Photonic Chips: The Hardware That Makes Photonics Real

A photonic chip is a semiconductor device that converts electrical signals into light and back again. It is the hardware component that makes the transition from copper to photonics inside AI data centers possible. Without photonic chips, the idea of using light to move data between GPUs remains theoretical. With them, it is a shipping product generating billions in revenue.

Jason Bodner’s Accelerated AI presentation focuses on photonic chips as the key enabler of the next AI boom. He reveals Broadcom (AVGO) for free because Broadcom debuted the industry’s first 400G optical chip, the specific component that converts electrical signals to light at the bandwidth required for modern AI workloads. As we explain in our Broadcom optical chips article, this is not a prototype. It is a product in production.

How Photonic Chips Work

A photonic chip, also called a silicon photonics transceiver, contains several components fabricated on a silicon substrate using standard semiconductor manufacturing processes:

Lasers. Microscopic lasers generate the light pulses that carry data. These are typically vertical-cavity surface-emitting lasers (VCSELs) or distributed feedback lasers, depending on the application.

Modulators. Modulators encode data onto the light by varying its intensity, phase, or frequency. The modulation speed determines the data throughput. A 400G chip can handle 400 gigabits per second, enough to transfer a high-definition movie in a fraction of a second.

Photodetectors. Photodetectors receive the incoming light signals and convert them back into electrical signals that the GPU or switch can process.

Waveguides. Waveguides are the optical equivalent of copper traces. They channel light through the chip with minimal loss, using total internal reflection to keep the light confined to a path only a few microns wide.

The manufacturing process leverages the existing semiconductor supply chain, which means photonic chips can be produced at scale using the same fabrication facilities that make conventional chips. This is important because it means the transition from copper to photonics does not require building an entirely new manufacturing infrastructure.

Why They Matter for AI

AI data centers require thousands of GPUs to work together simultaneously. The data moves between those chips over copper wires, and copper is the bottleneck. Electrons in copper travel at less than 1 percent of the speed of light. They generate heat, consume significant power, and cannot keep up with the bandwidth demands of modern AI workloads.

Photonic chips solve this by converting the electrical signals from GPUs into light, which travels at 124,000 miles per second. Light carries more data, barely uses power, and does not generate heat. Bodner says this makes AI “100 times faster and 100 times more energy efficient.” That is a significant claim, but the physics supports it. Nothing in the known universe moves faster than light.

For more on the broader technology transition, see our photonics explainer.

The Companies Making Photonic Chips

Broadcom is the leader in shipping volume photonic chips at 400G, but it is not the only company working on this technology. Nvidia has invested over $7 billion in photonics companies and partnered with Corning to build three optical factories in the United States. AMD is building a $280 million photonics research hub. Ayar Labs raised $500 million from ARK Invest and Sequoia Capital. Google and Microsoft are backing nEye, which raised $80 million. Bill Gates has personally invested over $200 million in two photonics companies.

Bodner also identifies three smaller companies behind the paywall that are involved in photonic chip manufacturing:

  1. A company that builds parts turning electrical signals to light inside AI clusters. It went on an acquisition spree for photonic firms, and Nvidia signed a multi-billion partnership with it. Jensen Huang calls it “the next trillion-dollar company.” Sales are up 42 percent year over year.

  2. A pure photonics play that moves data as light between data centers across cities and states. Its new product cuts networking power use by 70 percent. Sales are over $6 billion. It is 20x smaller than Broadcom.

  3. A secret weapon inside every major AI data center, with two of the three biggest AI spenders as customers. Nvidia just tapped it as a key partner. Sales are up 35 percent last quarter. It is 10x smaller than Broadcom.

For more on these picks and the full investment thesis, see our full Accelerated AI review.

The Historical Parallel

Bodner compares the current moment to the copper-to-fiber transition during the internet boom of the 1990s. Cisco, which dominated the networking layer, surged 3,800 percent. The companies that built the optical infrastructure for the internet were among the biggest winners of that era. Bodner argues that AI is a far bigger market and the photonic chip transition will produce even larger returns. For more on the pattern, see our copper to fiber article.

Considerations

The photonic chip market is real and growing, but it is also competitive. Multiple companies are investing heavily, and it is not yet clear which will dominate the next generation of optical components. Broadcom has a head start in shipping volume, but Nvidia’s $7 billion investment and AMD’s $280 million research hub suggest the landscape could shift. The right approach is to follow the capital flows and the shipping products, not just the announcements.

If you want to explore Bodner’s full thesis, you can access the Accelerated AI presentation through Brownstone Research.

This is not financial advice. Always do your own research before investing.