A San Jose-based semiconductor startup has raised $21 million to develop power technology for AI data centers, The Business Journals reports.

The round addresses a bottleneck that has moved to the center of the AI infrastructure buildout. As hyperscalers deploy racks drawing tens of kilowatts each to accelerate large-model training and inference, power delivery — not compute alone — has become a gating factor on data center expansion. Grid constraints across major US markets have already forced operators to cap new capacity, and component suppliers across the power chain, from voltage regulators to solid-state transformers, are seeing demand accelerate accordingly.

The $21 million commitment gives the San Jose company capital to advance its power technology toward that market. The company is one of a growing cohort of chip startups targeting the electrical architecture of AI facilities rather than the accelerators themselves — a segment where incumbent power semiconductor suppliers such as Infineon, onsemi and Texas Instruments are also racing to adapt portfolios originally built for servers consuming a fraction of today's per-rack power.

Silicon Valley remains the default launchpad for such ventures, with access to both venture capital focused on energy-plus-compute theses and to the design ecosystem supporting custom power management ICs. The Business Journals report did not disclose the investors in the round, the specific product architecture the company is developing, or a timeline to volume shipment.

The funding lands amid sustained investor appetite for AI-adjacent hardware plays. While accelerator developers command the largest rounds, suppliers of networking, cooling and power components have attracted steadily larger checks as data center operators confront the physical limits of delivering electricity to compute. Power semiconductors built on wide-bandgap materials such as silicon carbide and gallium nitride have drawn particular attention for their efficiency gains at high power densities.

Whether the San Jose startup can convert its $21 million into design wins with hyperscalers or their server OEMs will determine its position in a supply chain where qualification cycles are long and incumbents hold deep customer relationships. For now, the round signals that investors still see unclaimed territory in the power layer of the AI buildout, where every new accelerator rack sharpens the demand for better conversion and delivery of electricity.