Two startups are betting that optical metamaterials — the same physics behind "invisibility cloaks" — could be the key to solving the bandwidth crisis threatening AI data center expansion.
The Bandwidth Bottleneck
AI data centers face a critical infrastructure challenge. Conventional electronic switches that connect servers require constant conversion between light and electrons, creating both bandwidth limitations and excessive power consumption. As AI workloads scale, these bottlenecks become increasingly problematic.
"There's no market for" optical cloaks, noted Patrick Bowen, co-founder and CEO of photonic computing startup Neurophos in Austin, Texas. "Now companies are devising more practical uses for the science behind cloaks."
Lumotive's Programmable Light Steering
On March 19, 2026, Lumotive unveiled its new microchip featuring programmable liquid crystal elements built using standard semiconductor manufacturing techniques. The device is covered with copper structures, with liquid crystal elements sandwiched between them — similar to LCD displays but with electronically programmable optical properties.
The chip can precisely steer, lens, shape, and split beams of light reflected off its surface, performing functions that would normally require multiple optical components. Critically, it has no moving parts, addressing the reliability issues that plague microelectromechanical systems (MEMS).
"Having no moving parts significantly improves reliability," said CEO Sam Heidari.
Lumotive's chips currently support the industry standard of 256 by 256 ports but claim scalability to 10,000 by 10,000. The company plans to launch its first commercial optical switches by the end of 2026.
"We think this is game-changing for data centers," Heidari stated. The technology bypasses silicon photonics' energy efficiency problems and MEMS reliability issues simultaneously.
Beyond Optical Switching
Meanwhile, Neurophos is pursuing broader applications of metamaterial technology for optical computing, targeting what Bowen calls "transformative" potential for artificial intelligence infrastructure.
The Power Grid Challenge
Bandwidth isn't the only infrastructure concern. Experts are increasingly calling for development of an independent AI power grid to handle the industry's escalating energy demands. As computational requirements grow, power delivery and grid reliability emerge as parallel bottlenecks alongside bandwidth constraints.
The convergence of optical switching innovation and power infrastructure planning reflects the industry's recognition that AI scaling requires solutions across multiple infrastructure layers simultaneously.
What Comes Next
With Lumotive targeting end-of-2026 for product launch and Neurophos advancing optical computing research, 2026 could prove pivotal for metamaterial adoption in AI infrastructure. The technology's success will depend on achieving commercial viability — balancing cost, reliability, and manufacturing scale at data center volumes.