TL;DR · 30-second read
The Short Version
A young company in Portland, Oregon, called Mueon has raised $65 million from investors, including the investment arm of chipmaker Intel.
Its argument: computers that run artificial intelligence waste a lot of electricity moving information between parts that sit too far apart. Mueon wants to pack those parts into small stacked blocks so the data has less distance to travel.
It says this could make the same floor space up to ten times cheaper and more power-efficient to run. That is a goal, not a result yet, and the company has not named any customers.
Mueon Corporation, a Portland, Oregon semiconductor systems company, announced on October 15, 2026 that it has raised $65 million in a Series A round led by Maverick Silicon, with participation from Intel Capital, Geodesic Alliances Fund, Oregon Venture Fund, M-Ventures, Wellington Capital, IQT and Defined. The round follows a $15.5 million seed in 2025, when the company emerged from stealth, and brings total funding to more than $80 million.
The money will expand Mueon’s engineering and operations teams and move its Cubelets platform — modular, three-dimensional blocks that combine compute and memory — from development toward system validation and production readiness. Manish Muthal of Maverick Silicon and Srini Ananth of Intel Capital join the board.
Executive Summary
Mueon is not pitching a new AI chip. It is pitching a new way to physically assemble the parts around one. Cubelets stack compute and memory into compact 3D blocks and treat packaging (how chips are assembled and wired together), interconnect, power delivery and cooling as a single design problem. The company’s stated target is up to 10X gains in power and cost efficiency within the same rack footprint, with the architecture designed to scale toward megawatt-class density.
The significance is in where the efficiency is supposed to come from. Mueon argues that data movement, power density and deployment constraints increasingly define AI infrastructure economics, and that shortening the distance data travels cuts the power spent moving it. Endorsements from accelerator developer d-Matrix and from Ampere Computing, now part of SoftBank, frame Cubelets as a platform other chipmakers could build on.
The caution is equally clear: the 10X figure is a target with no published baseline, customers are unnamed, and the company describes itself as still moving toward production readiness. This round funds the proof, not the product.
Mueon’s Bet: The Power Bill Lives Between the Chips
The round’s headline number is a target: up to 10X gains in power and cost efficiency within the same rack footprint. (A rack is the standard cabinet that holds servers in a data center; its footprint is the floor space it occupies.) What makes the claim interesting is where Mueon locates the gain. The company does not promise a faster processor. It says Cubelets “densify compute at the system level, reducing the physical distance data has to travel and the power cost that comes with moving it,” and describes the architecture as “extending beyond any single accelerator technology.” An accelerator is the specialised chip, such as a GPU, that does the heavy mathematics in AI systems.
That positions Mueon one layer above the accelerator, in packaging, interconnect (the links that carry data between chips), power delivery and thermals — which it says it treats as “first-class design variables rather than separate problems to solve after the fact.” The d-Matrix endorsement fits this reading. Its co-founder and CTO, Sudeep Bhoja, describes combining Cubelets with d-Matrix’s in-memory compute (doing calculations where data is stored rather than fetching it) to build “smaller, denser infrastructure.” A chipmaker treating Mueon as a host for its silicon, not a rival to it, is the clearest evidence in the announcement that the bet is on data movement rather than raw chip speed.
Who would gain if the thesis holds? Operators who have hit what Ampere co-CEO Rohit Vidwans calls the “power and cooling limits” that have constrained large data centers for years, and accelerator vendors that could get a denser system without engineering it themselves. Who carries the risk? Everyone pricing the 10X figure before it is measured: Mueon gives no baseline system, workload or test method, so the number is best read as a design goal rather than a benchmark.
Investors Are Funding the System, Not Just the Silicon
Maverick Silicon led the round, and the board additions — Maverick’s Manish Muthal and Intel Capital’s Srini Ananth — give both lead and strategic investor a seat. Intel Capital’s stated rationale stresses integration rather than any chip: “bringing compute, memory, power delivery, and thermal management together into a single, integrated system.” Having the venture arm of a major chipmaker back a company whose differentiation is system-level assembly says something about where at least some sophisticated investors see room for improvement.
That supports the view that capital is flowing toward alternatives to conventional rack-scale designs, but one Series A is a data point, not a trend. Ampere’s statement that it is working with Mueon as the company “de-risks the technology on the path toward production” signals engineering engagement; neither Ampere nor d-Matrix discloses a purchase commitment or commercial agreement.
Production, Not Architecture, Is the Real Test
Mueon is plain that it is pre-production. The funds go to a transition “from the platform development phase towards broader system validation and production readiness,” and one of its four execution tracks, ecosystem validation, exists specifically to prove “the path from prototype to production” with manufacturing partners. More than $80 million raised to date pays for that proof; the company has not said what volume manufacturing will cost or who will carry it out.
Two engineering gates stand out. First, heat: stacking components in three dimensions concentrates power, and the goal of scaling “toward megawatt-class density” (a megawatt is a million watts) puts cooling and power delivery at the centre of the design rather than the edge. Second, software: Mueon says it is building a layer to make modular hardware usable “through familiar AI frameworks and deployment workflows.” New hardware architectures tend to win or lose on whether developers can run existing code without rewriting it, and that work is still in progress.
Background
Mueon Corporation is a Portland, Oregon semiconductor systems company that emerged from stealth in 2025 with a $15.5 million seed round. Its team brings experience across CPU and GPU architecture, memory systems, advanced packaging, high-speed I/O (input/output links), power delivery, thermal engineering and large-scale system design. The company organises its work into four tracks: silicon and system architecture, advanced integration, software and orchestration, and ecosystem validation with manufacturing and technology partners.
The company is entering a market in which large data centers, as Ampere’s Rohit Vidwans puts it, have been constrained by power and cooling limits for years. Most AI hardware today is built around accelerators mounted in conventional server racks; Mueon’s approach instead stacks compute and memory in three dimensions — placing components on top of one another rather than side by side — to increase how much computing fits in the same space. Source: Mueon Raises $65M in Series A Funding to Scale its Cubelets™ Exponential Computing Platform — Mueon Corporation’s announcement of its Maverick Silicon-led Series A and Cubelets production plans.Sources