When Data Turns to Light
The Photonic Shift Quietly Reshaping Compute
Most of the internet still runs on copper. Not in the long-haul pipes—those were swapped for fiber years ago—but in the short spans that matter most: inside racks, between chips, across servers. That’s where the physical limits are getting hard to ignore. Wires take up space, waste energy, throw off heat. And as models scale and systems get denser, those factors add up.
Photonics offers an alternative. Instead of moving data with electricity, it uses light. Photons don’t overheat. They don’t interfere with each other. They don’t require bulky shielding. That’s what makes them useful: not raw speed, but the possibility of scaling dense, high-performance systems without being throttled by power and bandwidth limits.
What you need to know
This month, MIT published new work on a photonic signal processor that uses light instead of electrons to handle wireless data. While not a general-purpose computer, the chip is designed for edge scenarios—real-time telecom processing, sensor fusion, and military radio applications. The team emphasized that photonics isn’t only for datacenter-scale systems. It’s also a candidate for making compact, intelligent devices more efficient.
Last week, researchers at Fudan University demonstrated a silicon photonic chip that reached 38 terabits per second using a new mode-multiplexing design. This is a functioning prototype built on standard fabrication processes, meaning it can be manufactured using existing semiconductor tools. This is fabricated silicon, not a theoretical exercise. It was designed with manufacturing constraints in view and built for systems that exist outside the lab.
Moreover, CHIPX has started limited production of a material that can move data using light instead of electricity. It’s an early step, but it shows the technology is getting close to being used in real hardware, not just experiments. The company—spun out of Shanghai Jiao Tong University—is producing thin-film lithium niobate wafers for photonic chips. Lithium niobate has long been valued for its optical properties, but it was considered too complex for modern chipmaking. CHIPX is now manufacturing 6-inch wafers with over 110 GHz modulation bandwidth and extremely low signal loss—bringing a lab material closer to commercial reality.
Adtran, a U.S. telecom hardware firm, also showed off new integrated photonics at the Laser World of Photonics event in Munich. Their messaging was direct: hyperscale data centers aren’t sustainable without a shift in how we interconnect compute. Their new hardware is designed to address that, using embedded optical components to cut power use and increase density inside racks.
Ecosystem Maturity, Not Just Materials
These developments are tied together by something more important than performance: manufacturability. The real shift is that photonic systems are becoming compatible with existing foundries and packaging techniques. Companies like OpenLight are now offering full photonic design kits to let chipmakers create light-based circuits using standard tools. Jabil, a massive electronics manufacturer, is expanding its packaging capabilities to support photonic die at commercial scale.
The hardware stack is finally starting to look buildable.
Who’s Betting on It—And Who’s Not
Governments are paying attention. China’s investment in CHIPX was part of a broader push to establish independence in strategic computing materials. The EU has designated photonic supply chains as part of its sovereignty agenda. In the U.S., companies like Nvidia, Cisco, and Intel are placing long-term bets on co-packaged optics, embedding photonic links directly alongside processors and switches.
Venture capital is beginning to follow. HyperLight, a Harvard spinout, raised $37 million to develop thin-film photonic platforms last year. Lightmatter and Ayar Labs both extended funding rounds this spring. Some funds see this as a repeat of the early AI wave, a deep tech that looked niche until the stack caught up.
That said, not everyone is convinced. Several chip designers remain skeptical about photonics in general-purpose applications. Lasers still require space and precision alignment. Coupling light efficiently into and out of chips is a known challenge. And most data center networks today can still be optimized further using improved copper routing and smarter signal processing. The skeptical view is that photonics is a specialty solution—not a general replacement.
Much of the difficulty lies not in making photonics work, but in making it standard. Companies have spent decades tuning their infrastructure around copper-based systems—every cable, switch, rack, and thermal design reflects the physics of electricity. Swapping in optics isn’t plug-and-play. It forces changes to packaging, cooling, signal integrity models, and often, the software stack that governs how components talk to each other. Even when the performance benefits are clear, the operational risk of changing something that already works is high—especially in environments where uptime matters more than raw speed.
There’s precedent. When data centers shifted from spinning disks to solid-state drives, it took years of incremental testing, cost reduction, and tooling improvements before SSDs became default. The same happened when cloud providers adopted GPU clusters for AI: no one flipped a switch overnight. It took pilot deployments, new scheduling software, redesigned power and cooling systems, and a willingness to rethink how performance was measured. Photonics demands a similar shift—not just in technology, but in mental model. Doubt doesn’t come from lack of vision, but from experience with how hard it is to unwind infrastructure that was never meant to shift.
What June Signals
This is a shift in how infrastructure gets built. And like most infrastructure shifts, it’s quiet until it’s everywhere.
This month didn’t bring a single breakthrough. It brought alignment: working chips, wafer-scale manufacturing, early product deployment, and investor interest. The gap between research and adoption got narrower. Fast.
We’ve optimized electrons for decades. Now we’re learning to work with light. The technology is catching up to the need—and the infrastructure is starting to adapt.