On July 29, the U.S. Department of Commerce announced 7 letters of intent to provide $874 million in federal research and development incentives under the CHIPS and Science Act, targeting domestic technologies to accelerate AI compute infrastructure and secure semiconductor supply chains. The funds will support breakthroughs in photonics, memory architectures, advanced packaging, and substrates—addressing the energy efficiency and bandwidth constraints that currently throttle AI deployment at scale.
GlobalFoundries received up to $300 million to accelerate co-packaged optics (integrating photonics directly alongside AI processors), reducing timelines by 2–3 years. Kepler secured up to $245 million for high-performance AI memory technology using 3D and ferroelectric innovations. Multibeam Corporation will get up to $140 million for advanced chip packaging and stacking to enable multi-chip AI systems. Extropic received up to $75 million for thermodynamic sampling units (TSUs) that solve optimization and sampling problems at reduced cost. Four other companies shared the remaining ~$114 million across photonics, memory, and connectivity work.
The Department will receive minority, non-controlling equity stakes in each company, aligning federal interests with commercial success. Commerce Secretary Howard Lutnick framed the investments as accelerating "America's innovation engine" against foreign competition. The funding targets bottlenecks—power delivery, memory access, packaging density, interconnect bandwidth—that currently limit AI lab ability to scale inference and fine-tuning beyond hyperscaler environments.
For architecture teams: these CHIPS investments address the structural constraints that make 22 GW of 2026 capacity expansions and the $630 billion hyperscaler capex feasible only for the largest cloud operators. Advances in co-packaged optics, novel memory, and thermal efficiency could extend affordable AI deployment to smaller data centers and improve cost-per-token across the board. Watch for commercialization timelines: the targeted 2–3 year acceleration means competing silicon and packaging advances from these 7 will begin reaching production by 2028–2029.