Zettascale Computing

Zettascale Computing is a Y Combinator backed company designing AI silicon, which it calls the XPU, for training and inference of large models at lower energy use. Its premise is that most energy in AI goes to moving data, so the chip is designed to keep data close to compute. The company says the first design, XPU Grasshopper, was co designed with AI. The product is hardware and its software stack, not a hosted coding assistant. XPU Grasshopper currently runs end to end inference on an FPGA prototype (VU47P) built in house, and it is not generally available. Pre orders are open, and the vendor lists a serving API as coming soon. XPU Frontends let developers keep PyTorch, tinygrad or JAX, with a short import and a call to move a model onto the XPU, and early access can be requested. Monolith is a planned hosted cluster of XPUs that behaves as one chip, rolling out in batches starting with early access ticket holders. For low level work, libxpu is a C interface over the whole machine, giving control of every buffer and data movement from any language that can call C, and it is planned to be fully open source. The audience is AI infrastructure teams and researchers working on training and inference workloads that depend on power efficiency. No prices are published for pre orders or early access. The site's job listings mention salary ranges for hardware engineers, and those are not product prices.