The $389 Million Power-Chip Award Is 77% of Everything Washington Has Actually Spent on Chip R&D

Commerce disclosed the awards as letters of intent, not definitive grants, and every recipient gives up equity to get the money. Each of the four companies will issue the government a minority, non-controlling, non-voting stake as a condition of the award. Helion's is sized at $175 million, struck against its most recent Series G preferred valuation, and it converts only when a definitive agreement is signed. Cost share is disclosed for exactly one award: Helion paired its $175 million with $50 million of private investment over a three-year award period.

The August 2026 audit of the program is the context every investor needs. Of the $11 billion appropriated for advanced microelectronics R&D in the 2022 CHIPS Act, Commerce canceled awards worth $7.8 billion: the National Semiconductor Technology Center award was terminated in 2025, advanced packaging awards were canceled or paused, and the industrial advisory committee charter was allowed to lapse. After those cancellations the department had spent $506.8 million — about 5% of the appropriation. The October 22 award of $389 million therefore equals roughly 77% of the program's entire spending to date, and only 3.5% of what Congress actually gave it.

The diagnosis is measurable. Power transformer lead times that ran 6 to 12 months before 2020 now run 24 to 48 months, with prices up 50% to 80%, and large units take more than 36 months; generators carry two-year lead times and high-voltage transformers 12 to 18 months. Substation transformer lead times stretched from roughly 140 weeks in 2023 to more than 160 weeks in 2026, and annual transformer demand is projected to pass 9,000 units by 2030 against roughly 1,500 today. This is the layer of the AI buildout with no spare capacity.

The industry has already rewritten the architecture around that constraint. NVIDIA is moving data centers to 800-volt direct current from 2027 to support racks of 1 MW and above, with named semiconductor suppliers including Texas Instruments, STMicroelectronics, ROHM, Monolithic Power Systems and Navitas, and named power-system suppliers including Eaton and Vertiv. The Open Compute Project's solid-state transformer specification reached release 0.3 in July 2026, and more than 80 equipment makers and infrastructure providers are already building to it. The megawatt rack gets decided in the conversion layer, not in the chip.

None of the four awardees is listed. But roughly $80 million of Helion's award is earmarked for semiconductor tooling, design and process development in Burlington, Vermont, in collaboration with GlobalFoundries — which directs roughly $80 million of Helion's award into GlobalFoundries' Vermont tooling program and makes the foundry the only listed company whose income statement this announcement directly funds. Helion's chief executive frames the prize in efficiency terms: the power electronics in the fusion path lose roughly 5% to 10% of the electricity flowing through them, and halving that loss is equivalent to doubling the output of the system.

GlobalFoundries has also bought its way into the same conversion layer from the commercial side. Its acquisition of Photeon Technologies' integrated voltage regulator business gives it power-delivery products aimed at AI data centers, and it reported second-quarter 2026 revenue of $1.786 billion at a 28.3% gross margin, with growth led by optical networking inside AI data centers. For a company whose economics turn on utilization, federal R&D contracts that fill tooling and process-development capacity are a better use of a fab than another logic wafer.

The materials layer is where the award is most revealing. Sibelco's $56 million targets next-generation purification and finishing of ultra-high-purity quartz — the material for the crucibles and process glassware used to grow silicon ingots and compound-semiconductor crystals. Spruce Pine in North Carolina has historically supplied more than 90% of the world's high-purity quartz sand, and Sibelco has spent $200 million there with another $500 million planned. The new leaching process is to be piloted in North Carolina.

Read the two materials awards together and the pattern is clear. Washington is funding purification of the quartz that silicon is grown from, and the power transistor that gallium nitride is made into — but not the gallium, and not the gallium-nitride substrates. A research program that pays for the device while leaving the metal and the substrate untouched relocates the bottleneck rather than removing it.

The downstream names are the ones already collecting revenue from the transition. Vertiv converted the demand most visibly: second-quarter 2026 revenue of $3.274 billion was 23.6% above the prior quarter, with gross profit of $1.235 billion and operating income of $638 million, a 37.7% gross margin. Eaton reported $8.531 billion of quarterly revenue, $2.855 billion of gross profit and $1.392 billion of operating income, and has been developing DC-distribution semiconductor transformers for the same architecture.

Washington's bottleneck diagnosis is right, and its portfolio is narrow. The AI power constraint is measurable in transformer lead times and in the gross-margin expansion at the power-chip vendors, and the money is landing on the least-substituted nodes in that chain: high-voltage pulsed-power devices, vertical power transistors, and the purification step for the hardest-to-make quartz. What the portfolio omits is as informative as what it contains.