Can silicon photonics overcome scaling challenges for AI and data centers?

Artificial intelligence (AI) is reshaping the world as we know it. The rise of generative AI tools like ChatGPT and other large language models (LLMs) is unleashing an explosion of new applications—real-time customer interactions, advanced analytics, and decision-making systems—that require unprecedented levels of computing power. These advancements come with challenges. The computational intensity of training LLMs and supporting their real-world applications demands massive datasets, ultralow latency, and scalable infrastructure.

As I’ve often said in discussions: this growth in AI isn’t a trend. It’s a fundamental shift in our world and how we think about data centers. The ability to scale AI effectively depends on rethinking the architecture and technology that power these systems. Silicon photonics, with its ability to transmit data at high speeds while reducing power consumption, offers a critical solution. Yet, like any transformative technology, scaling silicon photonics for widespread adoption requires tackling several technical and systemic barriers head-on.

Scalable infrastructure

For years, data centers served as the backbone of internet-driven workloads like video streaming and e-commerce. Today, AI applications—particularly ones driven by generative models—are upending this role. The rapid growth of AI and machine learning is redefining expectations, particularly in terms of bandwidth, latency, and energy efficiency.

Data center power demand is expected to grow at a compound annual growth rate (CAGR) of 15%, according to Goldman Sachs, while AI data center growth will be at 50% CAGR from 2023 to 2030. By the end of the decade, data centers are expected to account for 8% of total U.S. power consumption—up from 3% today. This trajectory is unsustainable with existing electrical interconnects. Optical interconnects, enabled by silicon photonics, are the only scalable path forward.

Silicon photonics is already proving to be a game-changer at overcoming bandwidth bottlenecks, addressing power challenges, and reducing heat generation. But it’s not a silver bullet. While the technology itself is compelling, its full potential depends on innovation and collaboration at every level—from materials science to manufacturing.

Silicon photonics stands out

Silicon photonics combines the well-understood manufacturing techniques of silicon semiconductors with the transformative capabilities of optical communication. It enables data to be transmitted using light, which is faster, more efficient, and capable of handling the enormous demands of AI workloads.

I’ve often described silicon photonics as more than an incremental improvement—it’s a paradigm shift. Unlike traditional electrical interconnects, which are quickly reaching their limits, optical interconnects offer a sustainable way to handle both the increasing volume of data and the complexity of AI architectures. And while challenges remain to reduce costs and scale production, the benefits of silicon photonics—higher bandwidth, reduced latency, and lower power consumption—make it an indispensable part of our future infrastructure.

Challenges of scaling silicon photonics

Scaling silicon photonics isn’t easy. Significant hurdles must be addressed, and I see four key areas that need focused innovation:

Energy efficiency. As AI workloads continue to grow, power consumption in data centers is becoming an issue. Without new approaches, projections show energy use could grow exponentially. Silicon photonics addresses it—consuming less power and generating less heat compared to traditional solutions—but we can go ever further. Innovations like co-packaged optics (CPO) reduce power consumption by as much as 30% compared to traditional pluggable solutions, and can bring optical components closer to the chip to minimize energy losses.

Manufacturing and packaging. Precision is critical for photonics—aligning optical elements at scale is one of the greatest challenges of manufacturing. Current processes are often manual, which isn’t scalable or cost-effective. We need to adopt automated solutions and improve integration with other technologies, such as graphics processing units (GPUs) and memory units, to truly unlock silicon photonics’ potential.

Ecosystem development. Silicon photonics doesn’t operate within a vacuum—to succeed, we need a robust ecosystem of tools, standardized packaging, and industry collaboration. Design tools, for example, need to evolve to model complex photonics systems with the same precision as traditional electronics.

Cost optimization. Silicon photonics must be affordable to achieve widespread adoption. This means improving manufacturing efficiencies and reducing the cost of materials, without compromising on performance.

Collaborative approach

One thing I’ve learned during my career is that innovation doesn’t happen in isolation. At Soitec, we develop engineered substrates like silicon-on-insulators (SOIs) to serve as the foundation for silicon photonics. Our materials are designed to enhance the optical properties and mechanical stability needed for high-performance applications to ensure the uniformity and surface quality critical to yield.

This is merely one piece of the puzzle. Success in scaling silicon photonics requires collaboration across the semiconductor ecosystem—from substrate suppliers to systems integrators and beyond. By working together, we can address the challenges of cost, efficiency, and scale.

AI and data center architectures

The potential of silicon photonics extends far beyond improving current data center operations. Its ability to enable innovations like co-packaged optics can fundamentally reshape AI architectures. By reducing the physical distance between processing and memory units, silicon photonics enhances speed, efficiency, and scalability. It opens the door to real-time analytics, more advanced AI models, and new applications in fields like augmented reality and autonomous systems.

Building a sustainable future

As we look to the future, sustainability must remain a priority. The power demands of AI can’t come at the expense of environmental health. Silicon photonics offers a way to dramatically reduce power consumption and cooling requirements within data centers. At Soitec, sustainability is embedded into both our materials development and manufacturing processes to help minimize waste and improve energy efficiency.

The journey to scale silicon photonics is as exciting as it is challenging. Demands of AI require us to think differently—to reimagine what’s possible for data center and computing infrastructure. Silicon photonics is an enabler of the next wave of technological innovation.

Reflecting on my years within the semiconductor industry, I’m inspired by what we’ve accomplished and energized by what’s ahead. The pace of change is accelerating and, while there are challenges, I have no doubt that through collaboration, innovation, and determination, we’ll rise to meet them.