Will AI Be a Bubble? What CIOs Should Plan For Either Way

"Should we pause until the bubble pops?"

A CIO told me about the board meeting where a director slid a newspaper across the table. The headline compared AI to the dot-com bubble. "If this thing is going to crash," the director asked, "why are we signing three-year AI contracts now?"

It is a fair question, and the room didn't have a crisp answer. Like the other scenes in this series, that story is a composite of conversations I have had with technology leaders. In the last year it has become the most common question I hear from boards.

My answer surprises people: it is the wrong question for an enterprise. Whether AI stocks are overvalued is a question for investors. Whether AI capabilities will keep getting better and cheaper is a question for CIOs, and the two have very different answers.

The case for a bubble

The bear case is not fringe. It rests on four observations.

1. The spending is enormous and accelerating. Gartner forecasts worldwide AI spending of about $2.67 trillion in 2026, up 49.5% from 2025, with AI infrastructure alone at about $1.48 trillion ^[1]. Based on their guidance, Amazon, Alphabet, Meta, Microsoft and SpaceX are on track to spend between $803 billion and $828 billion in capital expenditure this calendar year ^[2]. That figure covers all capex, not only AI, though executives at each company point to AI infrastructure as the main driver ^[2].

2. Spending is outrunning cash flow. In a February 2026 estimate, CreditSights put 2026 capex at about 54% of Meta's revenue, 47% of Microsoft's and 46% of Alphabet's ^[3]. One analysis of first-quarter filings found Amazon's trailing twelve-month free cash flow had fallen from $26 billion a year earlier to $1.2 billion ^[4].

3. Some of the money moves in a circle. In September 2025, Nvidia agreed to invest up to $100 billion in OpenAI to fund data centers that OpenAI would fill with Nvidia chips ^[5]. OpenAI also agreed to buy $300 billion of computing power from Oracle over about five years, and AMD issued OpenAI warrants to buy as much as 10% of AMD at a penny a share ^[6]. Critics compare these loops to the vendor financing of the telecom boom ^[6].

4. Enterprise returns are lagging. McKinsey's August 2026 survey found only 37% of respondents attribute any EBIT impact to AI, essentially flat from a year earlier ^[7].

The warnings come from inside the industry and from regulators. In August 2025, OpenAI CEO Sam Altman said he believed investors as a whole were overexcited about AI, while also calling it the most important thing to happen in a very long time ^[8]. In October 2025, the IMF said the AI investment surge echoed the dot-com boom of the late 1990s ^[9], and noted that stock market concentration risk was higher than during the dot-com era ^[10].

The case against

The bull case is just as evidence-based, and it rests on demand that shows up in revenue, not just in valuations.

1. Revenue is real and growing fast. In its fiscal third quarter of 2026, Microsoft said its AI business had passed a $37 billion annual revenue run rate, up 123% year over year ^[11]. Google Cloud grew 63% in the first quarter of 2026, and its backlog jumped to over $460 billion ^[11]. Those are contracted customer commitments, not projections.

2. The financing is sturdier than in 2000. The IMF's chief economist noted that AI-related investment has risen by less than 0.4% of US GDP since 2022, compared with 1.2% between 1995 and 2000, and that the boom is less likely to be a systemic event ^[9]. Much of today's spending is funded from the cash flow of profitable companies rather than from debt-laden startups ^[9][12].

3. The cost curve is falling fast. Epoch AI found that the price of reaching a fixed level of model performance has fallen between 9 and 900 times per year, depending on the task ^[13]. Falling unit costs expand the set of economically viable use cases, much as cheaper bandwidth did in the 2000s.

4. Individuals are already getting value. In McKinsey's 2026 survey, 80% of respondents said AI had improved their individual productivity ^[7]. The gap is converting that into enterprise financial results, which is an execution problem more than a technology problem.

What the last boom teaches

The closest historical parallel is not the dot-com stocks. It is the fiber.

