AI Prefill: The Workload Optical Computing Was Waiting For | Neurophos × Geometry

Optical computing has been “almost here” for decades—and many well-funded attempts have failed. So why should this time be different? Geometry’s Ben Levy sits down with Patrick Bowen of Neurophos to work through that question from first principles. They discuss how metamaterials let engineers design new optical properties, why the enormous size of conventional optical modulators held earlier systems back, and how Neurophos shrank those components by roughly 10,000×. The conversation moves from the physics to the market: why optical computing is about amortizing computation rather than simply using the speed of light; why Neurophos is focused on AI prefill; and whether specialized optical processors can help relieve the power and scaling constraints facing AI data centers. Geometry is an investor in Neurophos.