Japanese AI software market: a dataset of 2,830 listings

Whether a vendor’s own site can be read in Japanese is a selection criterion for Japanese buyers, and it is the first thing a reader checks before a trial. Of 2,830 products, 484 (17.1%) have a Japanese page on the official site, 1,757 do not, and 589 could not be fetched.

Measured by an automated probe on 2026-09-18, which looks at hreflang, html lang, a language switcher, a .jp domain and kana in the page body. It says nothing about the quality of Japanese input or output inside the product, and nothing about Japanese support or invoicing.

The spread across categories is narrower than the ranking suggests: every category sits between 11.8% and 27.2%. Read it as “most categories are mostly English”, not as a league table.

A Japanese buyer who cannot start without a purchase order will filter on whether the product can be tried alone. A recorded pricing type exists for 1,620 of 2,830 products (57.2%); for 1,210 the free-access condition is not recorded, usually because the vendor publishes no price.

“Not recorded” means we did not find a published condition, not that the product is expensive.

The billing unit decides whether a Japanese team can forecast a monthly bill before it signs. We hold a product-specific billing unit for 1,843 products (65.1%). The remaining 987 carry a placeholder sentence that only says to check the price list, so they are excluded from the table below.

One product can use several units at once, so the rows add up to more than 1,843. 49 products describe a unit that matches none of these six. Seats still lead, but a product whose bill moves with credits, minutes or volume needs a worked example in Japanese before a team can approve it.

This is the axis where our own data is thinnest, and we would rather publish it than hide it. A verification depth is recorded for 413 of 2,830 products (14.6%). Everything else has a checked date and sources, but no recorded statement of how deep the check went.

2,508 of 2,830 listed products (88.6%) come from vendors based outside Japan; 29 are Japanese, and the location of 293 could not be verified. That ratio describes what a Japanese reader searching in Japanese runs into on this site. It is not a measure of market share in Japan, and it partly reflects where we look for products.

These are editorial records built from vendors’ published pages, not from hands-on testing: almost no listing carries a hands-on check, so nothing here says a product works well in Japanese. The Japanese-page figure comes from one automated probe on 2026-09-18 and will drift as vendors change their sites. Checked dates vary across the catalogue (2026-09-05 to 2026-09-26), so a single “as of” date describes when we counted, not when each record was last confirmed. The catalogue is what we chose to list, so it is a sample and not a census of products sold in Japan. Counts change when we add products or correct a record; cite the date with the number.

You may quote or republish the figures on this page, including in commercial work, as long as you name 知能圏 (Chinouken) as the source, link to this page, and keep the date the figure was counted. We would rather be cited accurately than not cited at all.

Suggested form: “知能圏 (Chinouken), Japanese AI software market dataset, as of 2026-09-21, https://www.chinouken.com/en/japan-ai-market”. Spot something wrong? Tell us and we will correct it and say when.