Where does Japan stand in AI adoption today?

Over the past week, I have tracked the "current state of AI adoption in Japan" across three layers: regulations, effectiveness, and market evaluation. Today, I will synthesize the whole picture.

The trigger was a sense of discomfort with the figures of "72% for large enterprises and 28% for SMEs" in an article about AI adoption rates for executives. When I started verifying the facts, I landed in an unexpected place.

“Japan at 20%, US also at 20%?” — The discovery of parity
I often see the comparison "Japan at 20%, US at 90%," but these figures are based on different populations. Looking at figures from broad surveys of all companies, Japanese SMEs (20.4%), the US as a whole (17-20%), the EU as a whole (19.95%), and the OECD average (20.2%) are all roughly on par. The 80-90% figures from PwC and McKinsey are figures of a different granularity, targeting large enterprises, managers, and "use in at least one business task."

Tuesday: The government has endorsed the requirements for assetization
AI Business GuidelinesVersion 1.2 warns against excessive reliance on AI and requires those who approve it to have a basis for doing so. While the framework of "assetization of judgment logic" is not directly stated in the guidelines, it was a place that naturally approached that requirement when applied to practical work.

Wednesday: Adoption has caught up, but effectiveness has not
According to the PwC survey, the level of utilization and promotion is almost equal, with Japan at 87% and the US at 90%, but there is a gap in "effects that significantly exceed expectations," with Japan at 9% and the US at 38%. Expanding to the IPA survey, what becomes visible is not a lack of technology, but a lack of personnel who can judge adoption and verify effectiveness.

Thursday: If you confuse correlation with causation, there will be a price to pay
The stock prices of companies that mention AI outperform those that do not on average, but the median is the opposite. The SEC has taken action against companies that exaggerated their AI performance. "Stocks go up if you say AI" and "AI adoption increased profits" are different things.

The same composition can be seen domestically. Looking at companies that specifically mentioned AI in their financial results for the fiscal year ending March 2026 by industry, DataSection (3905), the leader in IT, was +172.8% from the end of March to the end of June, and RareJob (6096), the leader in EdTech, was +31.4%, while the top three companies in the drug discovery field (Daiichi Sankyo, Takeda Pharmaceutical, and Shionogi) all declined (-5.7% to -19.6%). The dividing line is whether AI is spoken of in a way that is directly linked to sales and profits for the quarter, or whether it remains limited to future research efficiency.

When you bundle the three, a common composition emerges.In terms of the volume of actions such as "adopted" or "mentioned," Japan and the US are not that different. The difference lies in the quality of the "verification" that follows.

The phase of chasing utilization rates and mention rates is almost over in Japan as well. What is asked next is how to create these three things: a system to confirm AI judgments and record the basis, personnel who can verify effectiveness, and disclosures that can explain AI effects in numbers. "AI" is not a label; what is being questioned is the depth of implementation.

Details (detailed figures for each survey, citations from government guidelines, and analysis of market evaluation) are written on Zenn.

Visualization of information gaps: The Organization for Small & Medium Enterprises and Regional Innovation, PwC, IPA, and AI Business Guidelines handled in this article are all domestic primary announcements, with 0 translation steps. The US Census Bureau, Eurostat, OECD, NBER papers, and SEC actions are overseas primary materials, with limited introduction in Japan, and 1 translation step (as of 7/18). FactSet is an overseas primary material, but there are introductory articles in Japan, and it has 2 translation steps. The stock price and financial data for the three domestic fields are primary screenings based on each company's financial results summary and timely disclosure, with 0 translation steps. This is not a comprehensive survey of all listed companies and is not intended as an investment recommendation.

Article Jack 4-line structure: AI hybrid evaluation and selection flow
This process is a general-purpose framework centered on "balancing comprehensiveness and cost" that an active PMO with a site for legacy migration applies when deciphering AI application cases regardless of industry.

Translation step index: An indicator that measures whether there is a primary announcement in Japan (= the information gap itself) (0 = domestic primary announcement (no translation required) / 1 = not yet announced in Japan (specified at the time of confirmation) / 2 = via third country/domestic media).

<!-- Keyword 1: Article Jack / Keyword 2: Assetization -->