SkyDeck Dinner Series: Hexabitz x Berkeley Grad Students
Berkeley grad students: come have dinner with Asaad Kaadan, Cofounder and CEO of Hexabitz (Berkeley SkyDeck Batch 23). Hexabitz is building a new kind of compute. Instead of server racks, we're building resilient, solar-powered compute sheets for terrestrial datacenters, space, defense, and harsh, off-grid environments.
Hexabitz models & simulations show that our Nature Inspired Modular compute Sheets (NIMS) can achieve up to 10x better throughput/$ compared to a highly optimized GW-scale B300 datacenter running large Mixture of Expert (MoE) models like Kimi K3. With the anti-datacenter movement spreading across the country, and the massive environmental impact GW-scale datacenters have on grid and water resources, our NIMS technology has the potential to solve one of the largest societal and technological challenges facing the US currently. We expect to be able to deploy GW-scale compute infrastructure in matter of weeks instead of years given our infrastructure-free architecture, resilient supply chain, and our planned fully automated manufacturing.
Our stack of technologies is so wide and diverse it spans AI inference optimization, custom silicon design, optical interconnects, distributed computing, advanced materials and silicone encapsulants, printed electronics and advanced integration, robotic and automated manufacturing, and space technologies, among others.
Bringing an entirely new class and formfactor of computing to market is no easy feat but it's a fun and challenging journey! We're looking for the most ambitious researchers, engineers, and tinkerers to join us on this journey. You will be working on the technology that might become someday the defacto computing architecture to embed intelligence everywhere and scale humanity to the next level.
If you're intrigued, bring your questions, ideas, and be ready for a geeky conversation! This is a small dinner with founders building a real product, so bring your own questions, open-source projects, or any crazy ideas you have about the future of computing. We're particularly interested with students obsessed about squeezing AI performance out of distributed edge devices from speculative decoding to smart resource management to adaptive model partitioning and other interesting ideas. However, if you also have any other skills that might fall in our wide technology stack, feel free to bring them and chat. Spots are limited, so RSVP below.