Japanese Game Developers Generative AI: 85.8% Now Use It

Japanese game developers generative AI use reached 85.8% in 2026, according to a preview of the CESA Video Game Industry Report published on September 17. Last year the same survey put the figure at 51%. In other words, the industry added roughly 34 points in twelve months. CESA runs the Tokyo Game Show, and it chose the show floor to release the number.

CESA is the Computer Entertainment Supplier’s Association, the trade body behind Tokyo Game Show 2026. On the first business day of the show it released a preview edition of its annual industry report, and the generative AI section is the part everyone quoted.

The sample is not small. Developers returned 1,349 valid responses, and 48 of the association’s 220 member companies answered a separate corporate questionnaire. Because CESA members include Capcom, Sega, Konami and Level-5, the survey reaches the studios that actually ship the games.

One number carries the story: 85.8% of respondents use generative AI in their work. In the 2025 edition that figure sat at 51%. As a result, what looked like an experiment a year ago now looks like standard equipment.

Adoption headlines can hide how shallow the usage is. Here the opposite is true. Of everyone surveyed, 63.0% said they reach for generative AI every working day, while 22.8% use it occasionally.

That leaves a small tail. Another 8.6% are still trialling or verifying the tools, 4.8% have never touched them, and 0.7% used them once and stopped. Consequently the holdouts are now a clear minority in Japanese development.

How Often Studios Reach for It

The full split from the 2026 preview edition looks like this.

The 2026 preview did not break down the specific tasks. However, the 2025 report did, and it gives the clearest picture we have of where the work lands.

Generating visual assets and images came first. Story and text generation followed, and programming support came third. On top of that, 32% of game companies said they used AI while building their own in-house engines.

That ordering matches what studios outside Japan report too. Concept art, texture passes and placeholder assets are the easiest wins, which is why AI 3D model generators moved from novelty to pipeline tool so quickly.

The corporate questionnaire asked companies what they expect to gain. Work efficiency and productivity led the list with 38 companies naming it, followed by shorter development cycles at 30 and lower development and operating costs at 29.

The rest of the ranking is more interesting than it looks. Multilingual support and faster global expansion drew 24 responses, new expressions and ideas 22, content quality 21, and labour shortages 20.

Read together, those answers describe an industry squeezed on two sides. Budgets keep rising, hiring keeps getting harder, and generative AI is being asked to absorb both pressures at once.

Adoption did not arrive without rules. Asked what measures they have in place, developers most often answered that humans handle verification, correction and supervision of anything a model produces.

Two other practices showed up across the majority of responses. Studios specify or restrict which tools staff may use, and they avoid shipping generated output without reworking it first.

CESA executive director Tsutomu Masuda framed the result plainly. “This demonstrates that generative AI is being widely and steadily adopted in development environments,” he said, before adding that the industry still has to answer for infringement risks.

Masuda’s caveat is the honest part of the report. He called it a key challenge to raise the quality of human-centred game development while protecting intellectual property rights.

Meanwhile the friction is visible from outside the survey too. Players and developers have pushed back on how models are trained, and the provenance of training data remains the loudest complaint against studio use.

Japanese studios have already felt it. Level-5 apologised this year over what it called careless use of AI in some videos, and Capcom engineers have said the technology handles routine tasks rather than replacing their creators.

Tokyo Game Show 2026 runs from September 17 to 21 at Makuhari Messe, a five-day format marking the show’s 30th anniversary. CESA timed the report to that stage for a reason.

The full industry report lands later, and it should say which tasks the 85.8% actually run through a model. Until then, the preview already settles the argument about whether Japanese studios use this technology.

Instead the open question has moved on. It is no longer whether generative AI belongs in a Japanese studio, but how much of a finished game a player is willing to accept from it.

If the programming-support side interests you, compare the leading AI coding tools and where each one fits a real pipeline. For the opposite view from inside the industry, read how one open-source project answered with the Godot engine AI code ban.

Sources

AUTOMATON WEST, Inven Global, PC Guide. *Photos: Tokyo-Game-Show-2025 0140 and Tokyo-Game-Show-2025 0030 by RuinDig/Yuki Uchida, CC BY 4.0, cropped to 16:9 and resized.*