Julia Lab algorithms to accelerate Neurophos photonic AI chips. Pairing advanced algorithms with photonic hardware to improve AI inference speed and energy efficiency. 01 ANNOUNCEMENT CAMBRIDGE, Mass. and AUSTIN, Texas, Oct. 21, 2026. The Julia Lab, a research group within MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), and Neurophos, a company building photonic AI processors, today announced a collaboration to push the frontier of high-performance computing using a light-based chip. The collaboration combines advanced mathematical approaches with hardware to make computing far more efficient. Julia Lab's numerical methods and efficient algorithms will run on Neurophos' Optical Processing Unit (OPU), which contains millions of optical processing elements on each chip. I've spent my career thinking about matrices, and it's thrilling to see light turn out to be so good at multiplying them. Matrix multiplication is the heartbeat of scientific computing and AI. I'm especially proud that this partnership with Neurophos grew out of its own lab, with Jake Chuharski now leading their work on making algorithms fit the hardware. Compilers have always translated mathematics into transistors; now Neurophos, Inc. get to translate it directly into light. ALAN EDELMAN MIT PROFESSOR · CSAIL PRINCIPAL INVESTIGATOR While GPUs use electricity to perform AI computations, OPUs also use light to carry out core matrix operations. With Neurophos' stated goal of cutting the resources used during AI compute by roughly 50 times with their first chip, the resource cost of doing the computation (in both energy and space) could decrease by over an order of magnitude, while also running faster than standard processors. They plan to release their systems commercially in 2028. On the MIT side of the collaboration is the Julia Lab, which was built on the Julia programming language created by Jeff Bezanson, Stefan Karpinski, Viral Shah, and Edelman. They focus on scientific computing and numerical algorithms, having made significant contributions to numerical linear algebra and high-performance computing. With OPUs, they can run efficient algorithms on efficient hardware. "We're teaming up with the Julia Lab because meaningful leaps in computing rely as much on numerical methods and efficient algorithms as the state-of-the-art hardware itself," says Jake Chuharski '24, MEng '25, a Julia Lab researcher leading Neurophos' quantization and optimization team. The collaboration reflects a shared focus on improving computing efficiency through advances in both software and hardware. "We're working to make computing with light over an order of magnitude faster and more energy efficient than today's best digital accelerators," says Neurophos CEO and co-founder Patrick Bowen. In a recent talk at the GlobalFoundries Tech summit, Bowen outlined how the company's optical architecture could deliver substantial gains in speed and energy efficiency over traditional GPUs. Like Google's Tensor Processing Units (TPUs), Neurophos' OPU is designed around a systolic array, an architecture specialized for the matrix calculations at the foundation of AI inference. The company's breakthrough in active metasurface technology enables optical modulators roughly 10,000 times smaller than conventional Mach-Zehnder devices, allowing millions of light-controlling elements to fit within a 5-by-5 millimeter area. On top of making computing systems perform faster, OPUs could help ease infrastructure demands of AI. By delivering more computation within a given power and space budget, OPUs will allow data centers to support larger workloads while reducing energy per calculation. Cloud providers could then help meet growing demand with fewer additional facilities and less expansion of power and cooling infrastructure than would otherwise be needed. About Neurophos. Neurophos Inc. is an Austin-based semiconductor company developing high-performance, energy-efficient photonic AI inference chips. Founded by Dr. Patrick Bowen and Dr. Andrew Traverso, the team includes industry veterans from NVIDIA, Apple, Samsung, Intel, AMD, Meta, ARM, Micron, Marvell, Mellanox, Lightmatter, and more. For details, visit www.neurophos.com.

Neurophos has raised $110M in a funding round led by Gates Frontier to build optical computing technology. The Austin, Texas, company uses silicon photonics circuits and metamaterials to run computations directly in the optical domain. M12 (Microsoft's venture fund), Carbon Direct Capital, Aramco Ventures, Bosch Ventures, Tectonic Ventures, and Space Capital also took part. The funding let the company quadruple its workforce in 2026. The approach targets the energy cost of data centres, where converting light to electrical signals and back wastes power. A hyperscale data centre uses 100MW to 1GW, about 50% of it for compute. "Moore's Law is slowing, but AI can't afford to wait," said chief executive officer and co-founder Dr. Patrick Bowen. Yole Group expects the first optical computing units to deploy in 2028, with the market reaching about $2.7 billion by 2034.