Raja Koduri's Oxmiq raises $35M to license AI chip designs instead of selling finished chips

Oxmiq Secures $35 Million Series A Funding to Scale Licensable GPU Architecture

Oxmiq Labs has closed a $35 million Series A funding round to scale its OxCore architecture, bringing the company's total capital raised to $60 million since its founding by veteran chip architect Raja Koduri 1

1

. The round was co-led by Fundomo and Samsung Catalyst Fund, with participation from MediaTek, AM Intelligence Labs, Pegatron Venture Capital, CDIB-TEN, Darwin Ventures, and Morgan Creek Digital 2

2

. This mix of investors, spanning memory and foundry-adjacent giants, contract manufacturers, and fabless chipmakers, signals interest that extends beyond pure financial returns.

Raja Koduri's Vision to Lower the Design Cost of Custom AI Silicon

Raja Koduri's Oxmiq Labs is tackling a fundamental barrier in AI chip design: the prohibitive cost and time required to build advanced AI chips from scratch. Developing a cutting-edge AI chip typically costs hundreds of millions of dollars and requires several years, keeping custom AI silicon out of reach for all but the largest cloud providers and chipmakers 1

1

. Koduri, whose background includes leading graphics at AMD, Apple, and Intel's graphics and accelerated-computing efforts, told Reuters that his company wants to "be the Arm of this next era" 2

2

. Just as Arm Holdings licenses CPU designs that power nearly every smartphone chip globally, Oxmiq aims to license GPU intellectual property through its flagship product OxCore, enabling customers to build custom AI processors without investing billions in developing their own silicon.

OxCore Architecture: Three Engines in One Licensable Core

At the center of Oxmiq's approach is OxCore, a licensable GPU architecture that integrates three distinct compute engines into a single core: a CUDA-compatible GPU engine, a tensor processing engine, and an orchestration engine that coordinates workloads across the system 5

5

. This tightly coupled design handles functions typically split across separate chips and was purpose-built for near-memory compute, reducing the data movement that drives up both cost and energy use in AI workloads 1

1

. The architecture scales efficiently from single-core AI deployments to large-scale datacenter configurations, and OxCore is currently running on FPGA with live demonstrations available 5

5

.

Chiplet Integration and Supply Chain Flexibility Through OxQuilt

Oxmiq is also developing OxQuilt, a chiplet integration architecture that combines heterogeneous compute chiplets and memory into a single package 2

2

. Unlike most AI silicon designs that lock customers to a specific foundry and memory type, OxQuilt adapts to any supply chain with configuration tools that let customers mix and match different process nodes, memory types, interconnect standards, and advanced packaging systems 5

5

. The architecture is designed to incorporate emerging interconnect technologies such as silicon photonics when they reach production readiness. This flexibility addresses a critical pain point, as Fundomo partner Rajeev Surati noted: "Most compute IP makes the customer bend their memory, packaging, and foundry around the chip. Oxmiq does the opposite, and that flips a cost center into leverage" 2

2

.

Jim Keller Joins Board as Oxmiq Targets Sovereign AI Demand

Perhaps the most significant validation of Oxmiq's approach is the addition of Jim Keller to its board of directors. Keller, CEO of Tenstorrent and one of the most influential chip architects in the industry, brings a career spanning Apple, AMD's Zen architecture, and Tesla's self-driving silicon 1

1

. Former Intel process technology fellow Valluri "Bob" Rao has also joined as an advisor 5

5

. The company, founded in 2024 and headquartered in Campbell, California, with engineering teams in Hyderabad and Bengaluru, is targeting semiconductor firms, neocloud providers, AI compute infrastructure builders, and governments pursuing sovereign AI programmes 3

3

.

Early Traction and Focus on Emerging Markets

Oxmiq launched a public beta of its software stack in November 2025 and reports it is now being used by 20 companies and 10 universities across nearly 300 GPUs 3

3

. The software stack includes OxCapsule for orchestration and OxPython, which runs existing CUDA and PyTorch code on OxCore without code changes, giving developers full portability across hardware 5

5

. Koduri told ET that the company's initial commercial focus is India and Southeast Asia, including Malaysia and Indonesia, where AI data centre expansion is driving demand for alternatives to Nvidia-led infrastructure 3

3

. He argued that "India cannot play in AI without taking control of the cost of it," noting that AI compute costs would need to fall by 50-100 times to drive mass adoption, and that chip design is where 92% of margins lie 4

4

.

Strategic Partnership Signals Renewable-Powered AI Compute Ambitions

Earlier this year, Oxmiq partnered with AM Intelligence Labs, part of the AM Green Group, to architect a 2 GW renewable-powered AI compute platform in Uttar Pradesh, with the first 1 GW phase expected to go live by the end of 2027 3

3

. This partnership demonstrates the company's ambition to address AI infrastructure from renewable power through to silicon IP. The Series A funding will be used to complete OxCore by the end of this year or early next year before customer deployments and commercial silicon tape-outs 4

4

. Whether Oxmiq's bet pays off will depend on whether any of its backers and customers actually put OxCore into silicon that ships, but the $35 million buys the company runway and credibility in a sector where trust in an unproven architecture remains the scarcest resource 1

1

.