Invisible Cloaks and AI Power: How Neurophos is Revolutionizing Computing with Metasurface Modulators
In the world of Harry Potter, an invisibility cloak is a magical device that allows its wearer to become, well, invisible. But in the world of materials science, a team of researchers at Duke University, led by Professor David R. Smith, created something almost just as impressive: a real-life invisibility cloak using synthetic composite materials called metamaterials.
Fast forward to today, and a startup called Neurophos is building upon that research to tackle one of the biggest challenges facing AI labs and hyperscalers: scaling computing power while keeping energy consumption in check. The company’s solution? A metasurface modulator, an optical device that can function as a tensor core processor, doing matrix vector multiplication – a fundamental math problem at the heart of many AI tasks.
Neurophos has raised $110 million in a Series A round to develop its “optical processing unit” (OPU), a chip that can cram hundreds of these modulators onto a single piece of silicon. According to the company, this OPU is significantly faster and more efficient than silicon GPUs currently used in AI data centers.
So, how does it work? Well, unlike traditional photonic chips, Neurophos’s metasurface modulators are incredibly small – about 10,000 times smaller than conventional optical transistors. This makes them much easier to mass-produce and integrate into a single chip.
“We’re solving the power efficiency problem first,” says Dr. Patrick Bowen, CEO and co-founder of Neurophos. “If you want to go fast, you have to solve the power efficiency problem. Because if you take a chip and make it 100 times faster, it burns 100 times more power. So you get the privilege of going fast after you solve the power efficiency problem.”
The result is an OPU that can wildly outperform Nvidia’s B200 AI GPU. The company claims its chip can run at 56 GHz, yielding a peak 235 Peta Operations per Second (POPS) and consuming 675 watts, compared to the B200, which would ship 9 POPS at 1,000 watts.
But here’s the thing: Neurophos isn’t just touting its speed and efficiency – it’s also tackling the mass-manufacturing challenges optical chips have historically faced by using standard silicon foundry materials, tools, and processes.
The company plans to use this funding to develop its first integrated photonic compute system, including datacenter-ready OPU modules, a full software stack, and early-access developer hardware. And it’s expanding its team, too – opening a new engineering site in San Francisco and growing its HQ in Austin, Texas.
As Microsoft’s Dr. Marc Tremblay notes, “Modern AI inference requires enormous amounts of energy 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.”
While Neurophos is entering a crowded market dominated by Nvidia, the company is confident its efficiency and performance advances will provide a sufficient moat. And with its recent funding, it’s well-positioned to make a splash in the world of AI computing.
