We’ve heard, often, that AI is a bubble. We’ve heard that it’s not like the dot-com bubble, and we’ve also heard that quantum computing will burst it. What, if any, of this is accurate? Let’s try to find out, taking these points in order.

First, of course AI is in a bubble. In an ad-sponsored media environment, there is simply nothing but dull and bubble available. Online media necessarily strives for clicks, and you don’t get them with stories that say “nothing new or interesting is happening.”. Vendors don’t advertise except where it does them good, which means places where prospects see the ads. As I pointed out in a blog last week, the real question is whether the AI bubble will burst, which leads us to the second point.

Is AI like dot-com? Let’s expand that one a bit to ask “Is AI like past bubbles, including dot-com?” That question is harder to answer, and we can’t approach it without looking at the anatomy of a bubble versus the anatomy of a real revolutionary development.

In a bubble, there is tech coverage and reader interest that feeds a positive view, and that view then feeds coverage. The key factor in bubble creation is that the coverage is linked with knowledge of something rather than to the “something” having achieved actual market potential. In most cases, in fact, the coverage precedes availability, as it did with 5G and is doing with 6G. Vendors tend to jump on this development, even though there’s no immediate sales prospect, if they believe that interest in it will promote willingness to engage among their prospects. However, this can also create “overhang”, where buying decisions are deferred because it’s perceived something better is in the pipeline. As a result, vendors may rush to deliver minimal implementations, or even “wash” current products with claims that are at best vaguely linked to the bubble’s underlying technology.

This development tends to encourage venture capitalists (VCs) to fund startups, in the hope that they can deliver something linked to the bubbling technology concepts and be acquired by vendors interested in getting something to market quickly. These startups feed the news cycle, but they also broaden the base of possible implementations. This is where the critical point of realization occurs. If something comes along from vendors or startups that can actually make a case for purchase at any reasonable scale, the bubble focus will tend to shift to it. This can redeem the bubble, to at least some extent, if the “reasonable scale” of the market exposed is large enough.

This is where quantum computing poses a threat. The Street, and the media in our ad-sponsored world, have an insatiable appetite for hype, and you can only hype a technology so far. AI hype has pushed us to the stage where the goal of AI as we hear it is artificial general intelligence (AGI), which is something no enterprise AI planner is even thinking about. Quantum computing today is like AI a couple years ago; what I’ve always called a “UFO”. The nice thing about UFOs is that they’re not landing in your yard to present themselves for examination, so proponents are free to assign them any characteristic they find interesting, without fear of contradiction. What is more a UFO than quantum technology, given that the whole quantum concept is almost murky by design? There’s no danger of quantum computing actually amounting to more than AI will in the near term, but a real danger in it being more entertaining, and that’s what drives interest and clicks.

That brings us to the difference between AI and past bubbles, including dot-com. With the dot-com bubble, one which I consulted through, the essential problem was created by Wall Street, who loves a bubble as long as it creates new winners and/or losers in the stock market. Worldcom and Enron, massive financial frauds created in the early 2000s, caused the government to pass the Sarbanes-Oxley Act, which improved governance and reporting and made that particular type of fraud much more difficult. Stock analysts played a big role in dot-com, and SOX (as it’s known) made that sort of thing much harder, not to mention illegal. There was no real substance behind the dot-com bubble, and in fact realization of huge valuations would be difficult if all you could really draw on was ad sponsorship.

The 5G and 6G stories are different. There were then, and are now, applications that could realize a significant market for incremental 5G and 6G, a market whose revenues could in fact lift a lot of vendor/operator boats. The problem was that up to 5G/6G, new revenue came from “personal communications” in some form, meaning humans consumed the service directly. The new stuff in 5G/6G was and is related mostly to M2M services, IoT and things derived from it. Think of these as “application communications”, which means you need applications to justify the services. Those applications have technology needs of their own, supplied by equipment and software vendors, not just services offered by telcos. There’s a whole ecosystem to be built, and 5G/6G doesn’t build it, specify it, or even define it. The 5G hype, in addition, was linked to the notion that somehow 5G would deliver significant enhancements to personal communication, enhancements users would be eager to get and pay for. It did not, obviously, and 6G won’t either.

So, what about AI? In a sense, AI is like all of the past bubbles in a way. Like dot-com, AI is a hype wave generated by interest, by coverage, and a lot of what’s said about it is really aimed at raising interest and generating clicks, not about building a market. But unlike AI, there is every reason to believe, based on my own experience and what I hear from enterprises and individuals, that there is real value to AI, even as a personal service. Not only that, there’s an application structure already in place that AI could fit into, and by doing so justify spending on expanding capabilities. A link to what’s missing in every major bubble of the past, and a vulnerability to each of their risks.

The risks? There’s the loss of financial credibility. There’s the ecosystem problem. There’s personal willingness to pay. There’s also loss of interest, and that’s what raises the question of quantum computing’s impact on AI.

The ecosystem problem can be glimpsed in a report from IBM, the company who knows the most about AI and its enterprise role. Glimpsed because the truth is offered peripheral to the main focus of the report. The truth? That enterprises say that one, if not the, problems associated with AI is data. Now, it would be fair to ask how you could possibly make a business case for AI adoption within your company without the data relating to how your business works. Thus, it would be fair to say that the report indicates that a whole lot of enterprises haven’t really done much helpful AI planning, enough to recognize even the critical elements in validating an AI business case. I think that the IBM acquisition of Confluent, who provides a kind of data collection and abstraction layer suitable for framing AI-related workflows, is aimed at packaging an integrated AI agent solution, something enterprises can pick off the shelf, because they’re not yet comfortable with engineering the whole framework themselves.

This same problem, I think, is related to HPE’s earnings miss relating to it. The problem HPE has is that they’re announcing stuff that plays in a realistic world of AI deployment without taking the trouble to promote that vision to prospects. Enterprises are generally positive about HPE’s recent Juniper integration announcement, for example, but they tell me that the sales force isn’t evangelizing, or perhaps even recognizing, the sort of distributed-agent infrastructure that it’s ideal for. That leaves things up to the enterprise, the people IBM’s report suggests are just realizing that data availability across the enterprise is a requirement for making an AI business case. It’s not enough to offer an integrated HPE/AMD/Juniper rack, they say. You need to make enterprise executives understand why that’s not only valuable, but essential.

That’s as important for AI as for HPE, and even perhaps for IBM. IBM is unique among AI proponents in that it’s promoted real AI business cases from the very first, but also unique in that it doesn’t have the breadth of products or customers needed to create an AI movement. Might IBM start doing even more M&A? Possible, and if we don’t want to see a kind of “dot-AI” bubble bursting, maybe we should hope they do.