Nvidia has leapt back into the AI game as panic over DeepSeek ebbs

During Tuesday’s trading this week, Nvidia’s stock price rose above $143 a share —higher than it had been just before the DeepSeek selloff. The revival was made complete by the arrival of Grok-3, the latest chatbot from Elon Musk’s xAI. It was presented on Monday night during an X livestream and came just days after Musk put in a trollish bid to buy rival OpenAI for $97.4 billion, which was summarily rejected by OpenAI’s CEO Sam Altman and the company’s board.

Also Read: Musk versus Altman: Get set for Act II of a Shakespearean drama

The Musk bid’s insincerity is arguably exposed by Grok-3’s performance, which appears to put it neatly ahead of rivals, though likely not for long (as such is the way it goes). Musk doesn’t need OpenAI to compete in the field of AI, evidently, but, by the look of things, he would take great satisfaction in making Altman’s life more difficult in any way he can.

Notably, the not-so-secret ingredient for Grok-3’s progress— which, as Ben Thompson notes in Stratechery, comes just 19 months after a standing start—was xAI’s enormous stockpile of Nvidia chips. 

The company’s Memphis data centre opened with 100,000 GPUs and could expand to as many as 1 million. Nvidia CEO Jensen Huang described the pace of xAI’s building as “superhuman” and suggested Musk was the only man capable of such a feat. 

Some might wonder what could be achieved if Musk stuck to what he was good at rather than meddling in things he doesn’t understand. Last week, Bloomberg reported that xAI is nearing a $5 billion deal with Dell to build servers to house even more Nvidia hardware.

Also Read: Welcome Elon Musk, the shadow president-elect of the United States

Nvidia investors should be encouraged by all this, as it further demonstrates that the route to leading-edge AI still runs through Nvidia chips—and seems set to remain that way for the foreseeable future. What Chinese engineers pulled off with DeepSeek was a striking achievement that has accelerated US efforts to build more efficient but smaller AI models.

But what DeepSeek wasn’t, it turns out, was a threat to the status quo of AI companies spending as much money on Nvidia hardware as possible.

Competitive pressures will come from other chipmakers, sure. Amazon.com, for one, is becoming increasingly bullish about its in-house Trainium and Inferentia chips, which it says are a cheaper alternative to Nvidia for training and running AI models. But Huang argues, and many agree with him, that Nvidia isn’t merely selling AI chips but an all-encompassing AI platform—its CUDA software platform is highly popular among AI developers and compatible only with its own microchips, though some question whether that will matter in the long run.

At the beginning of the year, I wrote that one certainty in 2025 would be repeated panics over Nvidia because investors are looking for any signal, however slight, that the company’s astronomical rise is built on a flawed assumption of AI’s impact. We didn’t have to wait long for these jitters to manifest thanks to DeepSeek. It won’t be the last time investors overreact.

Also Read: DeepSeek’s RI may be the first of many AI super-apps to come

In fact, it could be about to happen again. Nvidia is scheduled to report its earnings next Wednesday. Analysts at Bank of America warn that the company’s transition to selling its newer Blackwell chips, which have already hit some snags, could affect Nvidia’s forecast.

So, too, could the looming question of increased restrictions on exports to China, provoked in part by what the DeepSeek team achieved with sneaky workarounds. These are headwinds, for sure, but neither will likely throw Nvidia’s performance massively off course.

There are, of course, other questions, too, about Grok-3 and the rest. Crucial queries like ‘What do I actually do with this?’ and also ‘Will it still get a lot wrong?’ The answer to the first is we still don’t truly know. The answer to the second is it absolutely will. Beating other AI tools against a benchmark is neat, but it doesn’t necessarily mean much progress has been made in the field of actual usefulness. It’s still reasonable to ask: ‘So what?’

But if we limit ourselves, today at least, to the far more straightforward question of whether the January panic over Nvidia was justified, we can say the answer is ‘no.’

I’ll repeat what I have said in the past: Even if AI is overhyped, a great deal more Nvidia chips will be bought in the process of finding that out. ©Bloomberg