The Trump administration announced nonbinding letters of intent to invest up to $874 million in CHIPS Act funding across seven semiconductor companies in exchange for minority, non-controlling equity stakes in each firm. Kepler Computing will receive up to $245 million for R&D on a new class of high-performance AI memory using 3D and ferroelectric technologies. GlobalFoundries gets up to $300 million for co-packaged optics—integrating photonics directly with AI processors to enable ultra-fast, energy-efficient computing. Multibeam will receive $140 million for advanced chip packaging technology.
The remaining four recipients—Extropic ($75M for thermodynamic sampling units), Thintronics ($50M for low-loss insulating layers), OBSIDIA Semiconductors ($34M for component authentication systems), and Aeluma ($30M for photodetectors and lasers)—round out the awards. Commerce Secretary Howard Lutnick framed the strategy as generating taxpayer returns while building domestic semiconductor leadership. All agreements remain subject to further diligence and formal approval before funds are dispersed.
This marks an escalation in the Trump administration's "equity for funding" model across CHIPS Act investments. Since December 2025, the Commerce Department has announced 19 total awards (final or proposed) totaling up to $3.8 billion, all tied to government equity stakes. The policy shift—taking minority shareholdings rather than pure grants—differs from the Biden administration's grant-only approach under the same CHIPS Act appropriation, and expands a pattern established by the government's 10% stake in Intel ($8.9B for CHIPS Act conversion).
For architects evaluating supply chain and manufacturing partnerships, this signals U.S. government involvement now extends to equity ownership in strategic chip technology. Minority, non-voting stakes preserve operational independence but tie taxpayer returns to company valuations and exit events (IPO, acquisition). The focus on photonics, memory, and packaging reflects acknowledgment that chip bottlenecks sit across multiple layers of the stack, not just CPU/GPU fab capacity.