One of many greatest tales this week within the HPC-AI world entails — shock! — NVIDIA. Truly, make that two or three of the largest tales. One is NVIDIA’s stellar quarterly earnings announcement within the face of issues in current months about disruption of chip exports of GPUs to China.
The corporate shrugged off what it stated a couple of weeks in the past can be a $5 billion cost and charged forward with revenues ($44.1 billion — together with $39.1 within the knowledge middle market — up from $26 billion a yr in the past), and earnings ($18.8 billion, up 26 p.c YoY).
That the corporate largely exceeded expectations didn’t come as an entire shock contemplating the big and growing demand for AI compute to produce mammoth AI factories being stood up world wide, together with OpenAI’s AI knowledge middle deliberate for Abilene, TX, which is slated to deal with as much as 400,000 NVIDIA top-of-the-line Blackwell chips.
Nonetheless, challenges exist for NVIDIA, one in all which is the China AI market, which has grow to be more and more walled off to U.S. expertise distributors by the U.S. authorities. In an interview final evening with CNBC’s Jim Cramer, NVIDIA CEO Jensen Huang made the case for continued gross sales of his firm’s GPUs to China, the place he stated 50 p.c of AI builders reside. And he needs them to develop their AI functions on NVIDIAs CUDA growth platform to run on NVIDIA {hardware}.
“American expertise stacks will run AI the most effective all around the world, and so that is a very powerful strategic motive to be in China, as a result of there are such a lot of builders there,” Huang stated. “And since the world goes to undertake expertise from one nation or one other, then we choose to be it on the American expertise stack.”
Earlier this yr, exports to China of NVIDIA’s pared-down H20 AI GPU, beforehand allowed underneath U.S. commerce guidelines, had been closed to exports by the Trump White Home, forcing NVIDIA to take the $5 billion cost. This raises one other main story rising this week, that NVIDIA and AMD are engaged on growing AI chips for the China AI market that adjust to U.S. commerce restrictions.
In response to a narrative in DigiTimes, the 2 corporations have “rapidly adjusted their designs to diminished specs, and it’s anticipated that beginning in July 2025, they’ll launch a brand new wave of downgraded, however compliant, AI GPUs that may be bought to China.”
In response to the story, NVIDIA has tentatively named its China AI chip the B20, whereas AMD is transferring forward with the Radeon AI PRO R9700 and different merchandise. “Each can help the operation of fashions corresponding to DeepSeek,” DigiTimes reported.
NVIDIA’s anticipated B20 adopts the Blackwell structure however with main alterations in reminiscence and computing energy efficiency, in line with DigiTimes. AMD’s Radeon AI PRO R9700 and different merchandise are designed for AI workstations to speed up native inference, mannequin fine-tuning and different data-intensive workflows, in line with the publication.
As of 2021, NVIDIA’s market share in China was roughly 90 p.c, but it surely has now dropped to 50 p.c.
