Volantis raises $88M Series A to build photonic AI inference system

Volantis claims its A-1 system solves a ‘memory wall’ in AI hardware by combining high memory capacity and bandwidth—nearly two orders of magnitude better than current solutions. The company says it can run models with over 20 trillion parameters at 10,000 tokens per second per user, cutting inference costs per token. Benchmarks against NVIDIA’s Rubin highlight 15x better tokens per dollar and 30x lower latency for large-model inference. The system uses photonic interconnects to link compute chips and memory, scaling to 240 TB/s bandwidth with 10 TB memory capacity in a 15U datacenter rack.