Nvidia shows its plans for future AI chips. Hopper’s successor, Blackwell, is already in development and the chip designer is also planning an important announcement for 2025.

In 2024, Nvidia Blackwell will succeed the current generation of (very successful) Hopper chips. The chip designer himself announced this during a presentation for investors. Nvidia will release the Hopper H200 for the first time next year. This chip remains based on the current architecture and will receive derivatives with the GH200 series for ARM-based inference and L40S for classic inference.

Blackwell

Blackwell B100 will follow later in the year. The chip series takes its name from the American statistician and mathematician David Harold Blackwell. The chip would be based on TSMC’s 3nm process and in turn receive derivatives for ARM and c86-based workloads. Nvidia wants to bring Blackwell onto the market and deliver it to customers next year, although we wouldn’t be surprised if availability remains very limited initially. After all, there were many months between the introduction of the Hopper H100 and its availability on (cloud) servers.

Nvidia is also looking further into the future. The X100 is planned for 2025. It is somewhat questionable whether the AI ​​specialist really plans to follow B100 with X100 after a year. Traditionally, Nvidia aims for a two-year update cycle for its powerful AI GPUs. It is also not clear which mathematician the “X” in the name represents.

Nvidia isn’t losing sight of its Infiniband and DPU products either. New quantum chips and Spectrum-X accelerators are also on the roadmap for 2024 and 2025.