Every few weeks a new headline either declares that artificial intelligence will remake every industry or warns that the whole thing is a bubble about to pop. If you run IT for a company, both can feel true on the same day. Your vendors are shipping AI features faster than you can evaluate them, your board is asking what your AI strategy is, and meanwhile the financial press is comparing today's market to 1999.
This article does not try to predict the market. Nobody can do that reliably, and the people who claim to are usually selling something. Instead, it lays out what the strongest arguments on each side actually say, where the dot-com comparison holds up and where it breaks, and what a sensible IT leader should do when the honest answer is "we do not know yet." It reflects the picture as of November 2026, and parts of the financial story will have moved by the time you read this, so treat the market details as a snapshot and the planning advice as the durable part.
Is the AI boom actually a bubble?
Nobody knows for certain, and serious analysts are split. That split is itself the most useful fact in the whole debate.
On one side, some well-known investors and executives have openly described the current moment as bubble-like. CNBC's roundup of analyst and executive views quotes people who see overinvestment and late-stage behavior, including leaders of AI companies themselves acknowledging that money is being poured in faster than returns are showing up. On the other side, analysts point out that the companies spending the most are highly profitable, which was not true of many dot-com darlings. Even many of the skeptics accept that the underlying technology is real and useful. Their argument is about price and pace, not about whether AI works.
That distinction matters. "Bubble or breakthrough" sounds like an either-or question, but history says it can be both. The internet was a genuine breakthrough and the dot-com market was a genuine bubble, at the same time.
How does the AI boom compare with the dot-com bubble?
The two share a pattern of heavy spending ahead of proven demand, but they differ in who is spending and how they are paying for it.
The comparison is popular because the surface features rhyme: a transformative technology, soaring valuations, a rush of new entrants, and big promises about how fast the world will change. The differences are just as important, and they cut both ways.
| Factor | Dot-com era | Current AI boom |
|---|---|---|
| Who is spending most | Many young, unprofitable companies and telecom carriers | A small group of very large, profitable technology companies |
| Where the money goes | Fiber networks, websites, and customer acquisition | Data centers, specialized chips, power, and model training |
| Funding source | Public equity and debt markets, heavy vendor financing | Mostly cash flow from large companies, plus growing debt and investment deals between suppliers and customers |
| Proven demand at the start | Limited, with revenue far behind expectations | Real usage and revenue, though enterprise returns are uneven |
| Main risk | Companies with no path to profit running out of cash | Overbuilt capacity and hardware that ages faster than expected |
| What survived the bust | Fiber and infrastructure were later put to use; many companies vanished | Unknown, and the central question |
The telecom parallel is the one worth remembering. In the late 1990s, carriers laid far more fiber than demand needed at the time. Many of those companies failed, but the fiber did not disappear. It was eventually used, and cheap bandwidth helped power the next generation of internet businesses. Overbuilding hurt the investors, not necessarily the technology.
Where is the money behind the AI boom coming from?
Most of it comes from the largest technology companies spending their own cash flow on data centers and chips, with a growing share from debt and from deals between suppliers and customers.
Analyst estimates put annual AI-related capital spending by the biggest cloud providers in the hundreds of billions of dollars, and those estimates have kept rising. The exact figures vary by source and change quickly, so check a current, primary source before quoting a number. What matters for planning is the shape of the spending: it is concentrated in a handful of buyers, it is front-loaded, and its payoff depends on customers eventually paying enough for AI services to justify it.
One concern that comes up repeatedly is circular financing. In these arrangements, a supplier invests in or lends to a customer, and the customer then spends that money back with the supplier. It is not illegal or unusual in itself, and it can be perfectly sensible. But it can make demand look larger than it is, which is one of the patterns critics associate with the late dot-com years. For one academic take on how an AI bubble could end, see this analysis from the Yale School of Management.
What does the breakthrough case get right?
It rightly points out that AI tools are being used for real work today, and that the companies funding the buildout are not running on hope alone.
- Genuine usage. Writing, coding assistance, customer support drafting, document search, and data analysis are now routine in many workplaces. People pay for these tools and keep using them, which separates this from technologies that never found their audience.
- Profitable funders. Unlike many dot-com companies, the biggest spenders have large existing businesses that generate cash. A bad year for AI would hurt them without ending them.
- General-purpose potential. Supporters compare AI to electricity or the internet, technologies that took years to reshape industries but ultimately did. If that is the right comparison, short-term disappointment says little about the long-term outcome.
- Falling costs. The cost of using capable models has come down over time, which widens the set of problems where AI makes economic sense.
What does the bubble case get right?
It rightly points out that revenue from AI products has to catch up with spending on AI infrastructure, and that nobody has shown that it will on the timeline investors assume.
- The return gap inside companies. Plenty of organizations have run pilots that impressed people without producing measurable savings or revenue. Reports claiming that most enterprise projects fail to deliver should be read carefully, since methods and definitions vary, but the pattern of "great demo, unclear payoff" is familiar to many IT teams.
- Concentration. A small number of suppliers and customers account for a large share of the activity. If one major buyer slows down, the effects spread quickly.
- Hardware that ages. Chips and data center equipment lose value as newer generations arrive. If the useful life is shorter than the accounting assumes, the financial picture looks worse than reported.
