Latest news from Volta
Updates, insights and announcements from the team building next-generation AI infrastructure.
Designing a non-blocking fabric for 100,000 GPUs
At 100,000 GPUs the network is the computer – one congested link can idle the entire cluster. How we design a non-blocking, rail-optimised fabric, and why the real limits are radix, the NVLink bandwidth cliff and constant failure.
Volta bets on owning the grid to win the AI compute race
Bloomberg Technology on how Volta's substation-to-FLOP model is reshaping the economics of frontier AI compute.
Introducing Volta: The First AI-Native Hyperscaler
Intelligence has outgrown the ground it runs on. AI's binding constraint is no longer the GPU but the power, grid and cooling beneath it – which is why compute has to be built and owned as a single utility, from the substation to the FLOP.