The financing was co-led by Fundomo and the Samsung Catalyst Fund, with participation from a roster of strategic and financial backers including MediaTek, AM Intelligence Labs, Pegatron Venture Capital, CDIB-TEN, Darwin Ventures, and Morgan Creek Digital. The presence of MediaTek and Pegatron — companies that design and assemble hardware at scale — signals that OXMIQ's pitch is landing with the exact customers it needs to reach. For more on where this fits in the broader AI industry coverage, the funding lands amid an intensifying race to break Nvidia's grip on AI compute.
Selling blueprints, not chips
OXMIQ의 핵심은 확장 가능하고 라이센스가 가능한 GPU 코어인 OxCore라는 제품입니다. Rather than manufacture its own silicon, OXMIQ plans to license OxCore to semiconductor companies and AI infrastructure builders so they can create custom AI accelerators without launching a full in-house chip program — a notoriously expensive, multi-year undertaking.
OxCore integrates three compute engines into a single core: a CUDA-compatible GPU engine, a tensor processing engine, and an orchestration engine that coordinates workloads and AI agents across the system. OXMIQ says those functions are typically spread across three separate chips today, and that combining them enables near-memory compute designs that reduce the costly movement of data and improve energy efficiency — a critical metric as data center power consumption balloons.
The architecture scales from single-core edge deployments up to large-scale data center configurations, according to the company, and is currently running on FPGA hardware with live demonstrations.
일치하는 소프트웨어 스택
하드웨어는 AI 가속기의 절반에 불과합니다. 다른 하나는 기존 코드가 실제로 새로운 아키텍처에서 실행되도록 하는 것입니다. 이를 해결하기 위해 OXMIQ는 동반 소프트웨어 스택을 구축했습니다. OxPython is designed to let existing CUDA and PyTorch code run on OxCore without code changes, giving developers portability across hardware — an attempt to neutralize the largest moat Nvidia enjoys, its CUDA software ecosystem. OXMIQ는 OxPython이 타사 플랫폼에서 검증되었다고 말합니다.
The company has also developed OxQuilt, a chiplet integration architecture that combines heterogeneous compute chiplets and memory in a single package. Where many AI silicon designs are locked to a specific foundry and memory type, OXMIQ says OxQuilt is built to adapt across logic process nodes, memory types, interconnect standards, and advanced packaging options, letting customers design across whichever supply chain is available. A high-level orchestration layer called OxCapsule rounds out the stack, handling everything from agent orchestration to low-level kernel optimization.
야심 찬 논문 뒤에 숨겨진 친숙한 이름
Koduri는 GPU 설계 분야에서 가장 잘 알려진 인물 중 하나입니다. He previously served as a senior vice president at AMD leading the Radeon Technologies Group, had a stint at Apple, and later joined Intel as chief architect of accelerated computing and graphics before departing in 2023 to pursue an AI-focused venture. 그의 실적은 회사의 표현에 따르면 "Atoms에서 Agents까지" GPU 스택을 재설계하려는 OXMIQ의 야심찬 임무에 신뢰성을 부여합니다.
That mission is pitched as a top-to-bottom rethinking of the AI compute stack — spanning renewable power, data center infrastructure, silicon IP, and software — with the goal of lowering the cost of intelligence across every layer.
Why the licensing model matters now
타이밍은 의도적입니다. As the largest AI labs and cloud providers scramble for compute, a growing list of companies — including Amazon, Google, and Meta — have begun designing their own AI accelerators to reduce dependence on Nvidia. But only the largest players can afford the billions required to build a competitive chip from scratch. By licensing a proven, modular architecture instead, OXMIQ is targeting the much larger universe of companies that want custom AI silicon but lack Nvidia-scale budgets.
It is, in essence, the Arm model applied to AI GPUs: sell the design, let partners fabricate and customize it, and capture value across a fragmented customer base rather than fighting Nvidia head-on in the merchant-chip market. Whether OXMIQ can deliver production-quality silicon IP that matches Nvidia's performance — and convince enough partners to adopt OxCore before larger rivals cement their own custom-silicon strategies — will determine whether Koduri's latest venture becomes a foundational layer of the AI hardware stack or a niche alternative.
For now, the $35 million vote of confidence from Samsung, MediaTek, and a syndicate of deep-tech investors suggests the industry sees real promise in disaggregating the GPU.
---
AI 최신 소식최신 AI 뉴스, 분석 및 혁신을 한 곳에서 확인하세요.
AI 뉴스 더 보기 →
