The attitudes of Wall Street and the financial press on the AI giants has all the stability of teen-age romance, and that’s been proven by the swings in stock prices we’ve seen. The giants are spending way too much on AI one day, and are making money on it the next. Or so it seems. Let’s try to assemble some data points and get at what’s really happening.
One Wall Street report seems to me to capture the company/Street thinking. It talks about the “mega-cloud” and says that capex monetization (meaning ROI) is stepping up. What’s really interesting here, particularly to me, is that the chart that’s presented talks about cloud growth not AI growth. The report also favors Amazon and Microsoft as the leaders in mega-cloud, with Google trailing, despite the fact that there’s been pretty continuous and convincing evidence that Google has done better with AI ROI than the others.
The reason this is particularly interesting to me is that I noted in the past that the cloud giants were seeing AI as another way of easing everyone toward the “everything moves to the cloud” position. That was a popular view at one time, but it was overtaken by reality, repatriation, and slowing cloud revenues. Remember, the Street loves a bubble, and so it’s been concerned that AI was running its course, bubble-wise, and that quantum computing was developing too slowly (and was too geeky) to take up the bubble banner. So why not attribute a reversal of slowing cloud growth trends to AI, thus boosting the latter bubble? In the chart I cited, all three of the mega-cloud players showed better cloud growth in the second quarter than in the first. Since we don’t get firm AI revenue numbers from the Big Three, who’s to know?
We are, perhaps, or at least maybe we can.
Enterprises have been telling me about three trends they’ve seen this year, and both could be important in decoding whether AI investments are really monetizing.
The first trend is the acceleration of line organization interest in the personal-productivity form of AI. For all the years I’ve been in IT, there’s been tension between the line departments and the IT organization. In one of my prior employers, this tension resulted in a periodic shift between having all IT independent under a CIO who reported to the CEO, and having an IT CIO and organization for each line department, reporting to its VP. In all organizations, it tends to manifest itself by line attempts to break projects free of IT. Think low/no-code, citizen developers, forth-generation languages. Even SaaS is an example. This is easiest to do when the technology you need can be acquired as a service, meaning expensed.
Publicity, particularly publicity related to bubbles, thrives most where populist interest can be leveraged. Thus, stories about how wonderful AI is and how much it can do for you are targeted at average people, not IT professionals. The latter make up only a bit over 7% of the US workforce, and less than half of knowledge workers. This AI publicity focus, then, empowers the average at the expense of supporting the needs of the specialists, which actually makes it harder to develop an in-house AI capability. That’s our second trend.
The third trend is the increased competition for sales in a troubled global economy. Sellers know that financial concerns are impacting buyer confidence, and in many cases actual buyer budgets. What do you do to respond? Cutting prices cuts your revenue unless you can make up the discount with volume, which is hard to do in a challenged economy. So you look prettier. You try to make your stuff, product or service, look more attractive, make it easier to decide on, make it more accessible to consider. All of these things tend to focus seller attention on the front-end piece of applications, which happen to be what runs in the cloud in most cases.
This focus on prettiness is also smart if you consider that one reported impact of economic uncertainty is an increase in seller reliance on affluent buyers. These buyers are more likely to be tech-literate, more likely to comparison shop, and more susceptible to clever messaging delivered online.
In the early days of AI, meaning largely last year, enterprises were focused on in-house AI deployments just as they currently focus their core applications on self-hosting the core elements. There is still progress being made here despite the second of our trends (Palantir’s quarter shows that), but the first trend has driven what Palantir calls “token slop” and the third has emphasized the cloud-resident piece of applications. As a result, nearly all the enterprises who talked about self-hosted AI project plans in 2025 say they are delayed now. About one in 8 says they’ll not likely make much progress this year at all.
Many of those same enterprises do say that they’ve increased cloud spending, and nearly all of them say that cloud AI usage in their company has grown significantly, though most of my contacts are on the IT side and don’t have a direct way of measuring what line departments are expensing. Thus, there is anecdotal evidence that the mega-cloud boost may be more than a couple of quarters long, which would be good news for the cloud giants.
The challenge is the other things there’s evidence for. Every single enterprise who told me that AI usage was growing also said that it was out of control, that budgets were likely to be blown, and that constraints were likely to be applied in 2027. The IT organizations also say that there is little chance that the current AI missions, the “token slop”, would simply be brought in-house down the line to control costs. They think, as one put it, “we hit pause on rational AI planning this year.”
That’s not good news, people. Absent new business cases to drive new spending, the mega-cloud story is really a story of transference of hype. Everything didn’t move to the cloud, so now everything will move to AI, which happens to be in the cloud. The cost and data governance problems of the cloud, the problems that led to “repatriation” stories, have not been resolved.
Unfortunately, the AI hype and the “token slop” have covered up attempts to bring rationality, business cases, to AI. We may be overlooking a fundamental point, too. AI doesn’t have to “move” to the cloud, it has to move to the data center, and it’s far from clear that we know how to do that, or even want to know.
Why do students love AI? One told me it was simple, “It lets me work less, and someone else is paying for it.” That’s the same view that enterprises think workers have. Maybe AI is like a teen romance in more ways than one.
