DeepSeek sent Wall Street into a frenzy on Monday, sending shares of Nvidia (NVDA) cratering more than 17% and raising questions about whether Big Tech firms like Microsoft (MSFT) and Meta (META) are overspending on their own AI infrastructure buildouts.
Cut to Microsoft and Meta’s earnings calls Wednesday night and things were hunky dory. The CEOs of both companies, while acknowledging the efficiency improvements DeepSeek introduced to its AI models, said the Chinese-made AI will ultimately benefit their companies as well.
“That type of optimization means AI will be much more ubiquitous,” Microsoft CEO Satya Nadella said. “And so, therefore, for a hyperscaler like us, a PC platform provider like us, this is all good news as far as I’m concerned.”
Meta CEO Mark Zuckerberg, meanwhile, seemed to downplay the threat posed by DeepSeek, saying that the industry is constantly changing and DeepSeek’s announcement is simply a part of that ebb and flow.
“I think there’s a number of novel things that they did that I think we’re still digesting, and there are a number of things that they have … that we will hope to implement in our systems,” Zuckerberg said.
“I kind of expect that every new company that has an advance — that has a launch — is going to have some new advances that the rest of the field learns from. And that’s sort of how the technology industry goes,” he added.
DeepSeek set off alarm bells earlier this week as Wall Street caught wind of the company’s claim that it trained its open-source DeepSeek-V3 AI model, which matches or beats the performance of US models, for just $5 million using less than state-of-the-art chips.
US companies have spent tens of millions of dollars training their own AI models, leading investors to question whether US tech companies are being outpaced by Chinese rivals and if they actually need to spend so much training their AI systems. And if they can train those systems on less expensive chips, the thinking goes, why are they spending so much building massive, high-priced data centers?
However, analysts have questioned DeepSeek’s claims, noting that the company says it didn’t incorporate the “costs associated with prior research and ablation experiments on architectures, algorithms, or data,” into the $5 million figure.
Others have criticized the fact that DeepSeek developed its model using a technique called distillation, which “teaches” an AI model using questions and answers provided by a larger model.
During Meta’s investor call, Zuckerberg said that it’s too early to tell if there’s any need to pull back on investing in AI infrastructure based on DeepSeek’s data. Instead, the CEO said, Meta is better served by investing more in its data centers because the industry is moving to AI models that provide more accurate answers to questions depending on how much computing power they can pull from.
