Volantis Semiconductor raised an $88M Series A, bringing total funding to $97M, to attack AI’s memory bottleneck with photonics. Backers include: OpenAI's Sam Altman SemiAnalysis'S Dylan Patel Jeff Dean John Doerr and others The company is building an optical “fabric” designed to connect AI processors to much larger pools of memory while dramatically increasing bandwidth. Volantis says its architecture can boost both memory capacity and bandwidth by more than 10x, targeting inference for models larger than 10 trillion parameters at speeds of up to 10,000 tokens per second per user. The core problem it is trying to solve is the “memory wall.” Volantis argues today’s architectures force a tradeoff: on-chip SRAM offers very high bandwidth but limited capacity, while HBM-based GPUs and TPUs offer far more memory but eventually run into bandwidth constraints as models scale. Its approach replaces more of the electrical data movement with optics, using custom integrated micro-VCSEL lasers and eliminating traditional optical fiber to push much more data across short distances inside AI systems. The company says it has already transmitted data more than 10x farther than similarly tiny electrical wires inside a chip package, and its next-generation design has already taped out. The near-term use case is much faster inference. Volantis says a coding agent that currently takes around 30 minutes could potentially complete the same task in roughly 30 seconds, while also lowering cost per token through higher throughput and cheaper pooled memory. If the technology scales as planned, the bigger idea is enabling much larger context windows, entire code repositories in context, and AI systems that are currently limited more by memory movement than raw compute.