July 2, 2026: Artificial intelligence is advancing at an unprecedented pace, but its biggest constraint is no longer smarter models, it is compute. As demand for AI infrastructure surges, building custom silicon remains prohibitively expensive and largely restricted to a handful of technology giants. OXMIQ believes that has to change. By developing an open, licensable AI compute architecture, the company aims to make custom AI silicon accessible, helping businesses build faster, more efficient and cost-effective AI systems without starting from scratch.
Every breakthrough from ChatGPT to autonomous robots and industrial AI agents, depends on enormous computing power. Yet the infrastructure powering these systems remains concentrated among a handful of GPU vendors. That has made AI expensive to build, expensive to scale, and difficult to customize.
For most companies, creating AI-specific silicon is simply out of reach. Building a custom chip can cost hundreds of millions of dollars and take years of engineering. As a result, even organizations with unique AI workloads often settle for hardware that was never designed for their specific needs.
This is the problem OXMIQ believes it can solve.
The company, founded by semiconductor veteran Raja Koduri, has raised $35 million in Series A funding to build what could become a foundational layer of the next AI infrastructure era: open, licensable AI compute.
Making AI Compute Modular
Think about building a house.
Today, if you wanted a house designed exactly for your family, you would first need to build an entire cement factory, brick factory and steel plant before laying the first foundation.
That is how AI chip development works today.
OXMIQ wants to change that by offering the essential building blocks instead of forcing every company to start from scratch.
Instead of designing an entire processor, companies can license OXMIQ’s architecture and configure it for their own workloads, dramatically reducing both cost and development time.
In other words, OXMIQ wants to make custom AI compute as accessible as cloud computing made servers.
OXMIQ Is Building the Technology That Could Fix AI’s Compute Bottleneck
At the center of this strategy is OxCore, a new GPU architecture that combines three compute engines into a single scalable core.
The first is a CUDA-compatible GPU engine, allowing developers to run existing GPU workloads without rewriting their software.
The second is a Tensor Processing Engine, purpose-built for the matrix calculations that power modern AI models and inference.
The third is an Orchestration Engine (CPU), which coordinates workloads, manages AI agents and efficiently distributes tasks across the system.
Traditionally, these capabilities are spread across multiple chips that constantly exchange data, increasing latency, power consumption and cost.
OXMIQ integrates them into one tightly coupled architecture.
The result is less data movement, higher efficiency and lower energy consumption—three challenges that become increasingly important as AI workloads continue to grow.
For enterprises running thousands of AI agents or factories processing billions of AI tokens every day, those efficiency gains could translate into significant cost savings.
Beyond the GPU
OXMIQ is not trying to become another chip manufacturer.
Instead, it is building an open architecture that semiconductor companies, cloud providers, robotics firms and AI infrastructure companies can adapt to their own products.
Its OxQuilt platform allows customers to mix different memory technologies, chiplets, packaging methods and manufacturing processes instead of locking themselves into a single supply chain.
Its software stack—including OxPython and OxCapsule—ensures developers can continue running familiar CUDA and PyTorch applications without rewriting code, removing one of the biggest barriers to adopting new hardware.
Why Investors Are Paying Attention
The company’s investors are betting that AI’s next competitive advantage will come from infrastructure rather than models alone.
Rajeev Surati, Partner at Fundomo, explained the investment thesis simply:
“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. We backed this team because they will define how AI compute gets built this decade.”
Samsung Catalyst Fund echoed that view.
David Goldschmidt, who leads the fund, said OXMIQ’s architecture enables “efficient, custom inference solutions serving large-scale agentic workloads.”
The addition of legendary chip architect Jim Keller to OXMIQ’s board further reinforces industry confidence in the company’s direction.
According to Keller:
“As the industry concentrates around a few incumbents, this is more important than ever. OXMIQ’s open, configurable foundation, which developers can build on and own, is exactly where compute should be heading.”
Raja Koduri’s Bigger Vision
For Raja Koduri, this is about more than building another GPU.
It is about making AI infrastructure accessible.
As he explains:
“A licensable core with an open architecture means design teams everywhere can build the custom AI silicon their work needs. Today, state-of-the-art AI reaches most people through a handful of channels, and the cost of the compute underneath is the reason. Bring that cost down, and you widen who gets to build with it.”
That vision reflects one of AI’s biggest structural challenges today.
The future of AI will not depend solely on who builds the smartest models.
It will increasingly depend on who can build the most efficient infrastructure beneath them.
The Bigger Picture
Over the past decade, cloud computing democratized access to servers. Companies no longer needed to own data centers to build global software businesses.
OXMIQ is attempting something similar for AI hardware.
If successful, custom AI silicon may no longer remain the privilege of only the world’s largest technology companies. Startups, robotics builders, sovereign AI initiatives and enterprise infrastructure providers could all design compute tailored to their own workloads without embarking on billion-dollar chip programs.
As AI becomes embedded across every industry, the winners may not simply be those building better models, but those making compute cheaper, more flexible and available to everyone.
That is the opportunity OXMIQ is betting on, and one that could reshape how AI infrastructure is built over the coming decade.
