The chip most people picture when they hear “AI” is a monolithic slab of silicon from Nvidia, bolted to a heatsink and burning through electricity at a rate that makes data center operators wince. The less glamorous story is the one happening around it: the power electronics that decide how that electricity gets delivered, converted, and cooled, and the materials that make those components possible. That’s where the US government just placed a $389 million bet.
The funding, reported by Reuters, targets research and development for power semiconductors and critical materials. On its face it’s a modest number next to the tens of billions flowing into fabs and advanced logic. But the strategic logic is hard to argue with. Every GPU in every AI cluster is useless without a power conversion chain that can feed it cleanly and efficiently, and that chain runs on chips most consumers never see.
Power semiconductors are the unglamorous workhorses of the electronics world. They manage voltage, switch current, and convert AC to DC and back again. Silicon carbide and gallium nitride, the wide-bandgap materials that have been creeping into electric vehicles and industrial gear, are the current frontier. They handle higher voltages and temperatures than traditional silicon, which means less waste heat and smaller, cheaper systems. Data centers care about this intensely. Power delivery losses compound across thousands of servers, and at the scale AI operators are building, a few percentage points of efficiency translate into real money and real constraints on how much compute you can physically cram into a building.
Then there’s the materials side of the award, which is arguably the more interesting half. Wide-bandgap semiconductors depend on supply chains that are not especially diversified. Critical minerals processing has long been concentrated in a handful of countries, and the US has spent the past several years trying to change that through legislation and targeted grants. Funding R&D into materials is a slower play than building a factory, but it addresses the part of the problem that money alone can’t fix quickly: the fundamental chemistry and manufacturing know-how.
It’s worth placing this against the broader moment. AI infrastructure spending has become a defining theme for markets, with hyperscalers and chipmakers racing to build capacity, and investors scrutinizing every capital expenditure line for signs of discipline or excess. Tesla’s recent capex guidance, for instance, rattled shareholders worried about the payoff timeline on its AI and robotics ambitions. When the market is this twitchy about returns on AI spending, efficiency stops being a nice-to-have. Power semiconductors sit directly on that fault line, because the cheapest kilowatt is the one you don’t waste.
The funding also lands amid a strange week for the industry’s reputation. Reports of a sweeping AI-discovered cybersecurity flaw forcing website shutdowns have renewed arguments about whether the rush to deploy AI is outpacing the safeguards around it. That tension doesn’t map neatly onto power electronics, but it does shape the political environment. Lawmakers funding semiconductor R&D want visible progress on competitiveness and resilience, and they want it framed as a national capability rather than a subsidy. Critical materials and power chips fit that framing comfortably.
Skeptics will note that R&D funding is a long-fuse instrument. It won’t produce a commercial product next quarter, and it won’t solve the industry’s dependence on foreign processing overnight. The history of government semiconductor programs is mixed: some grants seed genuinely important research, others vanish into academic papers with little industrial uptake. The design of this particular program will matter as much as the dollar figure, particularly whether it connects research institutions to manufacturers who can actually scale what comes out of the lab.
Still, there’s a reason Washington keeps returning to this corner of the chip stack. The AI boom has made it obvious that compute is a strategic resource, and compute depends on power. Every model trained and every inference served is ultimately a story about electrons moving through materials. The companies that control the materials and the power conversion technology hold leverage that doesn’t show up in benchmark charts.
For investors and engineers alike, the signal is that the AI supply chain is wider than the GPU. Memory, networking, cooling, and power delivery all have their own bottlenecks and their own chokepoints. The $389 million is a small down payment on the least visible of those layers, and it’s a reminder that the most consequential chip in the building might not be the one with the fanciest spec sheet.