Dipole Labs

Dipole Labs (YC S26) builds high-speed optical circuit switches for AI datacenters. Between racks, data already travels as light, but routing it normally means converting to electrical signals and back, which burns power, adds heat and latency; Dipole's switch keeps traffic in the optical domain, targeting sub-microsecond reconfiguration at large port counts. An AI-native control layer learns each workload's GPU communication pattern and reconfigures the optical network in step with training and inference, so the network acts as an accelerator rather than a bottleneck. The company claims this could make a 10,000-GPU cluster perform like one with 12,500 GPUs — up to 25% more useful compute from the same hardware, or up to $1 billion of added capacity a year at 1 GW scale — which are company estimates, not measured results. It also describes high-speed spatial light modulators and says the platform extends from moving data with light to computing with it, with applications in quantum computing. Founded in 2026 by CEO Deepankur Thureja (physics PhD at ETH Zürich, where his thesis earned the ETH Medal; scalable quantum-optical hardware published in Nature) and CTO Gabriele Pasquale (applied physics and materials science PhD at EPFL; IBM prize from the Swiss and Austrian Physical Society), who met as postdoctoral researchers at Harvard. Headquartered in Boston with a European base in Zurich. It is seeking design partnerships with hyperscalers and neoclouds and collaborations with photonic foundries. No funding amount or valuation has been disclosed beyond YC backing.