Management
Japan subsidises both AI adoption and the retraining around it
The IT introduction subsidy became the Digitalisation and AI Introduction Subsidy in 2026, with a ceiling of ¥45 million and a rate reaching three quarters. Training costs are covered separately at up to 75%.
When AI adoption is blocked on budget, it is sometimes only that nobody checked the available schemes. Japanese support has shifted noticeably toward the adopting side over the last two years.
Why support moved toward adopters
The Basic Plan for Artificial Intelligence, a Cabinet decision of 23 December 2025, framed Japan's problem as implementation rather than research.
That framing redirects money. If research is the problem, support goes to institutes. If implementation is the problem, support goes to the companies adopting.
Accordingly, the IT Introduction Subsidy was renamed the Digitalisation and AI Introduction Subsidy in 2026. Putting AI in the name reflects a change in what is eligible. The ceiling is ¥45 million and the subsidy rate reaches three quarters.
The two schemes do different jobs
Subsidies and grants get conflated, but they are awarded on different conditions.
The Digitalisation and AI Introduction Subsidy covers the cost of tools and systems. It has application windows and a review, and you receive nothing unless selected. The quality of the business plan affects the outcome.
The reskilling course of the Human Resources Development Subsidy covers training costs and wages during training, at up to 75% of the training cost. It pays out where the requirements are met.
The first answers what you install. The second answers who becomes capable.
As covered earlier, only 3.8% of Japanese companies say their DX talent is adequate by quality. Installing tools with nobody able to use them is exactly the dead end that number describes. That is the reason to use both schemes together.
What weakens an application
Scheme details change annually, so what follows is only the part that does not change. Successful business plans share several features.
Current numbers exist. How many hours the target work takes today. A plan with no baseline cannot demonstrate an effect.
The target work is scoped as a process step. Not "improve sales efficiency" but something with a start and an end, like producing the first draft of a quotation.
The division of roles after adoption is written down. Who inputs, who checks, who corrects. Given that the fourth pillar of the Basic Plan asks for a division of roles between people and AI, this section also aligns with policy direction.
The metric is chosen. Handling time, volume, rework rate. State what you will watch after go live.
These four are not paperwork for a subsidy application. An internal approval requires the same content. Write it once and it serves both.
Applications usually fail not because the applicant misunderstood the scheme, but because the plan contains no numbers. Baseline numbers can only be collected before you start.
The order to work in
- Choose one target process and collect the current numbers
- Write the division of roles after adoption
- Read the current application guidelines and shape the plan to the requirements
- If training is needed, check the grant scheme in parallel
Most companies start at step three. Steps one and two are what make step three writable. And step one gets harder to collect the longer you wait.
Scheme names, ceilings, rates and application windows change each fiscal year. Confirm the figures in this article against the current guidelines before applying.
Sources
- MHLW, Human Resources Development Subsidy, reskilling course for business development
- SME Agency, Digitalisation and AI Introduction Subsidy (formerly the IT Introduction Subsidy, renamed 2026, ceiling ¥45 million, subsidy rate up to three quarters)
- Cabinet Office, Basic Plan for Artificial Intelligence (Cabinet decision, 2025-12-23)