Open-source software transformed how developers build applications. Open networking standards changed how enterprises design infrastructure. Open hardware architecture has the potential to do the same for AI compute — but only if someone builds the software bridge that makes it usable. That is precisely the gap Raja Koduri, founder of Oxmiq Labs and one of the most experienced GPU architects in the industry, has positioned himself to close.
The Education Behind the Engineering
Raja Koduri‘s path into GPU architecture began at two of India’s most rigorous technical institutions. He completed his undergraduate degree in electronics and communications engineering at Andhra University before earning a Master of Technology from the Indian Institute of Technology (IIT) Kharagpur — an institution consistently ranked among Asia’s most selective engineering programs.
That foundation in electrical and systems engineering did not simply prepare Koduri for a career in hardware design. It shaped a mode of analysis — one that begins with first principles, evaluates systems in terms of their dependencies and constraints, and identifies failure points not in individual components but in the architecture that connects them. That analytical orientation is visible across every major phase of his professional career, and it is especially evident in the problem Oxmiq Labs is structured to solve.
Building GPUs at Scale: The ATI-to-Intel Arc
Koduri began his industry career at ATI Technologies, which at the time was among the most consequential graphics hardware companies in the world. When ATI was acquired by Advanced Micro Devices, Koduri’s role expanded: he eventually served as Senior Vice President and Chief Architect of the Radeon Technologies Group, leading GPU architecture development across multiple product generations.
The Radeon work gave Koduri direct, sustained experience with what it means to compete against a dominant incumbent in a market shaped as much by software ecosystems as by silicon performance. AMD’s discrete GPU program was technically credible — but competing against NVIDIA required confronting the reality that performance specifications alone do not determine developer adoption. The software platform does.
A move to Apple followed, where Koduri contributed to graphics engineering at one of the world’s most tightly integrated hardware-software organizations. Apple’s approach — designing hardware and software in parallel, treating them as a single system — offered a different lens on the same underlying question: how does a hardware architecture achieve relevance in a developer ecosystem?
At Intel, as Chief Architect and Executive Vice President of the Architecture, Graphics and Software (IAGS) division, Koduri led the company’s most ambitious discrete GPU initiative in decades. The program involved building a new GPU architecture, assembling a new engineering organization, and attempting to establish a software developer ecosystem from a standing start — all while competing in a market where one player had spent years accumulating software lock-in. The experience was instructive in ways that cannot be replicated at a distance.
How GPU Lock-In Actually Works
Understanding why Oxmiq Labs exists requires understanding how CUDA’s dominance was constructed and why it has proved so durable.
NVIDIA’s CUDA platform, launched in 2006, gave developers a way to write software that ran directly on NVIDIA GPU hardware using a C-based programming model. In the years since, the AI and machine learning developer community built an enormous body of code, tooling, and institutional knowledge on top of that platform. PyTorch, one of the dominant deep learning frameworks, was optimized for CUDA. TensorFlow’s primary GPU support path runs through CUDA. The Python-based AI ecosystem — the environment in which the vast majority of enterprise AI development occurs today — is architected around CUDA as a foundational dependency.
This did not happen because developers are loyal to NVIDIA as a hardware vendor. It happened because CUDA was first, was well-engineered, and accumulated a critical mass of libraries, documentation, and community expertise that made alternatives functionally inferior not in hardware performance terms, but in developer productivity terms. Switching to a non-CUDA hardware platform means rebuilding or replacing the software toolchain — a cost that most organizations find prohibitive in practice, regardless of the hardware’s technical merits.
Koduri observed this dynamic from the inside at AMD and Intel. Both organizations built capable GPU hardware. Neither was able to fully overcome the software ecosystem gap. That experience is the direct origin of Oxmiq Labs.
Oxmiq Labs: The Compatibility-First Thesis
Founded in 2023 and based in San Francisco, Oxmiq Labs is a GPU software and intellectual property startup with a precise and well-defined thesis: CUDA workloads should run on non-NVIDIA hardware without requiring developers to modify their code.
The company’s technical architecture uses RISC-V-based GPU design as its hardware foundation. RISC-V, an open-source instruction set architecture, gives Oxmiq the flexibility to build GPU hardware and software interfaces without the licensing constraints and architectural rigidities that come with proprietary instruction sets. The open foundation at the hardware layer supports the portability objective at the software layer.
The target workflow is specific: Python-based AI applications written for CUDA execution environments should run on Oxmiq-enabled hardware as written, without recompilation, without toolchain replacement, and without performance degradation that makes the migration commercially unattractive. The company is not attempting to build a better CUDA. It is building a compatibility infrastructure that removes CUDA’s role as a hardware prerequisite.
The distinction matters. A new proprietary platform competes against CUDA and asks developers to make a choice. A compatibility layer removes the choice — the developer’s existing code simply works on more hardware. That is a fundamentally different value proposition, and it is one that addresses the market failure more directly than any previous hardware-centric approach.
Advisory Reach and Industry Positioning
Beyond his operational role at Oxmiq Labs, Raja Koduri serves in advisory and board capacities for leading semiconductor and AI companies. That network of relationships reflects both the depth of his technical reputation and the breadth of his industry experience.
For an organization attempting to establish a new software compatibility standard in the GPU ecosystem, advisory relationships with semiconductor manufacturers, cloud providers, and AI platform developers are not peripheral — they are part of the product. Ecosystem adoption requires trust, technical credibility, and the organizational relationships to convert interest into integration. Koduri’s professional history provides those assets in a way that few founders in this space can replicate.
His background also brings credibility with the developer community specifically. Engineers who have worked with Radeon architectures, who have followed Intel’s discrete GPU program, or who have studied GPU architecture at a technical level know Koduri’s work. That recognition reduces the friction of establishing Oxmiq’s technical legitimacy in a market where credibility is evaluated rigorously.
What Open GPU Architecture Means for AI’s Trajectory
The stakes of the problem Oxmiq is addressing extend beyond market competition between GPU vendors. AI infrastructure concentration in a single hardware ecosystem creates systemic fragility — in supply chains, in pricing, in the ability of enterprises and governments to build AI programs without a single-vendor dependency.
The open-source movement demonstrated across multiple technology layers that proprietary lock-in eventually yields to well-engineered open alternatives — but only when those alternatives address the actual source of lock-in rather than the symptoms. In Linux’s case, the lock-in was in operating system licensing. In the GPU compute case, the lock-in is in the software programming model. Oxmiq’s RISC-V-based, CUDA-compatible approach is structured precisely around that insight.
Raja Koduri has spent more than two decades inside the organizations where this problem was most clearly visible and most consequentially felt. Oxmiq Labs is the direct application of that accumulated understanding to a market structure that has remained largely unchanged — and that the broader AI industry has a significant interest in seeing evolve.
About Raja Koduri
Raja Koduri is an Indian-American computer engineer, technology executive, and founder with more than two decades of experience in GPU architecture and computing platform development. He holds a bachelor’s degree in electronics and communications from Andhra University and a Master of Technology from the Indian Institute of Technology (IIT) Kharagpur. Koduri has held senior roles at ATI Technologies, Advanced Micro Devices (AMD), Apple, and Intel, where he served as Chief Architect and Executive Vice President of the Architecture, Graphics and Software division. In 2023, he founded Oxmiq Labs Inc., a San Francisco-based GPU software and IP startup focused on enabling CUDA workloads to run on non-NVIDIA hardware through RISC-V-based designs and open software frameworks. He also serves in advisory and board capacities for leading semiconductor and AI companies.