This article digs into how optical networking might turn into the next big headache for AI infrastructure. Copackaged optics innovations like GlobalFoundries’ SCALE are in the spotlight, along with all the buzz around silicon photonics, interconnects, and data-center packaging. Investors are watching closely, and there’s a lot of back-and-forth over whether copper or optical interconnects will win out as AI workloads get heavier.
Optical Interconnects: The Next Urgent Bottleneck for AI Infrastructure
AI models just keep getting bigger and hungrier for data. That means the need for high-bandwidth, low-power links between servers, accelerators, and storage is soaring.
Goldman Sachs analysts have flagged optical networking as a looming bottleneck for AI infrastructure. They think it could outpace old-school copper wiring when it comes to scaling and efficiency. This has everyone eyeing copackaged optics and silicon photonics as ways to get more bandwidth and use less energy per bit.
Now, the industry’s looking for ways to put optical transceivers right next to compute chips. Streamlining physical interconnects inside data centers is a big part of that. The dream? Lower latency, less power, and way more bandwidth for AI workloads—without making networks massive, expensive, or unwieldy.
GlobalFoundries SCALE Copackaged Optics: How it Works
SCALE is GlobalFoundries’ take on copackaged optics. It pairs optical transceivers right with chips to trim power usage and ramp up bandwidth. In demos, SCALE pushed more data through a single optical fiber and improved bandwidth density compared to copper.
GlobalFoundries claims their platform beats the AI data-center optical interconnect specs set by the Optical Compute Interconnect Multi-Source Agreement, which includes big names like Nvidia, Broadcom, and AMD. The basic idea? Cut down on loss points and energy waste that usually come with separate optics and processor packages.
By putting optics and compute silicon together, data centers can cram more data into each fiber and deal with less heat. That means tighter rack density and more flexible network setups. If this scales the way they hope, it could speed up AI rollouts and cut costs for next-gen AI clusters.
Industry Implications and Supply Chain Considerations
Interest in copackaged optics and other optical tech is picking up as AI data centers keep multiplying. Packaging and interconnects could become supply-chain chokepoints, so tech like SCALE is getting attention at just the right moment.
Some folks still think copper will stick around for certain basic network connections. Others are betting optical interconnects will slowly edge copper out, especially where bandwidth and energy are tightest.
Wall Street can’t seem to agree on when copper will really fade. Even Nvidia’s top brass are pushing a mixed approach, using both copper and optical to build out networks. Cantor Fitzgerald analyst C.J. Muse thinks long-term trends like silicon photonics and edge computing could be a big deal for GlobalFoundries. Muse bumped GF to overweight, set an $80 price target, and sees their photonic integrated-circuit platform unlocking new revenue streams.
Market Response and Investor Outlook
The announcement gave GlobalFoundries’ stock a noticeable boost. Shares climbed as investors tried to gauge how integrated photonics could reshape data-center economics in the long run.
Still, the market’s a bit wary. The timeline for optical interconnects taking over isn’t clear, thanks to tricky manufacturing, standardization, and supply chain issues that come with copackaged optics.
Analyst Voices and the Potential Upside
Analysts who see real potential in silicon photonics often highlight GF’s photonic integrated-circuits business. They also mention GF’s exposure to edge computing as a possible driver for steady revenue growth.
Muse’s upgrade shows some faith that GF can take its photonics expertise beyond standard silicon devices. Data centers want higher bandwidth but less power per bit, so there’s a real incentive here.
The $80 target means about an 18–19% upside from the price mentioned earlier. That’s a sign investors believe CPO-enabled architectures could have a multi-year runway, though of course, nothing’s guaranteed.
Nvidia’s leadership keeps pushing a pragmatic approach to network scaling. They’re mixing copper and optical interconnects as AI systems get bigger.
This approach hints that ditching copper entirely is probably too soon, maybe even risky. Instead, a gradual, standards-based roll‑out of optical interconnects could still shake up data-center design before long.
- Key takeaway: Copackaged optics might boost AI data-center density and power efficiency—if those scalability claims actually play out in the real world.
- Industry risk: Supply-chain resilience for photonics components is still a huge concern for getting AI infrastructure up and running on time.
- Investment signal: Analysts are starting to weigh photonics more heavily in their AI infrastructure forecasts, and some are making selective upgrades and raising price targets because of it.
For folks building or researching this stuff, it’s clear that photonics needs to be part of AI infrastructure planning from the start. That means paying attention to thermal management, packaging standards, and how to scale up manufacturing.
As silicon photonics keeps maturing, and copackaged solutions start to show real-world results, the next wave of AI growth could really depend on how fast and efficiently data can move. It’s a big “if,” but the potential is hard to ignore.
Here is the source article for this story: Optics is the next big AI bottleneck. This company could be an underrated beneficiary.