For years, the semiconductor industry's arms race has been measured in nanometers — the ever-shrinking transistors that let a single chip pack in more compute. On July 29, 2026, the U.S. government signaled that the next front in the AI hardware war is not on the chip at all, but in the wires between them. GlobalFoundries said it had signed a letter of intent with the U.S. Department of Commerce for a $300 million CHIPS Act research award to accelerate silicon photonics — the technology that moves data between AI processors using light instead of copper.
The award, disbursed through Commerce's CHIPS Research and Development Office, is aimed squarely at the bottleneck that increasingly defines large AI clusters: not how fast an individual GPU can crunch numbers, but how fast data can travel between thousands of them. As AI models balloon, the copper links stitching processors together have become a wall — one that constrains both bandwidth and the power budgets of data centers.
Betting on light over copper
Silicon photonics transmits information as pulses of light, delivering far higher bandwidth and lower power draw than the electrical interconnects that have carried data inside computers for decades. GlobalFoundries plans to use the money to advance next-generation optical materials, wafer technologies and advanced packaging — including 3D hybrid bonding — that enable near-packaged optics (NPO) and co-packaged optics (CPO), architectures that fuse optical engines directly alongside AI chips.
The company anchored its pitch to its recently introduced SCALE platform — short for Silicon Photonics Co-Packaged Advanced Light Engine — and set concrete engineering targets: 400 gigabits per second of throughput and a fivefold gain in energy efficiency over current-generation implementations. The work will run through GF's existing facilities in Malta, New York, and Burlington, Vermont, which the company frames as a domestic path to high-volume optical manufacturing.
"Silicon photonics is essential to AI infrastructure. For a decade, the industry talked about the shift from copper to optical as something that was coming — today it is here, moving data at higher bandwidth and improved power efficiency as workloads grow more complex," said Tim Breen, chief executive of GlobalFoundries. "GlobalFoundries has spent more than a decade building the technology, footprint and ecosystem to lead this transition, and we have the proven manufacturing foundation to scale it — in the United States."
Commerce officials cast the award as part of a deliberate redirection of CHIPS money toward foundational bottlenecks rather than leading-edge logic alone. "The CHIPS R&D incentives will support a breakthrough in compute and communication networks moving past traditional copper bottlenecking to power next-generation AI," said Bill Frauenhofer, executive director for semiconductor innovation and investment at the Department of Commerce. Commerce Secretary Howard Lutnick added that the investment would "enhance our country's domestic capabilities, create high-paying jobs and keep America at the forefront of the semiconductor industry."
In a separate agreement, Commerce will take roughly a 1 percent equity stake in GlobalFoundries — a structure the Trump administration has used to let the public share in the upside of companies it funds.
Why this matters
The award is notable precisely because GlobalFoundries does not compete at TSMC's most advanced nodes. It is a mature-node and specialty foundry, and Washington's willingness to hand it $300 million reflects a broader recognition that AI performance is now gated by packaging, memory and interconnects as much as by transistor density. The bottleneck has shifted, and policy is following.
That shift was underscored by an unusually broad chorus of endorsements bundled into the announcement. Executives from AMD, Broadcom, Cisco, Corning, Lumentum, Marvell, Meta, Microsoft, Nvidia and Qualcomm all lined up behind the effort — a rare show of unanimity from companies that otherwise compete fiercely.
"The bottleneck in AI infrastructure is shifting from compute to connectivity — the ability to move data between and within systems without letting bandwidth or power constraints limit performance," said Chris Koopmans, president and chief operating officer of Marvell. Nvidia founder and CEO Jensen Huang put it in industrial-policy terms: "Silicon photonics is essential to that future, and GlobalFoundries brings the manufacturing expertise to help make it real in the United States."
The endorsements are not merely ceremonial. Co-packaged optics only pays off if hyperscalers and chip designers actually adopt it, and comments from Meta's Yee Jiun Song and Microsoft's Rani Borkar signal that the largest buyers of AI infrastructure want a domestic, multi-supplier optical supply chain rather than dependence on a single overseas source.
There is also a defense and high-performance-computing dimension. High-bandwidth, energy-efficient optical links are as relevant to secure military systems and national labs as to commercial data centers, and building that capability on U.S. soil is a stated goal of the CHIPS program.
What to watch next
The deal is still a letter of intent, not a finalized, disbursed award — GF's own filings note that CHIPS funding can be delayed or withheld. Watch for definitive terms and milestone schedules to firm up in the coming months. Investors will get an early read when GlobalFoundries reports second-quarter results on a conference call slated for August 5, 2026, where management is likely to detail how the photonics push fits its broader capital plans.
Beyond GF, the bigger question is whether $300 million meaningfully accelerates the copper-to-optical transition or simply subsidizes a shift the market was already funding. The engineering targets — 400 Gb/s and a 5x efficiency gain — are the numbers to track. If GlobalFoundries hits them at high volume, the U.S. will have a homegrown answer to one of AI's hardest scaling problems. If not, the interconnect bottleneck will remain the quiet ceiling on how large — and how power-hungry — the next generation of AI clusters can grow.
"The bottleneck in AI infrastructure is shifting from compute to connectivity — the ability to move data between and within systems without letting bandwidth or power constraints limit performance."- Chris Koopmans, President and COO, Marvell