Another AI chip maker has secured millions of dollars in funding. The startup, Neurophos Inc., just completed an oversubscribed early-stage funding round today, raising $110 million, bringing its total funding to date to $118 million.
Gates Frontier led the investment
The Series A funding round was led by Gates Frontier, with participation from numerous other investors, including Microsoft's venture capital arm M12, Carbon Direct Capital, Saudi Aramco Ventures, Bosch Ventures, Tectonic Ventures, Space Capital, DNX Ventures, Geometry, MetaVC Partners, Morgan Creek Capital, Silicon Catalyst Ventures, and Gaingels.
Like many other chip startups, Neurophos states that it is committed to addressing the growing shortage of computing power needed for the widespread adoption of artificial intelligence technologies in the coming years. The company explains that data centers face critical limitations in terms of total computing power and scalability, while also consuming enormous amounts of energy, and believes that the current trend of racing to build more data centers is not the solution.
Neurophos' solution is a novel AI acceleration chip called an "Optical Processing Unit" (OPU), which integrates over one million micrometer-scale optical processing elements onto a single chip. The company says its prototype chip is up to 100 times more powerful than existing AI processors, and will ultimately provide data center operators with a more powerful plug-and-play solution to replace the graphics processing units they currently rely on.
Neurophos founder
Neurophos is led by its co-founder and CEO, Patrick Bowen. He stated that the artificial intelligence industry can no longer wait for Moore's Law to meet its enormous demand for computing power. "Our breakthrough in photonics, by integrating massively parallel optical computing onto a single chip, has opened up a whole new dimension of scalability," he said. "This physical revolution means that as scale increases, both efficiency and raw speed will improve, breaking through the power consumption limitations of traditional GPUs."
The startup's core innovation lies in developing proprietary micrometer-scale metamaterial optical modulators, 10,000 times smaller than existing photonic elements, making photonic computing a reality for the first time. The company combines these modulators with in-memory computing technology, overcoming traditional hardware bottlenecks by fusing memory and processing power, thereby accelerating matrix multiplication operations—the core of AI—and reducing data transfer.
Neurophos's chip uses photons to bypass the speed and energy limitations of electrons, achieving clock frequencies exceeding 100 gigahertz. In other words, it is developing a completely new AI accelerator that promises significant efficiency improvements over GPUs. In early tests, the chip has demonstrated remarkable performance, exceeding 300 trillion operations per watt per second, far surpassing existing standards. The company states that its chip design also significantly reduces power consumption.
Partnership with operator Terakraft
This startup is partnering with Norwegian data center operator Terakraft to launch a pilot project for its optical AI accelerator in 2027. The company says it hopes to manufacture its first complete systems by early 2028 and scale up production later that year. M12 Managing Partner Michael Stewart said the timeline is realistic enough and he would be happy to fund it.
Holger Mueller of Constellation Research said investors are willing to pour money into promising chip startups because the anticipated demand for AI computing presents unprecedented opportunities that Nvidia cannot meet alone. The analyst noted that many companies are seeking faster, more economical ways to run their AI models as the scarcity of Nvidia GPUs continues to drive up prices.
"There's a lot of innovation emerging from startups, and Neurophos is a good example, working to integrate millions of light processing units onto a silicon chip," Mueller said. "This funding will help Neurophos achieve its next major milestone and prove its ability to mass-produce and operate Optical Processing Units (OPUs). The success or failure of these efforts will help determine whether OPUs can become as ubiquitous as GPUs."
Microsoft has become one of the startup’s most loyal supporters. It not only financially backs the company but also actively explores the advantages of its OPUs. "Modern AI inference requires massive amounts of computing power," said Mark Tremblay, Microsoft's Vice President and Core AI Infrastructure Technology Specialist. "We need breakthroughs in computing power to match the leaps in AI models themselves, and that's exactly what Neurophos is working on with its high-caliber technology and talent team."
Neurophos stated that the funds raised in this round will accelerate the delivery of its first integrated photonic computing system, which includes OPU modules for data centers, a complete software stack, and early developer hardware. In addition, the company plans to expand its headquarters in Austin, Texas, and open a new engineering center in San Francisco to showcase its technology to potential customers.
Source: Compiled from siliconangle
