The handbook of artificial intelligence (AI) is now well known. Build a larger GPU cluster. Add more Blackwell chips. Put more power into the problem. Once the chips are hot, build a data center next to the river. If you run out of bandwidth, install more copper wire.
in that way Amazon, alphabet, microsoftand meta platform And it works until you run into physics.
One company considered the same problem and arrived at a different answer. Global Foundries (GFS 0.88%) We believe that the real bottleneck in AI infrastructure is not computational power. This is the wire that connects the chip and replaces that wire with light.
Today’s changes
(-0.88%) $-0.71
current price
$79.92
Key data points
Market capitalization
$44 billion
daily range
$79.28 – $82.79
52 week range
$31.51 – $92.55
volume
190K
average volume
4.9 million
gross profit
November 26th%
The copper wall that no one talks about
Inside every AI data center, thousands of chips must share information at enormous speeds. Currently, the majority of communications occur over copper wires, which are running out of free space. They generate heat, lose signal over distance, and consume power, which can be costly at scale. As AI models get bigger, the copper problem gets worse.
The industry has known this for years. This solution is named co-packaged optics (CPO). The idea is to move the optical transceiver, a component that transmits data via light rather than electricity, right next to the chip, reducing the distance data travels over copper wires to almost zero. The result is a faster, cooler, and more power-efficient AI infrastructure.
In May 2026, GlobalFoundries announced SCALE (Advanced Light Engine Solution Co-Packaged with Silicon Photonics), the industry’s first platform to meet optical computing interconnect multi-source contract specifications for AI scale-up architectures. The platform uses both coarse wavelength division multiplexing (DWDM) and dense wavelength division multiplexing (DWDM) on each optical fiber to push bandwidth density and scalability beyond the limits of copper wire. Additionally, GlobalFoundries has already demonstrated 8 λ and 16 λ bidirectional DWDM natively on the platform. This is described by the company as a foundational milestone for everything to come.
Image source: Getty Images.
The part of the stack everyone is chasing
Here are some things about silicon photonics that get lost in the GPU coverage: It’s both a physical issue and a manufacturing issue. Designing silicon photonic chips is difficult. It is difficult to mass-build large-scale systems with the precision needed to align fiber optics for hyperscale data centers.
GlobalFoundries has spent years developing process technology to do just that. Its silicon photonics platform supports 50 Gbps and 100 Gbps microring modulators, wideband removable fiber interfaces, and flat insertion loss characteristics that future-proof the platform as the number of wavelengths increases. In November 2025, the company acquired Singapore’s Advanced Micro Foundry, a specialist silicon photonics manufacturer, adding manufacturing assets, intellectual property and engineering depth that would take years to build from scratch.
With this acquisition, GlobalFoundries gains silicon photonics production capacity in Singapore. Singapore is an important region for supply chain diversification amid rising U.S.-China semiconductor tensions. The company is building a platform that hyperscalers need, but few manufacturers can actually deliver.
GlobalFoundries is down 10% in one day, but we’re not worried
GlobalFoundries fell nearly 10% on May 27, dragged down by Mubadala’s stock sale allegations. Mubadala was previously the controlling shareholder of GlobalFoundries, an Abu Dhabi sovereign wealth fund. Despite the stock price decline, the Motley Fool’s long-term framework wins here.
The GlobalFoundries story remains the same. Just a few weeks ago, SCALE’s announcement sent its stock up 12% in just one trade. Earnings in the first quarter of 2026 continued to outperform. Silicon photonics revenues are expected to gain momentum with more than 500 successful designs recorded in 2025 and nearly double again in 2026. The sale does not change these facts.
The U.S. government is also backing GlobalFoundries with a proposed $375 million grant to help build domestic quantum manufacturing infrastructure.
All major hyperscalers are looking for ways to train larger models faster. GlobalFoundries is asking a different question. How do you move data between chips without melting the infrastructure?
Packing optics is the answer. Companies that are building manufacturing platforms to deliver it at scale are most often still filed as “semiconductor foundries.” On days when stock prices drop 10% for reasons unrelated to your most important business, that drop presents an opportunity.
