Neurophos Secures $110M Funding: Pioneering Tiny Optical Processors for AI with Innovations Like Invisibility Cloaks

Twenty years ago, visionary researcher David R. Smith from Duke University unveiled something extraordinary: a real-life invisibility cloak crafted from innovative materials known as metamaterials. While it didn’t quite reach the magical realm of Harry Potter’s creation, it marked a pivotal moment in material science and led to significant advancements in electromagnetism research. Fast forward to today, and Neurophos, an Austin-based startup born from that legacy, is championing an evolution in technology that promises to address a major challenge in artificial intelligence (AI) and hyperscale computing.

The Breakthrough: Metasurface Modulators

Neurophos has developed a game-changing “metasurface modulator”, designed to function as a powerful tensor core processor. This device enables rapid matrix-vector multiplication—a fundamental operation at the heart of AI tasks, particularly during inferencing. Currently, these computations are handled by specialized GPUs and TPUs utilizing traditional silicon technology. However, by embedding thousands of these modulators onto a single chip, Neurophos claims its optical processing unit outpaces current silicon GPUs significantly, all while enhancing efficiency during the often costly inferencing phase.

Funding the Future

To advance the development of its innovative chips, Neurophos recently secured $110 million in a successful Series A funding round. This investment, led by Gates Frontier (the venture firm of Bill Gates), also saw participation from notable players such as Microsoft’s M12, Carbon Direct, Aramco Ventures, and more.

The Promise of Photonic Chips

Although photonic chips aren’t a novel concept, they promise considerable advantages over traditional silicon-based processors. Theoretically, they deliver superior performance due to several factors:

• Reduced Heat Generation: Light produces less heat compared to electricity.
• Increased Speed: Photonic signals travel faster than electrical signals.
• Temperature Resilience: They are less influenced by temperature fluctuations and electromagnetic fields.

Overcoming Challenges

Historically, however, optical components have faced challenges related to size and mass production. They typically require converters to transition data between digital and analog formats—which can be bulky and consume considerable power.

Neurophos’s innovative metasurface aims to tackle these issues head-on. Remarkably, it boasts a size approximately 10,000 times smaller than conventional optical transistors, allowing thousands of units to fit on a single chip. This compactness results in efficiency levels that traditional silicon simply can’t match, enabling multiple calculations to occur simultaneously.

A Vision for Efficiency

Dr. Patrick Bowen, CEO and co-founder of Neurophos, shared insights on the importance of energy efficiency in chip design. He emphasized, “If you want to go fast, you have to solve the energy efficiency problem first. Making a chip 100 times faster typically results in a 100-fold increase in power consumption.” This strategic approach positions Neurophos not just for speed, but sustainable performance.

The result? An optical processing unit that claims to dramatically outperform Nvidia’s B200 AI GPU. With a peak capability of 235 Peta Operations per Second (POPS) at 56 GHz, it uses just 675 watts compared to the 1,000 watts needed by the B200 to achieve only 9 POPS.

Facing a Competitive Landscape

Despite its promising technology, Neurophos enters a highly contested market dominated by companies like Nvidia—recognized as the most valuable public company globally. Additionally, other firms in the photonics space are exploring their own avenues, and some, like Lightmatter, have shifted their focus toward interconnects. It’s important to note that Neurophos anticipates its first production chips won’t be ready until mid-2028.

However, Bowen remains optimistic about the future. He argues that while competitors focus on evolutionary improvements in silicon technology, Neurophos is pursuing a revolutionary path. “By the time we launch, we will have substantial advantages over existing products, as we start from a base of 50 times greater energy efficiency and speed.”

Manufacturing Innovations

In tackling traditional manufacturing hurdles associated with optical chips, Neurophos claims its products can be fabricated using standard silicon processes and tools. This approach enhances the potential for scalability as the company prepares to develop its first integrated photonic compute systems, inclusive of data center-ready optical processing unit (OPU) modules and a complete software stack.

Dr. Marc Tremblay, corporate vice president and technical fellow at Microsoft, aptly noted the importance of Neurophos’s innovations: “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, which is what Neurophos’ technology and high-talent density team is developing.”

As we look ahead, it’s clear that Neurophos is positioned at the forefront of an exciting shift in technology.

Are you ready to embrace the future of AI and optical computing? Share your thoughts with us and let’s embark on this journey together.