The new optics-based memory architecture enables inference speeds of up to 10,000 tokens per second per user for AI models exceeding 10T parameters. Volantis raised $88 million in Series A funding to develop these chips, which remove memory bottlenecks by increasing both bandwidth and capacity per chip. The current system already transmits data more than 10 times farther than electrical wires of equivalent size inside a chip package.

The company's technical team has a history of developing core semiconductor technologies, including the first CoWoS product and early HBM. Volantis will first apply its technology to AI coding agents designed to finish tasks in seconds rather than hours. The startup has already taped out a next-generation iteration that targets a world record for bandwidth density over relevant distances.