Optical scale-up is on the horizon.
2026-07-22
Sofia Garcia
The AI Architect, New York
Optical scale-up refers to the development of high-performance optical systems that can produce larger quantities of precision optics without sacrificing quality or consistency. Industry experts believe that significant advancements in manufacturing technologies and materials science are on the horizon, bringing us closer to achieving mass-scale optical production with unprecedented accuracy and reliability.
The Blurring Lines Between Optical and Copper Interconnects in Data Centers
As artificial intelligence (AI) continues to grow in power and complexity, the need for faster and more efficient interconnects is becoming increasingly pressing. Currently, most data centers rely on a combination of electronic and optical communication protocols for different tasks. However, as AI requirements continue to escalate, the industry is shifting towards a new paradigm: optical scale-up.
In this context, Nvidia has been quietly pioneering a new approach to optical interconnects, which could revolutionize the way we connect processors in data centers. The company's NVLink Fusion program allows hyperscalers and cloud providers to build custom AI systems around Nvidia's scale-up fabric, featuring photonics players such as Ayar Labs, Marvell Technologies, and Lightmatter.
The Challenges of Copper Interconnects
Traditional copper interconnects have limitations when it comes to bandwidth and latency. As AI systems grow in size, the number of GPUs and connections between them increases exponentially. Electrical links are being pushed toward terabit-per-second signaling, but higher frequencies increase attenuation, power consumption, and heat. This makes it increasingly difficult to route copper cables through crowded server racks.
Engineers have dubbed this phenomenon the "copper wall," which marks the limits of what can be achieved with traditional copper interconnects. To overcome this challenge, researchers are exploring optical interconnects as a viable alternative.
The Rise of Co-Packaged Optics
Co-packaged optics involve integrating lasers, photonic devices, and electronic chips into a single package. This approach has become increasingly practical due to advances in hybrid bonding and the maturation of the semiconductor manufacturing ecosystem. Companies like Ayar Labs and Lightmatter are developing optical chiplets that can sit alongside GPUs and other processors, converting electrical signals into light.
Ayar's director of product management, Vishal Chandrasekar, argues that co-packaged optics have become practical because the industry has matured around this technology. Advances in hybrid bonding now allow electronic and photonic chiplets to be manufactured separately and integrated into a single optical engine.
The Future of Optical Scale-Up
Nvidia is taking a gradual approach to adopting these technologies, starting with scale-out networks and eventually moving to scale-up. The company expects optics to become more prevalent as AI's bandwidth requirements continue to grow.
Keren Bergman, a professor of electrical engineering at Columbia University, views NVLink Fusion as an ecosystem that allows electrical and optical approaches to coexist. "The Fusion is sort of this umbrella—you can put copper in it, you can put photonics in it," she says.
However, not everyone expects photonics to become the only answer. Researchers continue to improve electrical interconnects through advances in signaling, packaging, and transceiver design. Other groups are pursuing alternative technologies, such as 3D photonic connectors and laser integration directly onto silicon.
The Road Ahead
If optical scale-up becomes a widespread industry standard, future AI systems could span multiple racks while behaving as a single computing domain, connected via a mix of electrical, optical, and other emerging technologies. Currently, most data centers rely on a combination of electronic and optical communication protocols for different tasks.
As the industry continues to evolve, we can expect to see more innovative solutions emerge from companies like Ayar Labs, Lightmatter, and Nvidia. The future of optical scale-up is bright, with potential applications in everything from high-performance computing to edge AI and beyond.
