Washington Takes an Equity Stake in a Two-Year-Old MIT Chip Startup to Fix AI's Power Problem
The Department of Commerce signed a letter of intent last week for up to $73 million in CHIPS and Science Act funding for Vertical Semiconductor, an MIT spin-out that has existed for about two years and raised its first institutional money, an $11 million seed round, just twelve months ago. In exchange, Commerce would take a minority, non-voting equity stake in the company. The deal is one of four announced together, totaling $389 million, with the other three going to fusion-power ventures Helion and Commonwealth Fusion Systems and to quartz supplier Sibelco. Only Vertical Semiconductor's award is pitched squarely at data centers.
The pitch is straightforward: GPUs don't just need more silicon, they need power delivered to them more efficiently, and conventional silicon power transistors are increasingly the weak link. Vertical Semiconductor builds gallium-nitride transistors in a 3D, "vertical" structure rather than the lateral layout used in most commercial GaN chips today, which the company says lets voltage capacity scale up without a bigger chip footprint. CEO Cynthia Liao has framed power, not compute, as the real bottleneck constraining how much AI infrastructure the US can build within existing grid limits — an argument that tracks with what grid operators and hyperscalers have been saying for over a year.
From $11 million to $73 million, with no product yet
What's notable is the pace, not just the sum. A company with no shipped product and a single seed round scaled by 6.6x into a federal award in a year, on the strength of a stated goal rather than a demonstrated one: a planned "2.5-megawatt AI rack" — a target design, not a built prototype — that the company says will cut energy losses by 30% and shrink footprint by 50% once developed. Neither figure comes with a disclosed baseline — 30% less than what, measured how, under what load — and no outlet covering the award has published independent voltage, current, or switching specifications for the transistors themselves. That's normal for a letter of intent, which is a non-binding statement of interest, not a disbursed grant with milestones attached. It's also exactly the gap between announcement and delivery that has characterized a lot of CHIPS Act R&D money this year, including an $874 million batch of letters of intent to seven companies announced in July that has produced little public evidence of installed capacity since.
Equity, not just a subsidy
The more significant shift is structural. The structural shift matters more than the dollar figure. When the Commerce Department took an equity stake in Intel last year, it drew bipartisan criticism as a step toward industrial policy by ownership rather than contract, and officials downplayed the idea that it would become a pattern. A minority, non-voting stake in a two-year-old startup with no shipped product suggests that pattern is continuing rather than fading. A minority, non-voting stake in a company this small carries little governance weight, but it normalizes the mechanism: federal R&D money increasingly arrives with Washington on the cap table, which changes who benefits if the bet pays off and raises the obvious question of what happens to that stake, and to the company's incentives, if it doesn't.
The deeper issue the award doesn't address is scale. Even a flawless 2.5-megawatt rack is a rack. The constraint hyperscalers are actually running into is gigawatts of grid interconnection and multi-year utility queues — a problem that better transistors inside the cabinet don't touch. Framing a power-electronics R&D bet as strengthening "US leadership in AI power infrastructure," as both the company and Commerce did in announcing it, borrows the scale of the real crisis to describe a fix for a much narrower slice of it. That doesn't make the technology worthless — more efficient power conversion at the chip level is a genuine, if marginal, lever — but it's a sliver of the buildout problem wearing the language of the whole thing.