- Narrative-driven pricing. When valuations rely on stories about what AI will be worth in ten years, small changes in expectations can move prices a lot, in both directions.
Have real-world AI adoption experience to share? We publish practical, experience-based articles from IT professionals, including honest accounts of what worked and what did not. See how to write for GrowthifyBiz.
What would a correction actually look like?
Most credible scenarios describe a selective shakeout, not the disappearance of AI: weaker companies fail or get absorbed, spending slows, and the technology keeps getting used.
The dot-com crash is the reference point people reach for. The Nasdaq composite fell by roughly three quarters from its 2000 peak before recovering over the following years, and a large number of companies disappeared. Yet the internet kept growing the entire time. A comparable AI correction could mean falling valuations for infrastructure suppliers, consolidation among model providers, cheaper capacity for everyone else, and a reset of expectations about how fast enterprises will pay for AI.
For a company that uses AI instead of building it, a correction could even help in some ways, through lower prices and stronger vendors. The real risk for buyers is narrower: depending on a vendor or product that does not survive, or that changes terms sharply once cheap promotional pricing ends.
Which signals should IT leaders watch?
Watch the signals that connect AI spending to real, paid demand, because those will tell you more than any headline about valuations.
- Whether AI revenue at major providers is growing faster than their infrastructure spending, or falling behind it.
- Whether large buyers are extending, pausing, or cancelling data center and chip commitments.
- Whether enterprise customers are moving from pilots to paid, widely deployed use, and renewing after the first year.
- Whether pricing for the AI tools you use is rising, falling, or being restructured.
- Whether the vendors you depend on are funded by their own cash flow or by deals with their own customers.
- Whether power and chip supply limits are slowing deployment, which can delay revenue even when demand is real.
Want more on the technology behind the headlines? Browse our AI & Data articles for practical coverage of machine learning, generative AI, and data engineering.
How should IT teams plan when nobody knows the answer?
Plan for both outcomes by funding small, measurable projects, avoiding hard lock-in, and building the data and governance foundations that pay off whether the market booms or busts.
- Tie every project to a number. Before starting an AI pilot, write down the metric it should move, such as hours saved, tickets deflected, or errors reduced, and the level that would justify continuing. Pilots without a target tend to linger.
- Start where the data is already good. AI projects stall most often on messy, scattered, or poorly governed data. Cleaning up data pays off under any scenario, which makes it a safe investment.
- Avoid single-vendor dependence where you can. Keep prompts, evaluation sets, and integrations portable so that switching models or providers is a project, not a rewrite.
- Budget in stages. Release funds in steps tied to results, not as one large commitment, so you can stop cleanly if the economics change.
- Put governance in place early. Decide which data may be sent to external AI services, who approves new tools, and how outputs get reviewed. This is cheaper before widespread use than after. Our cybersecurity coverage touches on the access and data protection side of that work.
- Invest in people. Training staff to use AI tools well and to spot their mistakes holds its value even if specific products come and go.
- Revisit the plan every quarter. The market and the tools are both changing quickly, so a yearly review is too slow.
Frequently asked questions
Is AI a bubble right now?
There is no consensus. Some analysts see late-stage bubble behavior in valuations and spending, while others argue the biggest spenders are profitable and the technology is delivering real value. Both can be partly right, as they were with the internet.
How is the AI boom different from the dot-com bubble?
The biggest difference is who is spending. Today's heaviest investors are large, profitable companies using their own cash flow, while many dot-com companies had little revenue and relied on capital markets. The similarities are heavy spending ahead of proven demand and some circular financing arrangements.
What happens if the AI bubble bursts?
Most scenarios describe a selective correction, with falling valuations, consolidation, and reduced infrastructure spending, but continued use of AI tools. After the dot-com crash, many companies failed while the underlying internet technology kept growing.
Should my company pause its AI investments?
Usually not entirely. A better approach is to keep funding small projects with clear success metrics, stage your spending, and avoid deep dependence on one vendor. That limits your downside if the market turns without leaving you behind if it does not.
What is circular financing in AI?
It describes deals where a supplier invests in or lends to a customer, and the customer spends that money on the supplier's products. It can be legitimate, but critics say it can inflate apparent demand.
Which companies are most exposed to an AI downturn?
Generally those that depend most on continued heavy spending: infrastructure suppliers, highly leveraged data center builders, and startups with large costs and unproven revenue. Companies that simply use AI tools are exposed mainly through vendor risk and price changes.
How can I tell real AI value from hype?
Ask for a measurable result tied to your own workflow, not a demo. If a vendor cannot show repeatable results on your type of data, or cannot say what you would pay at full, non-promotional pricing, treat the claims cautiously.
Will AI still matter if the market crashes?
Very likely, yes. The internet mattered after 2000, and a market correction does not make a useful technology disappear. It tends to reset prices, remove weaker players, and leave stronger products behind.
Bottom line
The AI boom may well be both a breakthrough and a bubble, and the two ideas do not contradict each other. The technology is useful now, the financing looks stretched in places, and the timeline for payback is unproven. For IT leaders the practical answer is neither to rush in nor to stand still. Fund measurable projects, keep your options open, fix your data, and watch the signals that connect spending to paid demand.
Questions or a topic you would like us to cover? Get in touch through our contact page, or read more in the GrowthifyBiz blog.