2026-07-18
Written by Lily Chen
The concept of optical scale-up has garnered significant attention in recent years as researchers and scientists explore innovative methods for scaling up photonic devices. This approach aims to integrate multiple layers into a single material, enhancing performance and efficiency while reducing size and energy consumption.
The Evolution of Optical Communication in Data Centers: A New Era of Scale-Up?
The shift towards more efficient and powerful computing systems has led to an explosion of data centers around the world. As artificial intelligence (AI) continues to grow in complexity and scale, traditional electrical communication protocols are struggling to keep up. The integration of optical communication protocols has emerged as a promising solution for long-distance connections in scale-out networks, but what about scale-up networks?
Scale-up networks connect multiple GPUs within a single mega-computer or rack, where latency is critical. In the past, dense copper interconnects, such as Nvidia's NVLink, have been the engineering solution of choice. However, as AI needs continue to grow, the limitations of electrical links are becoming increasingly apparent. Higher frequencies increase attenuation, power consumption, and heat, making it challenging to route cables through crowded server racks.
To address these challenges, researchers are exploring the use of optical interconnects in scale-up networks. Nvidia's NVLink Fusion program, launched in 2025, allows hyperscalers and cloud providers to build custom AI systems around Nvidia's scale-up fabric. This has attracted several photonics players, including Ayar Labs, Marvell Technologies, and Lightmatter.
One of the key challenges in deploying optical interconnects is converting electrical signals into light and integrating lasers, photonic devices, and electronic chips into a single package without blowing up cost, power consumption, or manufacturing complexity. However, engineers argue that these hurdles are no longer insurmountable.
Ayar Labs, for example, has developed optical chiplets meant to sit alongside GPUs and other processors, converting electrical signals into light only millimeters from the compute silicon. "The most optimal way to do that is having a photonic chiplet with an electronic chiplet and having them hybrid bonded together," says Ayar's director of product management, Vishal Chandrasekar.
Lightmatter has taken a different approach by designing a 3D photonic connector with input and output ports across the entire chip area. This allows for more flexible and efficient connections between components. Meanwhile, Lightmatter is also building a photonic interposer that serves as the packaging substrate itself, enabling future processors to be stacked directly on top of a silicon photonics engine.
The remaining challenge in optical scale-up is laser integration. Current pluggable laser modules take up valuable rack space and are hard to scale. Instead, Lightmatter is putting large numbers of lasers directly onto silicon, which they believe could soon support much denser optical scale-up fabrics.
As the industry continues to evolve, Nvidia is taking a gradual approach to adopting these technologies. Principal product marketing manager Jesse Clayton says that the company expects optics to move into scale-up networking eventually, as AI's bandwidth requirements continue to grow.
"I think we've taken an approach of migrating to optical when it makes the most sense for our platform," he says. "If you migrate the entire platform at once, you take a tremendous amount of risk on new designs. So, starting at the scale-out space and then in the future moving to scale-up is kind of a sensible, measured approach from our perspective."
The slow-and-steady attitude of Nvidia is one reason why some researchers view NVLink Fusion as more than an interoperability play. "The Fusion is sort of this umbrella—you can put copper in it, you can put photonics in it. It's very photonics friendly," says Keren Bergman, a professor of electrical engineering at Columbia University.
However, not everyone expects photonics to become the only answer. Researchers continue to improve electrical interconnects through advances in signaling, packaging, and transceiver design. And other groups are pursuing alternative technologies.
Ultimately, the question remains: can optical scale-up become something that the industry at large can do, rather than a proprietary feature of Nvidia's ecosystem? "Absolutely," says Chandrasekar. "There are going to be multiple implementations in the 2028 time frame in very high volume."
If this happens, future AI systems could span multiple racks while behaving as a single computing domain, connected via a mix of electrical, optical, and perhaps other emerging technologies. The future of data centers will likely involve a combination of these different approaches, each with its own strengths and weaknesses.
As the industry continues to evolve, one thing is clear: the shift towards more efficient and powerful computing systems will require innovative solutions like optical communication protocols. Whether it's through NVLink Fusion or other emerging technologies, the future of data centers looks brighter than ever.
