U.S. lines up $389 million for power chips linking fusion and AI data centers

The story

The U.S. Commerce Department has signed four letters of intent offering up to $389 million for research into power semiconductors and critical chipmaking materials. The proposed CHIPS and Science Act awards connect technologies that are usually discussed separately: fusion power systems, artificial-intelligence data centers and ultra-high-purity quartz. Their common constraint is controlling large flows of electricity efficiently and manufacturing the specialized components that make that possible.

The largest proposal is up to $175 million for Helion. The Washington-based fusion company plans to develop high-voltage semiconductor devices and advanced power modules for the pulsed electrical systems that drive its fusion generators. Those systems must switch, recover and redirect large bursts of energy repeatedly. Helion says more than $100 million would support semiconductor and packaging R&D in Everett, while roughly $80 million would fund tooling, process and design work with an industry partner in Burlington, Vermont.

Helion says the expected federal award would be accompanied by $50 million in private investment. Under a definitive agreement, Commerce would receive a $175 million non-voting interest valued against the company's latest preferred-stock round. That structure is unusual for a research incentive, but it does not make the government an operating owner: NIST describes the planned stakes across all four recipients as minority, non-controlling and non-voting.

Commonwealth Fusion Systems is in line for up to $85 million. Its proposed work covers semiconductors for radio-frequency plasma heating, power electronics, embedded fusion diagnostics and radiation-hardened components. The company also identifies fast controllers and electronics for robotic maintenance of future ARC plants. These are enabling technologies rather than a direct subsidy for a complete reactor, and the announcement provides no device specifications, production volume or timetable for commercial deployment.

The data-center component is a proposed award of up to $73 million for MIT spinout Vertical Semiconductor. The company is developing vertical gallium-nitride power transistors intended to reduce conversion losses and shrink electricity-delivery hardware. NIST says the program would demonstrate the technology in a 2.5-megawatt AI data-center rack. That is a target for R&D, not evidence that such a rack or the underlying transistors have already passed an independent commercial-scale test.

The remaining $56 million would go to Sibelco North America for ultra-high-purity quartz. Quartz components are used in crystal-growth crucibles, processing glassware and other hardware throughout semiconductor manufacturing. Sibelco says the project would research new leaching and hot-chlorination processes, product engineering and a North Carolina pilot and demonstration facility. Its letter of intent explicitly says a final award still depends on due diligence, negotiation and internal government approval.

Commerce frames the package around energy infrastructure. Better power conversion can lower thermal losses and improve the controllability of equipment ranging from grid systems to computing racks and fusion machines. Reuters independently confirmed the four proposed amounts and reported that the package follows 19 other semiconductor R&D grants from a $10 billion fund established by Congress in 2022. The four allocations sum exactly to the department's $389 million headline figure.

The announcement should not be read as $389 million already spent or as a forecast that commercial fusion is imminent. Letters of intent establish a negotiating framework; final awards may include technical milestones, reporting requirements and revised funding levels. The recipients' performance claims also remain company statements. None of the releases supplies independent efficiency measurements, manufacturing yields, qualification results or a date when funded components will reach customers.

INNOVOX analysis: the package shifts semiconductor policy toward the electrical layer surrounding advanced computing and energy systems. Leading-edge processors attract attention, but every AI rack and fusion experiment also depends on switches, converters, packaging, controllers and process materials. Funding four points in that chain could create shared manufacturing capabilities across markets. The test is whether public R&D money produces components that outside customers can buy, rather than bespoke hardware tied to one developer's roadmap.

What to watch next is the conversion of these letters into definitive awards. Commerce and the companies should disclose milestones, private cost sharing, equity terms and the point at which funding can be withheld. Technical evidence matters just as much: voltage, efficiency, thermal performance, radiation tolerance, production yield and lifecycle reliability. A credible outcome would be repeatable domestic processes and independently validated devices; a signed letter, by itself, is only the beginning of that path.

INNOVOX analysis

The package treats power conversion and materials as strategic infrastructure rather than secondary components. Its breadth is the signal: the same policy instrument is being used to tackle electricity control inside fusion machines, megawatt-scale computing racks and the upstream quartz needed to manufacture chips. The technical promise is real, but the announcement measures government intent—not device efficiency, production yield or commercial deployment.

What to watch

Watch for definitive award agreements, milestone and cost-sharing terms, the exact government equity stakes, named manufacturing partners, independently measured device performance and evidence that the funded processes move from laboratory development into repeatable U.S. production.