By now, almost everyone has heard about DeepSeek, the Chinese-made AI that has taught the US a thing or two about building cheaper artificial intelligence. For the chief executive officers of the big US tech companies, the timing is more than a little awkward.
This week will feature the first cluster of tech earnings — from Meta Platforms Inc., Microsoft Corp. and Apple Inc. — since DeepSeek’s emergence as a bona fide threat. Next week brings those of Google parent Alphabet Inc. and Amazon.com Inc. (Nvidia Corp., the company hardest hit, is perhaps grateful that its earnings fall outside the pack, due instead at the end of next month.)
With each company, investors will want to know why a Chinese startup could match the performance of top US-made AI at a mere fraction of the cost. Or, to quote Britain’s Daily Star tabloid, how China was able to make Silicon Valley’s “tech bros” look like “proper thickos.”
Don’t expect such colorful turns of phrase during these earnings calls, sadly. But this is a critical juncture, and the companies are under pressure to use their words to settle the mood. Here’s what I think those words should convey.
First, the DeepSeek moment must be put in a reasonable context. Before this bombshell, investors had already been getting twitchy over the mounting costs of the AI build-out. Yet, for the most part — as evidenced by sky-high valuations — Wall Street has been reassured that tech leaders knew what they were doing, were the best in the world at it and had the most money … so would, therefore, win (duh!).
DeepSeek shattered that perception, but that may be short-lived. The precise workings of its model and the process used to build it are still being dissected. It may be the case that its cheaper method is greatly reliant on piggybacking on the bigger strides made by US companies. Skepticism exists over the stated $6 million cost to build it.
Besides, DeepSeek’s learnings are all public, likely replicable, and, in any case, require the use of Nvidia hardware. It is a highly striking more-with-less feat, spurred mainly by the necessity of working around US export controls. Those export controls remain and become a heavier shackle by the day, leaving Chinese AI makers with increasingly inferior hardware over time unless they can build their own. That’s possible, but it’s undoubtedly a much taller order than matching US efforts on software. Without its own hardware, China will be forced to adapt and maintain outdated hardware, like an old Chevrolet on the streets of Havana.
Second, investors should be reminded that more, cheaper AI is a long-term net good as far as US big tech is concerned, even if this moment has been humbling. DeepSeek indicates that basic AI (as we’ll see it in the future) is becoming commoditized more quickly than first thought. Great! This only increases the opportunity ahead — though such thinking requires Wall Street to think beyond the quarters in front of its nose.
Microsoft CEO Satya Nadella posted late Sunday a short message reminding us of the “Jevons paradox” — a mid-19th-century economic observation that states the more efficient something becomes, the more ways people will find to consume it. The example at the time was coal. Nadella thinks it can now be applied to AI. Microsoft and other US players are still ideally placed to provide the engine rooms and applications for that use. Investors shouldn’t expect any wavering on big goals like the Stargate project confirmed last week — if you believe the AI thesis, all of it (and more) will be needed.
Third, companies should use these earnings seasons to finally start laying out real revenue numbers on their AI efforts so far. Investors have been frustrated by cryptic or ambiguous disclosures, such as AI “contributing 12 percentage points” to Microsoft’s Azure cloud growth in the last quarter. We don’t know enough about the who, what, or why of AI spending.
Now would be a good time for tech companies to pin some numbers down. It can take this moment to remind investors that, unlike DeepSeek, US big tech has existing huge vectors to sell AI, such as Microsoft’s Office 365 or Meta’s advertising platform.
Above all, the companies reporting this week and next need to lay out what hasn’t changed since DeepSeek grabbed the headlines — which is plenty. AI models don’t just need to be built; they need to be run for millions of clients at the same time in a reliable, secure environment fit for corporations, governments and the military.
Those models still need to get better if they’re ever to reach what can be reasonably described as “artificial general intelligence,” something that DeepSeek hasn’t brought us much closer to. Rather, by showing how to do less with more, it potentially directs US tech to do more with more. That’s the narrative that will need to be spun.
A message from Advisor Perspectives and VettaFi: To learn more about this and other topics, check out some of our webcasts.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Dave Lee
