One of the more interesting aspects of the early-week DeepSeek panic on Wall Street was Meta’s resilience.
While Nvidia was busy logging the single-largest one-day market cap loss in history (a truly “impressive” $590 billion rout, costing Jensen Huang $20 billion on paper), Meta actually notched a sixth straight daily gain.
Indeed, Meta’s up 13% already in 2025, with some of that likely due to Mark Zuckerberg’s efforts to ingratiate himself with the new regime in Washington. If you didn’t listen to Mark’s Joe Rogan interview, you should. Ol’ Zuck’s metamorphosed into a full-on, red pill, right-winger, although I’d suggest a lot of that’s an act in the service of protecting his company and shareholders. Shareholders like himself.
Meta’s outperforming Microsoft by 10ppt this month and as the figure shows, it’s not a recent phenomenon. Zuck’s run circles around Satya Nadella since early 2023, when investors began to reward Meta’s “year of efficiency” plans following a year of — no sense mincing words — catastrophic losses for the stock in 2022.
Consider this: At the seemingly hopeless lows for Meta in October of 2022, the META/MSFT ratio was 0.4. Today, it’s 1.5.
How did Meta dodge the DeepSeek shock? Simple: Investors actually viewed DeepSeek’s ostensible breakthrough (specifically the open source aspect) as an attestation to Llama’s promise. Already, the likes of Gene Munster are calling Llama “the DeepSeek of the West.”
The idea that OpenAI’s suddenly an also-ran because some hedge fund manager in China cobbled together a capable chatbot with Mag7’s couch cushion change is obviously laughable. Or probably laughable, anyway. And it’s not exactly as if Meta’s more frugal than Microsoft when it comes to AI outlays.
Further, Zuckerberg’s Zuckerberg, but Nadella’s not exactly a slouch. The guy’s one of the best CEOs on the planet. Zuckerberg’s a legend — a legend among legends, even — but I doubt seriously the idea that he’s head and shoulders above Nadella when it comes to long-term corporate strategy, and that’s what the stock discrepancy noted above seems to suggest.
Meta’s resilience early this week, JonesTrading’s Mike O’Rourke said, is predicated on the assumption that in the long run, “the commoditization of AI models will be a benefit to the hyperscalers and Meta will be the biggest beneficiary.” “That may certainly prove true, but the ‘long run’ is far away, and Meta just announced a massive capex boost,” O’Rourke went on, cautioning market participants that “chasing new all-time highs in a market meltdown is hardly getting a bargain.”
Both Meta and Microsoft report this week. You can bet DeepSeek will come up on the calls.
META is a developer of AI models, but what makes it money is being a user of AI models. By adopting DeepSeek’s innovations, META may have better models to use and need less infrastructure to develop them. It will still need plenty of infrastructure, but investors won’t complain if capex undershoots plan. “If” – it is not a given that more efficient training means less total inference compute; there’s an argument to the contrary.
MSFT is not a developer of AI models, they bought use of OpenAI’s models in return for a very large funding commitment (anyone kept track of how many tens of billions?) Maybe it now looks like it overpaid. MSFT is already not keeping up with OpenAI’s infrastructure ambitions, hence OpenAI is wriggling away from its formerly monogamous relationship with MSFT and left its patron out of the “StarGate” menage-a-trois. Potentially two black eyes in quick succession? Also, the CoPilot 365 rollout and forcible upgrade is getting panned. MSFT Azure can make money hosting models, but it would be better if MSFT apps also led in AI value-generation.
It’s also worth pointing out that Facebook has blown out earnings in all past Q4s of election years. All that election advertisement spending seems to catch the analysts off base every time.