photonics AI News & Updates
Neurophos Raises $110M for Optical AI Chips Claiming 50x Efficiency Over Nvidia
Neurophos, a Duke University spinout, has raised $110 million led by Gates Frontier to develop optical processing units using metamaterial-based metasurface modulators for AI inferencing. The startup claims its photonic chips will deliver 235 POPS at 675 watts compared to Nvidia's B200 at 9 POPS at 1,000 watts, representing a claimed 50x advantage in energy efficiency and speed. Production is expected by mid-2028 using standard silicon foundry processes.
Skynet Chance (+0.01%): More efficient AI hardware could enable larger-scale deployment of AI systems and reduce barriers to running advanced models, potentially increasing proliferation risks. However, the technology is primarily focused on inferencing rather than training, limiting its impact on developing fundamentally more capable systems.
Skynet Date (+0 days): If successful, dramatically more efficient inference hardware could accelerate AI deployment timelines by reducing cost and power barriers, though the 2028 production target limits near-term impact. The technology addresses scaling bottlenecks that currently constrain widespread AI system deployment.
AGI Progress (+0.03%): Breakthrough hardware efficiency could enable more complex AI architectures and larger-scale continuous learning systems that are currently constrained by power and cost. Removing compute bottlenecks historically accelerates progress in AI capabilities by enabling new research directions.
AGI Date (-1 days): A 50x improvement in inference efficiency could significantly accelerate AGI timelines by making continuous learning, massive-scale deployment, and more complex architectures economically viable. However, the 2028 production timeline and focus on inference rather than training moderates the near-term acceleration effect.