Why the US is investing $300 million in GlobalFoundries silicon photonics for AI

The US government has awarded semiconductor manufacturing giant GlobalFoundries $300 million to accelerate the development and domestic production of silicon photonics for next-generation artificial intelligence chips. This funding, a strategic allocation from the federal CHIPS and Science Act, directly addresses the most critical physical bottleneck in modern AI datacenters: the speed at which silicon can transmit data. By transitioning from traditional copper-based wiring to light-based optical interconnects, the initiative aims to bypass the hard physical scaling limits of modern microelectronics while securing a crucial domestic supply chain.

The Looming Copper Wall in AI Datacenters

For decades, scaling computer performance was governed by Moore's Law—the predictable doubling of transistors on a sliver of silicon. But as artificial intelligence models scale exponentially, the bottleneck is no longer just how fast an individual graphics processing unit (GPU) can compute, but how quickly thousands of GPUs can talk to one another. Today's massive AI training clusters are choking on their own copper plumbing.

Traditional copper interconnects suffer from severe physical limitations at high frequencies. As data rates climb, copper wires generate unsustainable levels of heat and experience signal degradation, requiring massive amounts of power just to push electrons across a circuit board. This "communications wall" threatens to stall the progress of frontier AI models, which require seamless, low-latency synchronization across tens of thousands of distributed chips.

This is where silicon photonics becomes a necessity rather than an alternative. By using laser-generated light instead of electricity to transmit data, optical interconnects can move information at the speed of light with virtually zero resistance, negligible heat generation, and orders-of-magnitude higher bandwidth density. The technology effectively replaces copper wires on the chip and system level with microscopic optical waveguides.

Why GlobalFoundries and Why Now?

While cutting-edge logic manufacturing (sub-3nm processes) is dominated by Taiwan Semiconductor Manufacturing Company (TSMC), the transition to optical computing requires specialized foundry capabilities. GlobalFoundries, the largest US-based contract chipmaker, has quietly positioned itself as a pioneer in this space with its specialized GF Fotonix platform, which integrates silicon photonics components directly onto standard silicon manufacturing lines.

"Silicon photonics is no longer a futuristic research project; it is the physical infrastructure that will determine which nation and which companies lead the next wave of artificial intelligence."

US Department of Commerce

The $300 million federal award will be used to scale and upgrade GlobalFoundries' manufacturing facilities in Malta, New York, and Burlington, Vermont. Manufacturing optical chips is notoriously difficult because it requires aligning lasers, modulators, and photodetectors onto silicon with nanometer-level precision. The federal funding is designed to transition these complex optical packaging processes from low-yield, specialized labs into high-volume, cost-effective commercial production.

The Geopolitical Race for the Optical Physical Layer

The US Department of Commerce’s decision to fund GlobalFoundries silicon photonics development is deeply rooted in geopolitical supply chain defense. While the US has aggressively funded domestic leading-edge logic fabs via Intel and TSMC's Arizona facilities, those chips still rely heavily on overseas packaging and advanced networking ecosystems centered in Asia.

By establishing a robust, onshore supply chain for optical interconnects, the US is attempting to secure the high-value physical layer of future AI hardware. Hyperscalers like Meta, Microsoft, and Google are already designing custom AI silicon that heavily leverages chiplet-based architectures. In these modular designs, different functional blocks of a chip are bound together using optical interfaces—a design paradigm that is impossible to execute at scale without domestic foundry support.

Furthermore, this funding serves as a vital catalyst for the wider silicon photonics ecosystem. Startups like Ayar Labs and Celestial AI, which design optical I/O chiplets and memory interconnects, rely on foundry partners like GlobalFoundries to fabricate their designs. A well-funded domestic foundry lowers the barrier to entry for these innovators, ensuring that the next generation of optical computing IP remains anchored in the US.

The Shift from Transistors to Systems

The industry-wide pivot toward silicon photonics represents a fundamental shift in how we define computing power. The historical race to shrink transistors to the single-digit nanometer scale is yielding diminishing returns. The future of AI hardware scaling lies in system-level integration—how efficiently we can link compute, memory, and networking into a single, cohesive optical supercomputer.

With this $300 million injection, the US government is betting that the winner of the AI race won't just be the country with the smallest transistors, but the one that can move data between those transistors the fastest. By backing GlobalFoundries, the US is laying the physical foundations to ensure that light-speed AI computing is built on domestic soil.