Source: Finance Association
① At the Nvidia earnings conference, CEO Hwang In-hoon revealed that after Blackwell, Nvidia is developing a new chip and releasing it once a year; ② Faced with questions about current products that may be slow to sell, Hwang In-hoon said that the faster the update will only make the company have more demand for chips.
$NVIDIA (NVDA.US)$The earnings report ignited Wednesday's stock price. Revenue and profit surpassed expectations, driving the entire market to look forward to the artificial intelligence concept.
After the good data, Nvidia also quietly revealed its updated “hard core” strength. Nvidia CEO Hwang In-hoon pointed out during the earnings call that after Blackwell, Nvidia is developing a new chip at the pace of releasing it once a year.
Previously, Nvidia's chip design frequency was stable once every two years, from Ampere, which was first released in 2020, to the industry darling H100 chip (Hopper series) in 2022, to the highly anticipated Blackwell in 2024. But apparently, two years seems like too long for Nvidia.
Earlier this month, well-known analyst Guo Mingyi revealed that Nvidia's next-generation AI chip architecture, Rubin, will be launched in 2025, and the market can get R100 AI GPUs as early as next year. Now, the credibility of this news seems to have skyrocketed.
Hwang In-hoon said that Nvidia will move at a very fast pace, and new CPUs, new GPUs, new network cards, new switches... a large number of chips are coming soon.
It was replaced quickly, so it sold quickly
Earlier this week, it was revealed that Amazon suspended purchases from Nvidia due to the rapid iteration of AI chips, pending the upcoming release of Blackwell chips. Amazon later denied the suspension of the contract, but admitted that it would switch to purchasing Blackwell chips in the supercomputer project.
This has also raised concerns in the market that the speed with which Nvidia chips are being updated may slow sales of the company's contemporary products. In response, Hwang In-hoon responded that Nvidia's next-generation GPUs are both electrically and mechanically backwards compatible and run the same software, so customers can easily transition from H100 to H200 to B100 in existing data centers.
He believes that as the market transitions to H200 and Blackwell, we expect demand to exceed supply for some time. Everyone is anxious to get their infrastructure online to be profitable as soon as possible, so their orders for Nvidia GPUs aren't stalling.
Hwang In-hoon asked an interesting question: Does the company want to be the first company to release a major milestone in the next phase of the AI competition, or the second company to announce a 0.3% increase in effectiveness just a few days later?
In addition, Nvidia's chief financial officer Colette Kress revealed that automobiles will become the biggest industry in Nvidia data centers this year, and for example, Tesla has purchased 35,000 H100 chips to train its fully autonomous driving system. In addition to automobiles, consumer internet companies such as Meta are also growing strongly.
editor/tolk