Austin-based startup Neurophos has secured $110 million in a Series A funding round to develop ultra-efficient optical processing units (OPUs). The company, which spun out of Duke University and the Metacept incubator, aims to solve the AI industry’s most pressing challenge: scaling compute power without triggering unsustainable energy consumption.
From Metamaterials to AI Breakthroughs
The technology roots back to research on metamaterials—engineered substances capable of manipulating light in unconventional ways. While early applications focused on experimental “invisibility cloaks,” Neurophos has pivoted this physics foundation toward a “metasurface modulator.” This component functions as a tensor core processor, capable of performing matrix-vector multiplication—the mathematical engine behind AI inferencing—far more efficiently than traditional silicon-based GPUs.
Solving the Efficiency Bottleneck
Traditional photonic chips have historically struggled with size constraints and the energy-intensive need for digital-to-analog data conversion. Neurophos claims its proprietary metasurface design is 10,000 times smaller than conventional optical transistors. This extreme miniaturization allows thousands of units to be packed onto a single chip, drastically increasing calculation density.
“If you want to go fast, you have to solve the energy efficiency problem first,” explains Dr. Patrick Bowen, CEO and co-founder of Neurophos. “By shrinking the optical transistor, we can perform more math within the optics domain before requiring a conversion back to electronics.”
Performance vs. Nvidia’s Blackwell
Neurophos is positioning its OPU as a direct competitor to industry-standard hardware, including Nvidia’s B200 GPU. According to company data, the Neurophos chip is designed to run at 56 GHz, delivering a peak performance of 235 Peta Operations per Second (POPS) while drawing 675 watts. For comparison, the B200 is rated at 9 POPS with a 1,000-watt power draw.
The startup intends to leverage standard silicon foundry processes to ensure mass-market manufacturing viability, aiming for a commercial launch by mid-2028.
Strategic Backing and Future Roadmap
The Series A round was led by Gates Frontier, with significant participation from Microsoft’s M12, Carbon Direct, Aramco Ventures, Bosch Ventures, Tectonic Ventures, and Space Capital. The capital will be utilized to build out a full software stack, develop data center-ready OPU modules, and expand engineering operations in San Francisco and Austin.
Dr. Marc Tremblay, corporate vice president of core AI infrastructure at Microsoft, emphasized the urgency of this innovation: “Modern AI inference demands monumental amounts of power and compute. We need a breakthrough in compute on par with the leaps we’ve seen in AI models themselves.”
While the company faces a competitive landscape dominated by entrenched players like Nvidia and specialized firms like Lightmatter, Bowen remains confident that their 50x advantage in speed and efficiency over current architectures provides a substantial moat for the future of AI infrastructure.