AI bubble or real AI boom: how to tell them apart

What separates an AI bubble from a real AI boom in four criteria?

A boom and a bubble differ in what pays for growth. In a boom, customers pay for results, revenue rises with spending, and prices follow earnings. In a bubble, new investors pay for hope, revenue lags spending, and prices follow other buyers. Four criteria turn this into a test: funding, revenue, productivity, and durability. The table shows each criterion for both cases.

The four criteria work because each one is measurable with public data and a date. They also suit any reader, from a Singapore family office to a regional manufacturer. A criterion you can measure belongs in the decision, and one you can only argue about belongs in a note.

Who pays for AI spending, and what does that say about a boom?

Follow the money to its source. If customers repay the spending through subscriptions, usage fees, and savings, the system funds itself, and the growth is a boom. If the spending relies on fresh share sales and loans against high valuations, the system funds itself only while prices rise, and the growth leans toward a bubble.

The EQOS Asia AI study gives the scale: about $200B a year goes into AI research and deployment. Against that, MIT 2025 finds that 95% of deployed AI projects fail in production. A project that fails in production returns little to its buyer, so check the customer-funded share of spending against current revenue data.

How fast must AI revenue grow for the boom reading to hold?

Revenue is the second test. A boom needs revenue that rises with spending, quarter after quarter. Look at revenue per customer, renewal rates, and the share of customers who move from pilot to production. Each figure tells you whether buyers keep paying after the first year. Ask for these three numbers from any vendor before you accept a growth story.

The study names why pilots stall: insufficient or biased data at 42%, no alignment with the strategy at 31%, and team resistance at 24%. These causes sit inside the buyer’s organisation, so the boom reading depends on whether buyers fix them. The study’s four corrections lift the score from 0.337 to 0.401 and cut the predicted failure rate from 94% to 82%.

What do productivity gains say about an AI bubble or a boom?

Productivity is the third test, and it favours the boom reading. The EQOS Asia AI study projects productivity gains of +40% and 300 million jobs that change. Gains of that size give buyers a reason to pay. The detailed data, including where gains show up and where they stay hidden, sits on the page about AI productivity gains.

A projection becomes a measurement only after delivery, so apply a discount. The study also notes that fewer than 15% of companies have a retraining plan for affected employees, and McKinsey estimates that 30% of hours worked could be automated by 2030. Gains reach the income statement after teams adopt the tools, which is why this test needs repeated checks.

Why can a real technology still produce an AI bubble?

The fourth test is durability, and history gives a lesson here. A technology can be real and useful while the prices of the companies around it run far ahead of earnings. Railways and the internet each changed the economy, and each saw periods when investors paid for the future at one price and received it at another. The page on how a real technology still creates a bubble covers both cases.

The lesson for AI is that usefulness and valuation are separate questions. A boom reading says AI will matter. A bubble reading says today’s prices overshoot what it will earn. Both can hold together, so durability asks a precise question: which revenue would survive if prices fell by half, and which spending would stop?

Can the AI market be a boom and a bubble at the same time?

Yes, and this is the most useful reading as of October 2026. Different parts of the AI chain can sit in different states. A product with paying customers and measured gains behaves like a boom. A valuation that rose ×4 in 18 months while 95% of projects fail in production behaves like a bubble. The study’s verdict, bifurcation, describes this split: the system can settle on a soft landing or break.

Mixed readings suit staged decisions. Commit capital to the parts that pass the four tests, set smaller positions where tests are red, and write down the review date. The score φ(S) = 0.337 places the market in the fragile zone, so the mix can shift quickly when a shock arrives.

How can you score this quarter’s AI bubble or boom result?

Give each criterion one line: green when the evidence supports a boom, red when it supports a bubble, amber when mixed. Write the source and the date beside each line. Three or more red lines in one quarter raise the probability of a bubble, and your commitments can move with it. The record turns an opinion into a series you can defend.

The Λ simulator gives a first estimate for your own exposure in 10 minutes, free and anonymous. If the decision involves AI spending in Singapore, ASEAN, or Hong Kong, the EQOS Asia 30-minute pre-diagnostic returns a score and a ranked set of options (EQOS Asia AI study, April 2026).

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile