The AI Bubble 2026: Should Thai Businesses Panic — And the Question That Actually Matters

What is the AI bubble, and why is everyone talking about it now?

Through 2025 and into mid-2026, the largest technology companies poured money into building data centers and buying compute chips at a scale the world had never seen. At the same time, the valuations of some AI companies soared even while they were still losing money. That picture led a number of analysts to warn that we are in a bubble, while others insist this is a structural shift backed by real revenue. The truth is that both pictures can be true at once, and understanding which parts are bubble and which parts are real is what helps a business decide well.

To avoid leaning either way, we will put the evidence from both sides on the table first, then return to the question Thai businesses should actually ask. Every figure in this article is sourced at the end, and we will be explicit about which numbers are reported facts and which are estimates or forecasts still under debate.

The evidence for a "bubble"

The bubble camp points to the sheer size of investment running ahead of revenue, and to funding patterns that circle among a handful of companies.

• Infrastructure spending is surging: In 2026 the major hyperscalers (Google, Amazon, Microsoft, Meta) have combined capital expenditure (capex) in a range of roughly US$600–725 billion (a range compiled by various trackers from company guidance and analyst aggregation), up sharply from about US$410 billion in 2025.
• Circular financing: Some 2026 analyses estimate more than US$800 billion in cross-investment arrangements within the AI ecosystem (an analyst estimate, not audited figures). A large deal such as the Nvidia–OpenAI US$100 billion, 10 GW arrangement was reported to be "on ice," and Jensen Huang said in March 2026 that the US$100 billion figure is "not in the cards."
• The leaders are still losing money: OpenAI is widely reported to be on track to lose around US$14 billion in 2026 (a projection, not audited accounts).
• Warnings from analysts: Ruchir Sharma argues AI checks all four bubble boxes; David Woo sees a possible burst in the second half of 2026; Capital Economics expects a burst within 2026; and a US Treasury draft report warns of dot-com-style spillover. All of these are opinions and forecasts from individual observers — not outcomes that have already happened.
• A depreciation question: Michael Burry argues hyperscalers may understate hardware depreciation (his own model estimates around US$176 billion) and called the situation "clearly Cisco." The changes to equipment useful-life schedules that the companies disclosed themselves are facts; the estimate is his view.

Another bubble signal is the concentration of value in a handful of companies. The following are reported figures as of mid-2026.

The evidence for a "real revolution"

The other side points out that revenue and real-world usage are growing very fast, and that the investment structure differs significantly from the dot-com era.

• Revenue is growing by leaps: Enterprise generative-AI revenue grew from about US$1.7 billion in 2023 to roughly US$37 billion in 2025, and more than 10 AI products now exceed US$1 billion in annual recurring revenue (ARR). (This is an analyst/market compilation, and some of it comes from NVIDIA, an interested source, so treat it with care.)
• Run-rate figures for the leaders: Anthropic reported an annual revenue run-rate of about US$47 billion in May 2026, up from about US$30 billion in April 2026, while OpenAI was around US$25 billion annualized in February 2026. Note that run-rate is not the same as audited annual revenue, and both firms are still loss-making.
• Adoption is real: Surveys find 88% of organizations use AI in at least one function, and 72% use generative AI (up from 33% in 2024). These are self-reported survey figures.
• A structural difference from dot-com: This round is funded largely by highly profitable hyperscalers, not debt-laden telecom companies as in the dot-com era — though debt is now creeping in through data-center special-purpose vehicles (SPVs) and private credit (an analyst opinion).

The reasonable middle: "localized bubble, real revolution"

Recent research tries to pinpoint where the bubble actually sits. An arXiv preprint titled "Boom, Bubble, or Buildout?" (Wang & Chen, May 2026, not yet peer-reviewed) concludes this is "a real technological revolution with localized bubble dynamics" — infrastructure leaders are stronger, while private valuations, application-layer software and speculative data-center projects show the most bubble signs. The paper also cites US private AI investment of US$285.9 billion in 2025.

On the primary-data side, Epoch AI (published 5 June 2026) estimates that AI data-center investment was around 0.8% of US GDP in Q1 2026 (broader computing infrastructure around 1.5%, up from about 0.7% in 2015–2022). That means in level terms it is still a small share of the economy, but in growth terms it is a large contributor. So a pause would hurt momentum, but it would not automatically mean systemic collapse.

