The AI Bubble Will Burst. And That Might Be the Best Thing for AI.
Is the AI bubble about to burst? Explore the gap between AI hype, investment and real business value—and why a market correction could benefit AI.
DWSX Perspectives / Issue 01 — November 2026
The technology can be revolutionary. The investment thesis can still be wrong.
There is an increasingly expensive assumption running through the technology industry: because artificial intelligence can do remarkable things, almost every AI investment must eventually make economic sense.
That is not how technology works. And it is certainly not how markets work.
We are being asked to believe two things at once: that AI will transform virtually every industry, and that the companies spending enormous sums to build and deploy it will all capture enough value to justify those costs. The first proposition may prove broadly correct. The second deserves much more skepticism.
We've seen this movie before
At the turn of the millennium, the internet was genuinely changing the world. Yet investors funded businesses whose valuations depended on extraordinary growth and business models that were sometimes barely tested. The Nasdaq fell roughly 77% from its 2000 peak to October 2002, according to Goldman Sachs' historical account.
The crash destroyed capital and companies. It did not destroy the internet. Online commerce, search, cloud infrastructure and countless digital businesses went on to become indispensable.
This is the useful historical parallel – not a prediction that AI stocks must follow exactly the same trajectory. A technology can be transformative while its market is overheated. Those are not contradictory statements.
Productivity is not profit
McKinsey's August 2026 global AI survey offers a revealing contrast. Eighty percent of respondents say AI improves their individual productivity. Yet only 37% report a positive contribution to enterprise EBIT from AI. Around 6% qualify as AI high performers under McKinsey's definition.
These are self-reported survey measures, not a controlled estimate of AI's return on investment. They do not show that 63% of companies lost money. They do, however, expose a real management problem: helping an employee finish a task faster is not automatically the same as improving the economics of an entire organization.
The saved minutes must translate into capacity, quality, revenue or avoided costs. Otherwise, the gains can remain invisible to the bottom line.
The demo is cheap. The system is not.
An AI prototype can be impressive in an afternoon. Production is another story.
The bill includes integration with existing software, data preparation, security controls, monitoring, evaluation, human review, failure handling and continuous maintenance. Model usage is only one line item. In the same McKinsey survey, about 20% of respondents say AI-related operating costs have constrained use.
None of this makes AI a bad investment. It makes it an investment that must compete with alternatives.
Would a conventional rules engine solve the problem more reliably? Could an API integration remove the bottleneck without a model? Is the task frequent enough to justify the complexity? What happens when the system is wrong? Who owns the failure?
If those questions sound unglamorous, good. They are the questions that separate engineering from theater.
The danger of putting AI everywhere
Some organizations are starting with the solution and searching for a problem. 'We need an AI strategy' is becoming a substitute for 'We need to improve this business outcome.'
That inversion is costly. It encourages teams to introduce probabilistic systems where deterministic ones would suffice, add vendors where integration would do, and optimize for demonstrations rather than sustained operational performance.
There are excellent uses for AI: language-heavy workflows, coding assistance, discovery, classification, and complex tasks where adaptability matters. But there are also processes where predictability, auditability and low operating cost matter more than flexibility.
Mature technology leadership means being willing to say no—even to fashionable technology.
A correction could be healthy. It could also hurt.
Here is the uncomfortable position: a market correction could improve discipline by forcing vendors to prove value, customers to scrutinize costs and investors to distinguish durable businesses from expensive stories.
But it would not be painless. Funding cuts could slow useful research, damage suppliers, eliminate jobs and make customers more cautious. We should not romanticize a crash simply because it might clear away hype.
And we should acknowledge the strongest counterargument: some infrastructure investment needs to precede demand. The early internet required enormous upfront spending too. Today's apparent excess capacity may become tomorrow's essential foundation. Nobody knows with confidence where that balance lies.
What survives after the hype?
The most useful question for a business is not 'Are we using AI?' It is 'Would we still fund this project if the letters AI disappeared from the pitch?'
If the answer depends on a measurable reduction in errors, faster service, better decisions, new revenue or lower total cost of ownership, there may be a real business case.
If the answer depends on impressing a board, copying competitors or making a product sound current, the project is on much weaker ground.
At DWSX, we believe technology should earn its place in the architecture. Sometimes that means AI. Sometimes it means simpler software, better integration, or removing unnecessary complexity.
The AI bubble may or may not burst. But the industry would benefit from one change regardless: less worship of technology, and more accountability for what it actually delivers.
What is your view: are businesses underinvesting in AI—or paying too much to deploy it everywhere?
