AI bubble guide - EQOS Asia

EU AI Act 2027 compliance costs and the AI bubble

Summarise this page with AI

The essentials in 30 seconds

  • As of October 2026, EU AI Act compliance is due on 2 December 2027 and falls on AI suppliers and on buyers.
  • Five cost lines cover most of the work: documentation, testing, oversight, monitoring, and legal review.
  • In a worked example, 4 systems at 100 each take 8% of an AI revenue of 5,000.
  • The EQOS Asia AI study scores explainability of deployed models at 0.25 (April 2026).
  • Singapore and Hong Kong firms with European customers can carry the cost.

The EU AI Act turns a general worry about AI rules into a dated bill. This page sits under AI regulation as a trigger for the AI bubble and covers the compliance side. It lists the cost lines, shows a worked budget with round numbers, explains who pays, and links the deadline to the fragile state of the AI market. It is general information as of October 2026 and not legal advice.

What does the EU AI Act deadline of 2027 mean for the AI bubble?

The EQOS Asia AI study lists AI Act compliance on 2 December 2027 among the facts that shape the market (EQOS Asia AI study, April 2026). For the AI bubble, the date matters because prices assume future profits, and compliance subtracts from them. A cost that is known and dated can be priced. A cost that is underestimated surprises the market when the bills arrive.

The deadline also compresses work. Companies that start late compete for the same lawyers, auditors, and testers in the same months. Early starters spread the cost and learn what suppliers can document. For a decision maker, the useful step is to treat the date as a project milestone with a budget and an owner.

Which cost lines does EU AI Act compliance add for an AI company?

Five lines cover most of the work. Documentation records how a system was built and what data it used. Testing checks accuracy and bias before release. Human oversight designs who can review and stop a system. Monitoring tracks behaviour after launch. Legal review confirms which rules apply to which system. The first four recur each year, and the fifth is mostly a one-time cost.

The lines grow with the number of systems in scope and with the risk of their use. A drafting assistant needs little. A system that scores people for jobs or credit needs much more. The first task is therefore a count: how many AI systems do you run, and what decision does each one support?

How much of revenue can EU AI Act compliance take?

Take an illustrative AI company with 5,000 of AI revenue and 4 systems in scope. Each system needs 60 for documentation and testing, plus 40 a year for oversight and monitoring. The first-year cost is 4 times 100, or 400. That equals 8% of AI revenue, and for a company with a 16% margin it removes half of the profit.

The figure is not a forecast. It shows how the pieces add up, and it shows why a margin that looks comfortable can shrink fast. If the company passes 25% of the cost to customers, the burden falls to 6% of revenue. Replace the round numbers with quotes from your own suppliers and advisers before using the method.

Illustrative EU AI Act budget: 4 systems, 100 per system, 400 total, 8 percent of revenue
An illustrative first-year compliance budget on 5,000 of AI revenue.

Who pays the EU AI Act compliance costs, the supplier or the buyer?

Both can pay. The supplier carries the cost of building a compliant system and recovers part of it through prices. The buyer carries the cost of using the system in a compliant way, such as oversight, staff training, and records. Contracts decide the split, so the wording on responsibility deserves a careful read.

A supplier that cannot recover its costs may leave a market or stop a product. A buyer then loses a tool it depends on. Ask each supplier three questions in writing: which of your systems fall under the Act, what is your plan to comply by 2 December 2027, and which costs will you pass on?

Which companies outside Europe face EU AI Act compliance costs?

A company outside Europe can be in scope when its AI system reaches European users or its output is used in Europe. A Singapore software firm with customers in Europe, or a Hong Kong group that scores European applicants, can carry compliance costs although its servers and staff sit in Asia.

A simple map helps. List the markets of each AI product, mark the ones in Europe, and note who in the chain is responsible. The result is a short table that can go to a lawyer. This route is cheaper than asking a lawyer to review everything, and it shows where the budget should go first.

How does the EU AI Act deadline interact with a fragile AI market?

The EQOS Asia AI study scores the enterprise AI market at 0.337 and records explainability of deployed models at 0.25, with transparency of automated decisions at 0.30 (EQOS Asia AI study, April 2026). Documentation and oversight depend on explaining what a model does, so low explainability means a harder and costlier compliance job.

That link is why this branch matters for the AI bubble. Companies that deploy models they cannot explain face the largest bill, and they are often the ones whose projects fail in production. The cost lands where resilience is lowest. It is a pressure on top of rates, earnings, and funding, and it can add to the others.

How can you budget for EU AI Act compliance before 2027?

Plan in four steps. Inventory your AI systems and their uses. Classify each by the decision it supports. Price the five cost lines for each class using supplier quotes. Add the totals to your 2027 budget with a buffer. Review the numbers each quarter, because rules and guidance can change, and your system list will change too.

The EQOS Asia method can turn the budget into a score of exposure, and the Λ simulator lets you test a cost shock on your plan. A session with the EQOS Asia team applies the same steps to a live decision. This page is general information as of October 2026, not investment or legal advice.

Four steps to budget for EU AI Act compliance
A four-step compliance budget.

Frequently asked questions

When does the EU AI Act compliance deadline fall?

The EQOS Asia AI study lists AI Act compliance on 2 December 2027. Plan for it as a project milestone with a budget and an owner, because companies that start late compete for the same advisers and testers.

What are the main cost lines of compliance?

Five lines cover most of the work: documentation, testing, human oversight, monitoring, and legal review. The first four recur each year, and the number of systems in scope and the risk of their use drive the total.

Can a company outside Europe be affected?

Yes, when its AI system reaches European users or its output is used in Europe. A Singapore or Hong Kong company with European customers can carry compliance costs although its servers sit in Asia. Map your markets to find out.

Why does the Act matter for the AI bubble?

It adds a dated cost to companies whose profits prices already assume. The EQOS Asia AI study scores explainability at 0.25, so companies deploying models they cannot explain face the largest bills, often where resilience is lowest.

Test your knowledge

1. What is the compliance date listed in the EQOS Asia AI study?

  • 2 December 2027
  • 1 June 2025
  • 31 December 2030
Show the answer

2 December 2027.

2. In the worked example, what share of AI revenue does compliance take?

  • 2%
  • 8%
  • 25%
Show the answer

8%.

3. Which cost line is mostly a one-time cost?

  • Monitoring
  • Human oversight
  • Legal review
Show the answer

Legal review.

4. Who can be in scope outside Europe?

  • Nobody
  • Companies whose AI reaches European users
  • Only European banks
Show the answer

Companies whose AI reaches European users.

5. What score does the study give explainability of deployed models?

  • 0.25
  • 0.50
  • 0.95
Show the answer

0.25.

About the author

Portrait of Stéphane Jambu

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

Sources: EQOS Asia AI study (eqos-asia.com/case-studies/ia-2026.html), EQOS Asia method page (eqos-asia.com/method.html), consulted on 22 October 2026. Page updated on 22 October 2026.