The end of the beginning for AI

 News that Microsoft is slashing sales expectations for its CoPilot application seem to suggest that all isn't quite what it's cracked up to be in the world of AI, or should we perhaps more correctly say, the subset of AI that is LLMs.  ChatGPT, Claude, Gemini, CoPilot and others have taken the world by storm, leading to stories that border on science fiction, where machines do all the work.  Of course, when folks like Elon Musk argue that robots will do all of our work, it's not hard to get caught up in the wave of advocacy that's going on right now.

What we need to realize is that almost every technology is overhyped initially, often fails to deliver on its first and second broad deployments, and eventually finds its product market fit.  Beyond these truths, we have the added disadvantage of having the largest figures in the AI market - OpenAI, Microsoft, Google and others, along with their infrastructure supporters such as Nvidia, making bold claims about the future.  

Let's remind ourselves that in a world where Nvidia is the most valuable company in the world, it has reached that pinnacle through technical artistry and market sophistry.  Nvidia does have powerful GPUs, but also has a pitchman for a CEO, who cannot afford to see his stock price, which is tied to the deployment of data centers and the broad use of AI, decline.  Thus, he, Altman and others are driving a wave of advocacy for AI that is not being realized in the general marketplace, and they are doing a disservice to what AI will eventually become - a much broader, more diversified and generally more useful set of applications than the narrow solutions we are being presented with (and which for the most part aren't delivering).  

So, we have at least two intertwined issues - a still relatively new technology that is finding its product market fit, while the leaders of these technologies and their infrastructure supporters are hyping the technologies and their potential solutions in a way rarely seen previous.  This in turn drives interesting and opaque investment models between the data centers, Nvidia and the larger AI application providers, with a lot of cross investment that smacks of Enron cross-selling internet bandwidth a few decades ago.

People should not be surprised that a general LLM provides general answers to general questions.  When your entire dataset is the broad internet, a reversion to the mean is to be expected.  Smaller AI companies are already taking steps to create more focused and narrow LLMs, aimed at solving specific business problems and leveraging my new favorite Orwellian title - forward-deployed engineers.  A forward deployed engineer is a new title for what we in the past called a consultant, someone who is educated in both how to use the application and can tailor the use to the client's needs, rather than have the client ask inartful questions and create inadequate prompts.

So, we are at the end of the beginning for AI, really LLMs.  It's a mistake to lump all AI into the sturm und drang around LLMs.  There are more use cases and plenty of additional development going on in the broader AI sphere that could be as attractive, if not far more so, in just a few years.  But, with that said, LLMs have plenty to offer and we'll see a lot of learning deployed very quickly in the next 8 to 12 months which will deliver more value than we've seen to date and likely result in a slightly more realistic and possibly more humble pitch from some of the market leaders.

Of course, hyping a new technology isn't new.  One of my favorite examples was for the Segue, yes, the scooter that mall cops and terrified tourists use the most.  When the Segue was first introduced, several notables, including one of the leading venture capitalists of the time, said that they thought the Segue would revolutionalize transportation.  Those claims were clearly debunked only a few months after the launch.  Steve Jobs said that he thought cities would be redesigned around the Segue.  Bezos felt it would solve the last mile problem.  So, no, hyping technology isn't new.

What is new is hyping an entire industry and ecosystem.  Right now, we are seeing hype from the chip makers, the AI software developers, the data center developers and managers and many others, who are hyping an entire infrastructure and ecosystem.  This isn't the Tulip Bulb bubble or the dot com bubble, although the dot com bubble more closely approximates it.  When your measurement of success is "eyeballs" as it was in the dot com bubble, you have a problem.  The same could be said of the current LLM craze.  We've yet to find the real killer app, that drives a significant amount of revenue and solves a significant problem.  Right now, we are hawking a solution without a clearly identified problem.

Are we currently experiencing an AI or LLM bubble?  I think the answer is yes.  We simply haven't seen the killer app from AI in business yet.  There are clearly a lot of experiments going on with AI in the consumer market and in the business world, but few really large and beneficial solutions that have been deployed and demonstrated to drive revenue and profits.  So far, the result has been that AI is interesting, saves some time or improves documents somewhat, and may reduce research efforts.  This means it will likely have an impact on consulting but has yet to identify a broad brush solution.

Does that mean that the AI bubble will crash?  Not necessarily, but we should keep a closer eye on the data center market, where implied demand for AI processing is driving accelerated development of more and more data centers.  These data centers and the implied compute are based on the expectation of larger and larger deployments of AI, which should really only happen if businesses and to some extent consumers see value.  The sites are being built as if the value has already been recognized and calculated, but we are far ahead of the game in that regard.  There could be a correction in late Q1 2026 because no one wants to disrupt the market this close to Christmas 2025, but at some point, some correction will almost have to occur to reset to more realistic expectations.

Comments

Popular posts from this blog

Orwell, the surveillance state and what's next

Executives: narrow your focus for success

Why Taiwan is like Venezuela and the Ukraine