Mueon, Developer of AI Infrastructure and 3D Cubelet Architecture, Raises $65 Million

As the parameter counts and context windows of generative AI models grow, the biggest bottleneck in data centers no longer comes from the raw compute power of chips alone; it comes from the cost of moving data between processors and memory units and from enormous energy consumption. Components laid out side by side on traditional two-dimensional (2D) server motherboards have hit the limits of speed and thermal efficiency that modern AI workloads demand. Mueon, a US-based semiconductor and AI infrastructure startup that is redesigning data center architecture from scratch to overcome these fundamental physical limits, announced that it has raised $65 million in a Series A round.

Total Funding Tops $80 Million in a Round Led by Maverick Silicon

The $65 million Series A round of Portland, Oregon-based Mueon was led by Maverick Silicon, known for its semiconductor-focused investments. As part of the deal, Maverick Silicon Managing Partner Manish Muthal also joined Mueon's board of directors.

Other strategic and financial investors in the round included strong funds such as Intel Capital, Geodesic Alliances Fund, Oregon Venture Fund, M-Ventures, Wellington Capital, IQT (In-Q-Tel) and Defined. Together with its earlier $15.5 million seed round, the total capital Mueon has raised to date now stands at more than $80 million.

3D Packaging Expertise From Founders With Intel Roots

Behind Mueon, founded in 2025, lies deep engineering experience spanning decades in the semiconductor industry. The company's co-founders Wilfred Gomes (CEO) and Jack Hwang (CTO) spent a large part of their careers at Intel developing advanced microprocessor architectures.

Wilfred Gomes in particular is known as someone who, during his time as an Intel Fellow, played a key role in the integration of logic and memory units and in the development of Foveros three-dimensional (3D) chip packaging technology. Jack Hwang, as a Principal Engineer, led silicon manufacturing, advanced packaging and system integration processes. The pair founded Mueon to take the logic of 3D vertical integration inside the chip beyond a single microprocessor and bring it to the scale of the entire server and data center.

Farewell to Traditional Motherboards: What Is the “Cubelets” Architecture?

In the traditional data center servers used for decades, processors (CPU/GPU), memory modules, power regulators and networking components sit side by side on a wide, flat motherboard. In AI computation, however, terabytes of data traveling centimeters along the copper data buses on the motherboard cause both serious latency and the loss of a large share of energy as heat.

To eliminate this structural inefficiency, Mueon is developing a three-dimensional, modular hardware architecture it calls “Cubelets”:

• Vertical (3D) Stacking and Short Data Distance: Instead of spreading compute and memory units across a horizontal plane, the Cubelets architecture stacks them vertically on top of each other. This shrinks the physical distance between processor and memory to the millimeter level, multiplying data transfer speed and radically cutting the energy spent per communication.
• Integrated Power Delivery and Thermal Management: Heat and power delivery, the biggest challenge of dense 3D structures, are handled by innovative cooling channels and vertical power delivery layers integrated directly into the Cubelet modules.
• A Target of Up to 10x Efficiency: According to engineering projections shared by the company, Cubelets-based systems aim to deliver up to 10x improvement in energy consumption, space usage and total cost of ownership compared with traditional data center racks.

Post-Funding Roadmap and Software Ecosystem

For a radical change in hardware architecture to gain traction in data centers, it is critical that it works seamlessly with existing AI models and software frameworks. Mueon will put its new $65 million in funding toward three main focus areas:

1. Moving to Production and System Validation: Completing the physical design of the Cubelet platform, running system validation tests and making it ready for mass production at commercial scale. 2. Building the Software Stack: Completing the compiler and orchestration software that will let customers run PyTorch, TensorFlow and common large language model (LLM) inference/training software directly on Cubelets hardware without rewriting their code. 3. Growing the Team: Expanding the engineering staff with experts in hardware design, thermal engineering and systems software.

At a time when the load AI data centers put on global power grids is under debate, the 3D modular approach Mueon offers has the potential to set a new standard for the hyperscale computing infrastructure of the future.