A dot-com lesson: the fiber survived, the builders went bust

People like to say "even if the bubble bursts, the technology stays." That is true, but it needs to be understood correctly. In the dot-com era, enormous amounts of fiber-optic cable were laid. When the bubble burst, less than about 5% of that fiber was "lit" (in active use). The glut bankrupted the companies that laid it, and their original equity holders were wiped out — yet the fiber itself remained in the ground and powered the internet for the next decade.

The lesson is that "the technology stays" applies to the asset, not to the shareholders. The data centers and chips being built today may be used for a long time, even if the companies that built them go bust first. This is why a retail investor chasing AI stocks and a business that merely wants to use AI face completely different kinds of risk.

The question Thai businesses should actually ask

Here is the heart of the article. An ordinary Thai business — especially an SME — has no need to bet on whether OpenAI, Anthropic or Google will be the winner, no need to guess whether GPU prices will rise or fall, and no need to predict which quarter a bubble might burst. Those are speculative bets an SME does not need to make. What pays off no matter which way the market goes is durable value: your own data, plus a back-office system that stores it in an orderly way.

Conversely, the real risk for an SME is not the stock price on an exchange — it is the decision to tie yourself to the wrong place.

Where Thai businesses sit on this map

The good news is that Thai businesses are not being left behind. The UOB Business Outlook Study 2026 found that more than 7 in 10 Thai SMEs are implementing AI — above the ASEAN average — and about 8 in 10 have adopted digital solutions, with ongoing concern about the cost of adoption and workforce readiness (see the details in our analysis of Thai SME AI adoption).

On the infrastructure side, Microsoft is investing more than US$1 billion to build an AI data-center region in Thailand (helping keep data resident in-country), and AIS Business, together with Microsoft Thailand, launched an "AI Ready for SMEs" program on 4 June 2026. All of this means the cost of accessing AI for Thai businesses is likely to fall — but it also reinforces that the deciding factor is not "do you have an AI tool yet" but "is your data ready to feed it."

Where ERP fits in (put plainly)

Return to the core of the MIT/NANDA finding: roughly 80% of the work of making AI deliver is the data plumbing behind the scenes, not the model itself. This is where a back-office system like ERP fits in — not because ERP is AI, but because ERP is the layer that makes your data "clean, structured, and yours," which is a condition every AI model needs.

Saeree ERP stores core operational data — accounting, inventory, trading partners and budgets — in a single PostgreSQL database, with an audit trail of who changed which record and when. That data set is a layer that is durable, portable, and yours, not tied to any single AI vendor. When the time comes to build on top with AI, you will have data ready to plug into whatever model wins the market at that moment, whichever it is. That is the hedge against both vendor lock-in and a bursting-bubble scenario, because the real value is held in your data, not in anyone's stock (the same idea we wrote about in ERP and AI in 2026 and in consolidating data in a corporate data warehouse).

Conclusion

The AI bubble of 2026 is not black or white. The bubble-side evidence (capex running ahead of revenue, circular financing, leaders still losing money) and the real-revolution evidence (revenue growing by leaps, real enterprise adoption, a stronger capital structure than dot-com) are both true at the same time. The most accurate description is "a localized bubble, but a real revolution," and the dot-com fiber lesson reminds us that the technology stays while speculative shareholders may not.

For an ordinary Thai business, the question to ask is not "will the bubble burst" but "is our data ready." Clean, orderly data pays off whichever way the market goes, while betting on who wins or when a bubble bursts is a wager an SME does not need to make. Use this period to get your data house in order, and you will be ready to capture value from whichever AI model wins in the future.

"Whether the bubble bursts is not the question for a business that simply wants to use AI well. The real question is whether your data is clean and ready — because that is the value that stays with you no matter which way the market goes."

- The Saeree ERP Team

References

• Epoch AI — AI Data Center Investment as a Share of GDP (5 June 2026)
• CNBC — Big Tech AI Capital Spending 2026 (6 February 2026)
• arXiv — "Boom, Bubble, or Buildout?" (Wang & Chen, May 2026, preprint, not peer-reviewed)
• MIT Project NANDA — "The GenAI Divide: State of AI in Business" (2025 report, non-peer-reviewed, methodology contested)
• Global Finance — AI's Circle Game (circular financing in the AI ecosystem)
• Forbes (James Broughel) — AI Can Change The World And Still Be A Bubble
• IEEE ComSoc — Lessons from the dot-com fiber-optic bubble
• UOB Business Outlook Study 2026 (via ThaiPR) — Thai SME AI adoption

Information last verified on 20 July 2026.