In the late 1990s, telecom carriers laid fiber far ahead of demand. By early 2002, an FDIC analysis cited estimates that only about 2 to 5 percent of US fiber-optic cable was actually lit and carrying traffic ^[15]. More than 60 telecom companies failed between 2001 and 2003 ^[16]. Investors lost fortunes.

But the fiber stayed in the ground. It outlasted the companies that built it, and the glut kept bandwidth cheap for the data centers and internet businesses that came after ^[16].

The lesson for CIOs has two halves:

A financial bust and a technology boom can happen at the same time. The internet did not fail in 2001. Specific business models and balance sheets did.

Buyers usually fare better than builders. When capacity is overbuilt, prices fall for the people consuming it. An AI correction would hurt some vendors and investors, while potentially making compute and models cheaper for enterprises.

What it means for the enterprise

Whether or not valuations correct, four practical risks deserve a line in your AI plan.

1. Vendor survival. The AI vendor landscape is crowded, and a shakeout is likely in any scenario. Gartner estimated in 2025 that only about 130 of the thousands of vendors claiming agentic AI capabilities were real, and predicted that over 40% of agentic AI projects would be canceled by the end of 2027 ^[17]. Some of your vendors will not be here in three years.

2. Price volatility. Unit prices for a given capability are falling quickly ^[13], but vendors under margin pressure may reprice, bundle or restrict tiers. A correction could push prices down for buyers; a capacity crunch could push them up. Build budgets that survive both.

3. Lock-in at the wrong layer. If your prompts, data pipelines and evaluation suites are tied to one provider's proprietary interface, switching becomes expensive exactly when you might need to switch.

4. Value you can't prove. In a downturn, CFOs cut what they cannot measure. With only 37% of organizations reporting any EBIT impact from AI ^[7], programs without hard metrics are the most exposed.

Implementation considerations

Abstract the model layer. Route calls through an internal gateway so you can swap or mix model providers without rewriting applications. Keep a tested fallback model for critical workflows.

Own your data and evaluation. Your knowledge base, retrieval index and test sets are the durable assets. Keep them in your environment and in portable formats, so a model change is a configuration change.

Contract for exit. Negotiate data export in usable formats, transition assistance, notice periods for price and model deprecations, and clear terms if the vendor is acquired.

Tie spend to outcomes. Fund AI by business process with a named owner and a baseline metric, not by tool. Review quarterly and stop what isn't working.

A bubble-proof AI playbook

The goal is a program that wins in both scenarios: if valuations hold, you have compounding capability; if they correct, you have cheaper inputs and fewer, stronger vendors.

Board-ready answers

When the director asks "Is AI a bubble?", three sentences usually land:

1. "There are signs of financial excess, and the IMF and even AI leaders have said so ^[8][9]."

2. "That is a question about vendor valuations; our exposure is to vendor survival and pricing, which we are managing."

3. "Our AI spend is tied to these business outcomes, and here is what each delivered last quarter."

CIO checklist

• Map every AI vendor by criticality, funding source and financial health
• Route model calls through an internal gateway with a tested fallback
• Keep data, retrieval indexes and evaluation sets in your own environment and portable formats
• Add exit, data-export and repricing-notice clauses at each renewal
• Report AI cost per transaction alongside total spend
• Tie every AI initiative to a named owner and a baseline metric
• Stop or restructure any initiative that misses its metric for two consecutive quarters

A founder's perspective

I run an AI company, so you should weigh my view accordingly. Here it is anyway: I think parts of the AI market are priced for perfection, and some companies, possibly including well-funded ones, will not survive the next few years. I also think the enterprises that keep building through the noise will look back on this period the way companies that built on cheap bandwidth looked back on 2002.

The distinction that matters is between the market and the capability. Markets overshoot. Capabilities that get cheaper and better every year tend to keep spreading, whatever happens to the stock prices of the companies that sell them.

Go back to that board meeting. The right answer to "Should we pause until the bubble pops?" is: "We are building a program that doesn't need us to know."