Bill Gates Backed Startup Unveils Light-Powered AI Chip That Could Challenge Nvidia

US startup Neurophos, backed by Bill Gates’ Gates Frontier Fund, has unveiled a new class of artificial intelligence processor that uses light instead of electricity to perform computations an approach the company says could dramatically outperform today’s power-hungry AI chips.

The announcement, first reported by Tom's Hardware, has sparked interest across the semiconductor industry, which is increasingly searching for alternatives to conventional silicon-based computing as AI workloads strain power and thermal limits.

How Optical Computing Changes the Equation

Unlike traditional GPUs that rely on electrons moving through silicon transistors, Neurophos’ processor performs calculations using photons. Because light can switch at much higher speeds and generates significantly less heat, optical computing has long been viewed as a potential leap forward for AI performance and efficiency.

Neurophos claims it has overcome a key historical bottleneck in optical computing by shrinking photonic components enough to be densely packed on a chip manufactured using existing fabrication processes. This enables a single, massive optical compute matrix operating at extremely high speeds—rather than hundreds or thousands of smaller electronic cores.

A Different Path From Nvidia

Today’s AI hardware market is dominated by Nvidia, whose GPUs remain the backbone of data centres powering large language models and generative AI. While Nvidia has begun incorporating photonics to speed up data transfer between chips, the actual computation still relies on electronic silicon.

Neurophos is pursuing a more radical approach: true optical computation, where light performs the core mathematical operations. If successful at scale, this could represent a fundamental shift in how AI hardware is designed.

Is Nvidia’s Lead Under Threat?

Industry analysts caution that Nvidia’s dominance is not under immediate threat. Neurophos’ technology is still several years away from mass production and faces significant challenges, including manufacturing yields, software compatibility and integration with existing AI frameworks.

Even so, the project highlights a possible future in which optical processors complement or eventually replace parts of today’s electronic AI stacks. As data centres grapple with rising energy costs and physical scaling limits, such alternatives are gaining attention.

For now, Neurophos represents an ambitious bet on the next decade of AI computing—one that underscores how the race for post-silicon innovation is accelerating alongside the global AI boom.