Nvidia suffered a rocky — albeit not exactly disastrous — post-earnings trade on Thursday, when the shares were back-footed along with the rest of the market as US macro concerns and trade jitters continued to vex investors.
The company’s report was good, and so was the current-quarter guide, but as I put it while recapping Nvidia’s results, “when the weight of the world’s on your shoulders, ‘good’ sometimes isn’t ‘good enough.'”
That said, there was scant evidence to suggest an abrupt slowdown’s in the offing for AI spend. If anything, it’s the opposite. Maybe DeepSeek created OpenAI with $42, a rubber band and two paperclips, maybe it didn’t, but for now anyway, it doesn’t look as though Nvidia’s staring at any sort of existential crisis.
As the figures above, from Goldman, remind you, Q4 results from the hyper-scalers all tipped massive outlays going forward, and Nvidia’s the biggest beneficiary of that spending.
Meanwhile, strategists continue to suggest the investment case around AI’s transitioning towards “adopters,” which is to say investors are likely to disproportionately reward companies which put AI to use going forward, and even more so companies which quantify the benefit.
Indeed, the figures below, from Morgan Stanley’s Mike Wilson, suggest companies which talked up efficiency through AI on their calls were rewarded to the tune of 80bps of outperformance over the next five days. The median company which, to quote Wilson, gave “concrete, numerical examples of efficiency gains tied to AI” enjoyed nearly 3ppt of outperformance post-earnings.
Not surprisingly, the same dynamic’s visible in analysts’ profitability outlook for such companies, illustrated by the bottom row of charts. Margin revisions for the same group of corporates were meaningfully stronger post-earnings versus the index, with those providing specific examples of AI efficiency enhancements enjoying a 50bps boost in forward net margins post-earnings versus 20bps of incremental projected margin contraction for the S&P.
One way to look at that is to argue that any way you cut it, it’s safer to bet on “adopters” at this point than overvalued hyper-scalers, Nvidia and “picks and shovels” (so to speak). Even if you think massive hyper-scaler capex is justified and that “Remember Nvidia?” won’t be a punchline one day, there are presumably better bargains to be had across the universe of AI adopters and beneficiaries, which would also outperform in a scenario where it turns out DeepSeek was right to suggest there isn’t a one-to-one relationship (or anything like it) between dollars spent and AI performance.
Your last sentence sums it up nicely.
Each new greatest LLM is bring less & less improvement for a greater & greater cost. Now it is time to see commercial end users (beyond coders) benefiting enough to pay to use it. The Salesforce experience with their agents is not very encouraging but it’s early yet.
I am taking UNDER on these Capex numbers – and am as comfortable as can be.
I’m not a mega-fan of polls and surveys, but if this one is even partly right it suggests that AI use in most real world businesses will not be an overnight sensation. And a reminder that coding and call centers are not a major portion of the overall economy.
https://www.zdnet.com/article/most-us-workers-dont-use-ai-at-work-yet-this-study-suggests-a-reason-why/?utm_source=Iterable&utm_medium=email&utm_campaign=campaign_12759717
Curious if anyone here uses AI at work in a significant way? And for what, specifically?
I use AI at work constantly. I am a coder, but I don’t use it just for coding. You can ask AI to get a sense of what ppl are working on based on ticketing systems, you can have it generate a podcast to tell you who is doing what.
When I have a new technical project, I ask AI the best way to get it done.
It’s absolutely incredible.
There is a very useful discussion of the point you make in your last paragraph found in H.Igor Ansoff’s insightful book: Implanting Strategic Management, Prentice-Hall International,1984, Section 2.2.2, pp 40-44. The top graph on p 41 illustrates the demand cycle curve for what may be called “turbulent technologies.” The envelope curve refers to the overall demand for products which are long-lasting but over time have been produced with a number of different technologies. Radio, for example began in the late 1800s. The first radio transmitters and receivers where based on simple crystal technology. Now a radio system can be microscopically small contained on a chip in an Apple watch.
The lower graph represents the demand curve for a “fertile technology.” These types of technologies can be used in myriad products. Total demand will be driven by the number of applications of the tech, eg. chips, transistors, etc. The most desirable investments will typically be found on the upper curves. AI is very probably a fertile technology so the best investments will probably be made in those who develop and produce the technology. There will be thousands of uses for AI and few of them will produce more revenue than the tech used to facilitate each individual applications. On the other hand, the tech will evolve over time and get cheaper as production increases, so perhaps complex value-added applications will produce more value than the tech itself. You can see how difficult it is to find the best way to go.