There’s no question that DeepSeek stunned the US tech world with its AI revelations this week. The Chinese startup released DeepSeek R1, a reasoning model as good as ChatGPT o1, saying it trained the AI at a fraction of the cost of what OpenAI has to spend. It’s all thanks to a new approach and several breakthrough innovations in software.
The stock market shed nearly $1 trillion on Monday, as investors feared that AI hardware spending might slow down now that DeepSeek researchers have found ways to match US innovation without having access to the latest chips. Much of those losses were recovered on Tuesday, however.
As the sky is clearing, it sure looks like it’s business as usual for the big players in AI. Anthropic CEO Dario Amodei explained why DeepSeek’s advances will not slow down the race to buy millions of chips to advance AI development and ultimately reach AGI and beyond. He also showed that DeepSeek’s spending on AI is significant despite the software breakthroughs.
Seemingly echoing that chain of thought, Meta CEO Mark Zuckerberg reaffirmed his commitment to spending “hundreds of billions of dollars” on AI hardware in the long term. The comments came during Meta’s quarterly earnings call on Wednesday, during which Zuckerberg said he also wants the upcoming Meta AI Llama 4 model to lead the market rather than follow it.
At the same time, reports say that Meta is already studying DeepSeek and might use the Chinese AI to help power its own advertising products.
Last week, Zuckerberg said Meta will spend $60 billion in 2025 on AI, focusing on data centers. The comments came around the time of Trump’s announcement of the $500 billion Stargate AI infrastructure program.
On Wednesday, the CEO reconfirmed that Meta will invest “very heavily” in AI over the long term. Per TechCrunch, Zuckerberg answered questions about DeepSeek’s impact on Meta AI, saying that AI infrastructure will continue to be a “strategic advantage” for the company.
Zuckerberg said Meta sees DeepSeek as a new competitor that it’s learning from. But it’s “way too early” to tell if chip demand will stop increasing. High-end components remain crucial for inference, or the speed with which the AI processes and responds to prompts.
Also, Meta has to make its AI available to billions of users. “At this point, I would bet that the ability to build out that kind of infrastructure is going to be a major advantage for both the quality of the service and being able to serve the scale that we want to,” Zuckerberg said.
Zuckerberg addressed more than just hardware during the earnings call. He also teased the upcoming Llama 4 model. The CEO wants it to be the world’s most competitive and lead the market. Llama 4 will get agentic capabilities like ChatGPT and Claude, and it will support multimodal input.
I’ll remind you that Meta’s Llama is an open-source project that anyone can use. However, Meta is employing it to power the Meta AI features in its social networks, including Facebook, Messenger, WhatsApp, and Instagram. Meta AI is available in these apps by default, whether you want it or not.
It’s not just Llama 4 and future Meta AI models that Meta is developing. Reports say the company has a “war room” studying its newest rival, DeepSeek. And if there’s one thing Meta excels at, it’s studying rival products and then copying them.
I did say earlier this week that it’s only a matter of time before US AI firms look at DeepSeek’s software breakthroughs aiming to reproduce them. They’re probably doing it already, Meta included. On the same note, Seeking Alpha relays a report from The Information that says Meta is already looking to test DeepSeek for advertising applications.
An unnamed Meta employee claims that Meta is considering the use of DeepSeek because of its “high performance.” The AI might be used to deploy generative AI tools for advertisers. Apparently, some advertisers complained that text- and image-generation tools powered by Llama are not very good right now. They often had to rewrite or edit AI-generated copy.
If the claims are accurate, it’ll be interesting to see how Llama 4 can lead the AI space when current tools aren’t good enough for a key type of Meta AI user: The people whose creations will bring in lots of ad-based revenue to Meta.
Meta did not confirm the rumors but told The Information that it regularly works on understanding all models.
Rumors aside, the biggest takeaway from this week’s Meta earnings report is that the company will spend hundreds of billions on AI hardware. This should inevitably help Meta create better AI experiences in the near future, regardless of whether Llama 4 will lead the way or whether it uses DeepSeek tech in advertising tools.
