Nvidia reported its fourth quarter earnings this week, with a stronger-than-expected 78% growth in revenue, despite comparisons to another historically strong quarter in 2023.
That quarter ended in December, so it doesn’t include the impact of DeepSeek, an AI model released earlier this year that can go toe-to-toe with models from OpenAI, despite being developed on a fraction of the budget. But that didn’t stop CEO Jensen Huang from predicting the growth would continue, which has sent its stock price back up and recover from investor unease.
Catch-up: Based in China, DeepSeek not only used fewer chips to create its R1 model, it had to use the Nvidia H800 — a less-capable version of the H100 chips most other developers use, but doesn’t contravene U.S. trade policies that have limited AI chip supplies in China. There has been suspicion that DeepSeek may have used backchannels to get its hands on other chips, but it almost certainly didn’t have the massive stockpiles other companies have been gathering.
The rosy view: Huang expects that so-called “reasoning” AI models — which are capable of more complicated, multi-step tasks and include DeepSeek’s R1 — will still need a lot of computing power, which will drive orders for its next-generation Blackwell GPUs from those looking to train and run them. Even if those models are leaner, like DeepSeek’s, it will still fuel more adoption and development, meaning more chips will still get sold.
Okay but: It’s not neccessarily that DeepSeek itself is going to take the knees out from every AI developer and their hardware providers. It’s that there a company out there credibly suggesting that tossing more money and chips at AI isn’t the only way to make it work better, which has been the industry’s central thesis for growth and created what most in the industry agree is an AI bubble.
That means more AI competitors — especially those who can’t afford rooms full of servers — might try and succeed at that other path, helped by the fact that DeepSeek open-sourced its methods.
Bottom line: DeepSeek is probably not going to cause Nvidia’s collapse this quarter — or even the next one, or the one after that. But most of the billions being pumped into AI are going straight to Nvidia, and downplaying the impact a processor-light model could be another example of a “don’t worry about the bubble” statements that have come before other economic bubbles popping through history. But hey, with how the stock price has been performing, investors seem fine with squeezing some more gains out of the bubble, at least for a little while longer.
Oh, and: The other potential snag that wasn’t included in Nvidia’s Q4 was some of its RTX-series GPUs — used primarily in PC gaming — having a hardware issue that caused them to underperform. This came after production on Blackwell chips was delayed due to design flaws and issues with overheating — and even though Nvidia says the issues have been fixed, some believe the chips could fry again, since not every customer will use them in ideal conditions.