Nvidia will add two new chips to its lineup over the next two years, as the company scales up its output amid strong demand for artificial intelligence technology.
Blackwell Ultra, a family of graphics processing units, will be released later in 2025. Its Vera Rubin chip will launch in 2026, offering significant performance upgrades on AI-related tasks.
“AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance,” said Nvidia CEO Jensen Huang.
“We designed Blackwell Ultra for this moment — it’s a single versatile platform that can easily and efficiently do pretraining, post-training and reasoning AI inference.”
The Ultra chips will include nearly 100 gigabytes of additional memory over the standard Blackwell chips. It will retain the standard chips’ capacity of 20 petaflops, or floating-point operations per second, while running AI models.
Nvidia’s Blackwell Ultra chips will be released in the second half of 2025, the company said.
Vera Rubin, set for a 2026 release, is capable of 50 petaflops of AI performance. This is well above the current 20 petaflops of Nvidia’s existing Blackwell chips.
The Vera Rubin chips’ central processing unit will be around twice as fast as the Blackwell chips. A Rubin Ultra chip will follow in the second half of 2027.
Another new chip family, called Feynman, will be available in 2028.
The company also announced two new desktop AI computers, known as the DGX Spark and DGX Station. The Station will include one Blackwell Ultra chip, while the Spark features a GB10 Blackwell Superchip.
Huang revealed these new chips at Nvidia’s annual GTC conference. The company said in May 2024 that it would aim to release a new chip family each year, twice as frequently as its previous rate.
Nvidia’s (NASDAQ: NVDA) share price closed at US$115.43, down from its previous close at $119.53. Its market capitalisation is $2.82 trillion.
Related content
