IBM, this week, became the latest casualty of AI…so they say. Is that true? To a degree, it is, but not totally, nor primarily for the reasons you’re likely hearing. They’re a very important player in the space, more important in my view than any of the “giants” I blogged about yesterday. What’s happening to IBM is also more important to AI than the positioning of these giants, so we need to look at the story in more detail.
IBM’s CEO attributed the problem to a failure by IBM to recognize the shift of spending to open-architecture servers and storage that IBM didn’t make. Mainframe hardware, which had been strong, softened. The Financial Times attributed the problem to a shift in spending away from software to AI. Wall Street research generally dismissed the shift as a short-term reaction. Let’s start with what enterprises have said.
First, it’s important to note that IBM’s systemic strength has always been with its mainframe products. Mainframes go back to the System/360 of the mid-1960s, and they were expensive and powerful giants of the data center, typically purchased by large enterprises. Not only that, these enterprises regularly wrote their own software to support core missions, and often this software was difficult to redo. Arguably, IBM made it easy to stay with mainframes, and IBM’s decision to largely stay out of the open-model Linux server business surely didn’t facilitate any shift.
The mainframe was the basis for much of the enterprise transaction processing activity, and “transaction processing” is the handling of the routine business paper flow that gradually became computerized. Over time, IBM customers tell me that their work to “modernize” their core applications really focused on shifting the front-end portion of those applications off the mainstream, the GUI, and leaving the core data management and analysis portion on the mainframe.
Transaction processing load advances, generally, at the pace of commercial activity. Industry-wide, GDP growth is a reasonable measure. Shopping, the front-end and GUI processing stuff, has tended to grow faster as companies shifted to more online shopping support. Thus, one could expect mainframe growth to be linked to business results, and business aspirations could reasonably be seen as the driver of front-end activity, the things that have typically been handled by open servers, the Internet, and the cloud. Economic uncertainty, of the sort created by the Iran war and oil crisis, impacts buyer confidence, and as a result impacts the sellers’ IT plans.
AI enters into the picture here, but not as an alternative to mainframe software. No enterprise has ever told me they would be likely to even consider moving transaction processing to AI, even to the self-hosted form. Some (about a third) think AI might impact some of that front-end stuff, but the impact is as much the impact of a shift from capital hosting to expensed cloud/AI as it is a statement that AI could actually do a better job. Right now, enterprises sometimes find that AI pricing is artificially low, making an AI solution to what was originally seen as a cloud/SaaS application more financially attractive.
But what AI has done is drive up the cost of that open server technology, memory and CPUs and so forth. That is generally seen by enterprises (almost 80%) as a short-term impact, and Wall Street agrees, so it’s prudent to defer projects that could require more open-server investment to a later time when components are cheaper. The actual suppliers of this are somewhat insulated by the fact that AI is also hosted with the same sort of gear, but IBM doesn’t sell that stuff.
How about IBM’s position as the smart player in the AI game from enterprises’ perspective? IBM is the champion of self-hosting AI, but their role is mostly in consulting, in software, and in prepping core data repositories for access by AI. The problem with that isn’t just that it’s a limited upside—it was enough of one to be beneficial to IBM’s profits last year and up to this quarter—but also that it isn’t how the vast majority of AI influence is being directed. What do you hear, read, about AI? The stuff that favors the plans of those AI giants I blogged about earlier this week. This resulted in something I’ve noted in past blogs, a growing disconnect between how IT organizations saw AI and how line departments saw it. Even CEO/CFO types tended to think more about the personal productivity applications of AI than the applications of AI to data analytics, which is where the real business cases could be made. Part of the reason was that you could expense cloud AI, and usually adopt it without approvals. Not so with self-hosted AI. As a result, well over three-quarters of enterprises who talked about self-hosting plans early this year now say that their plans have been delayed. Almost 20% say the projects are no longer being actively pursued.
OK, where does this leave us with IBM? I do think that IBM has benefited a bit from the AI hype wave, and now that benefit, and the stock appreciation associated with it, could rightfully be withdrawn. However, I think that much of the impact of that has already been felt, and that the Street is right saying that the second-quarter is a blip that will correct later in the year. A drop of over 20% is short-selling, not a threat to IBM’s value. Less than half that would be appropriate.
But…there is a signal here, maybe even two. IBM’s power in the market is linked to a small field of giant customers who are perhaps a bit of a captive audience. Captive now, but forever? If IBM could have pushed self-hosted AI successfully in the first half of this year, they could have created a market wave of reality that would have served as a counter to all the AI hype, and they could have spread the influence of self-hosting to a broader market. Red Hat was the path this should have taken, but IBM didn’t take it effectively.
Which raises the second signal. Could this be a hint that IBM might acquire an open-server player? Might they have pre-announced a reason for an M&A deal that would/could become public before their earnings date? Tough call, since this might impact the willingness of other open server vendors to endorse a Red Hat strategy with their customers, doubly undesirable now given that Red Hat is trying to win over disgruntled VMware customers. Could it mean some new AI direction for Red Hat, a software direction? Or something new for IBM? Obviously, we’ll have to see what develops, but something could very well be in the wind.
IBM isn’t the only one at risk here, though; it isn’t even the major one. Who should really be sweating? The AI chip players. What AI has proved up to now is that if you give something away, people will take it. AI has also proved that people, voters, don’t like the giant AI data centers. What it’s already starting to prove is that if you invest without a return, you’re going to be punished by Wall Street. The real hope for AI to sustain itself is new, real, business cases, and IBM is proving that we’re not working hard enough to develop them, to distribute both the value and the impact of AI. If that doesn’t change before the end of this year, 2027 could be a very bad year for AI.
