Volantis 宣布 8800 万美元 A 轮融资，累计融资 9700 万美元，投资人包括 Sam Altman、Jeff Dean 和 John Doerr，目标是在超过 10T 参数的模型上实现每用户 10,000 tokens/秒的推理。
This team is working to bring a MASSIVE 10,000 tokens/second/user on models above 10T parameters.
Volantis wants to cut coding-agent runs from 30 minutes to 30 seconds by feeding AI chips memory over light.
Sam Altman, Jeff Dean and John Doerr are among the backers of the $97M it has raised so far.
A single HBM memory stack is about 11mm long, longer than copper can reach. That limit is why Nvidia's best GPUs fit only 8 stacks each. Volantis uses light to wire over 220 memory chips into a single shared pool.
Volantis builds its optical waveguides directly into the chip package instead of using fiber cables. Each waveguide is over 2500x smaller than optical fiber. That density packs tens of thousands of light channels into the package.
Volantis claims well under 1 picojoule per bit, under 5ns latency and over 200TB/s in total.
Today's AI chips force a choice between memory capacity and memory speed. SRAM moves data very fast but holds too little to fit giant models, while HBM holds far more but moves data slower, which caps tokens per second.
But Volantis is connecting hundreds of memory chips over light, and their bandwidth adds up as each joins.
So Capacity and speed grow together here.
Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: https://volantissemi.ai/news-insights/our-88m-series-a-demolishing-the-memory-wall-with-photonics-post在 X 查看被引用的帖子
来源：Rohan Paul · x.com