearly triple the growth rate of merchant GPUs. Broadcom and Marvell currently control approximately 95% of the custom AI ASIC co-design market, with Broadcom reporting record quarterly AI revenue of $10.8 billion in mid-2026. This dominance is built on multi-year co-design partnerships with hyperscalers like Google, whose TPU v6e Trillium reportedly offers significantly better price-performance for specific inference workloads compared to off-the-shelf competitors. Oxmiq is entering a landscape where advanced packaging and interconnects have become the primary bottlenecks for performance. According to SemiEngineering in February 2026, leading-edge mask sets now cost between $30 million and $50 million, driving the industry toward chiplet-based strategies to improve yields and SKU flexibility. In parallel, the transition to optical I/O is accelerating. Per Reuters in June 2026, NVIDIA secured a multi-year deal with Lumentum to integrate silicon photonics directly into its AI infrastructure, responding to physical limits on copper-based electrical interconnects. By designing OxCore to support silicon photonics, Oxmiq is positioning its architecture for a market where power efficiency and data density are the ultimate competitive measures.
Read full article at siliconangle.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source