A second data point from this week’s Tokyo Game Show 2026 coverage is reshaping how the games industry talks about AI vendors, not just AI adoption. While CESA’s headline figure, 85.8% of Japanese game developers now use generative AI in their work, has dominated the news cycle, a related industry survey of Japanese online game companies carries a number that matters more to Google, Anthropic, and Microsoft: which AI tool developers actually reach for. According to AUTOMATON West’s July 2026 reporting, generative AI tool adoption among Japanese online game companies has hit 100%, and among the tools named, Google’s Gemini leads at 94%, Anthropic’s Claude follows at 84%, and GitHub Copilot trails at 76%. That ordering, not the adoption headline, is the story publishers, platform holders, and AI vendors are now parsing.
This is a genuinely different read than the CESA total. The 85.8% figure describes how many developers touch generative AI at all; the Gemini/Claude/Copilot breakdown describes which company is winning the fight for developer workflows inside one of the world’s most influential game-development markets. For AI vendors, that distinction is worth billions in future enterprise contracts. For studios in the US, Europe, and Korea watching Japan’s trajectory, it’s a preview of which tools are likely to become the default stack for game production over the next two years.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the new Japan gen-AI tool survey actually found
The tool-level numbers come from a July 2026 AUTOMATON West report covering Japanese online game companies, a category distinct from CESA’s broader Tokyo Game Show 2026 developer survey but drawn from the same industry moment. Within that surveyed group, generative AI tool adoption reached 100%, meaning every responding company reported using at least one generative AI product somewhere in its pipeline. Among the tools those companies use, Gemini topped the list at 94%, Claude followed at 84%, and GitHub Copilot came in at 76%. Those percentages aren’t mutually exclusive; many studios run more than one tool depending on task, so the numbers reflect overlapping usage rather than a single winner-take-all market split.
The broader CESA figure, previewed at Tokyo Game Show 2026 and confirmed by Denfaminicogamer’s Japanese-language report, found that 85.8% of surveyed game developers and CESA member companies use generative AI in their work, with 63.0% using it daily and 22.8% using it occasionally. That combined 85.8% is up sharply from 51% in CESA’s 2025 survey, which polled 54 Japanese game companies between June and July 2025. A jump of roughly 35 percentage points in a single year is the kind of shift that normally takes three or four survey cycles in enterprise software adoption. Coverage of the 2026 figure has run across PC Guide, Tech4Gamers, AUTOMATON West, Denfaminicogamer, and Seoul Economic Daily (Sedaily), which tied the number to productivity gains and shorter production cycles cited by Japanese studios attending TGS 2026.
Why Gemini is ahead in Japan’s game studios
Gemini’s 94% usage rate among Japanese online game companies puts it well ahead of both Claude and GitHub Copilot in this specific survey population. Reporting on the underlying numbers doesn’t break down exactly why Gemini leads by discipline, but the pattern lines up with Google’s recent coding gains: Gemini 3.7 Flash scored 65.3% on the DeepSWE v1.1 benchmark in August 2026, up sharply from 49.0% for Gemini 3.6 Flash, giving Google a concrete technical argument for why studios reach for Gemini on production coding tasks rather than just art and narrative generation. That capability jump sits alongside Google’s distribution advantages in the Japanese enterprise market: deep existing relationships through Google Workspace and Google Cloud accounts at major publishers, integration with Android and mobile-first development pipelines that matter heavily for Japan’s mobile-gaming-dominant industry, and aggressive enterprise pricing bundled with cloud infrastructure many studios already use for backend services.
None of the coverage of this survey includes named quotes from CESA officials or studio executives explaining the Gemini lead specifically, so any explanation beyond the raw adoption numbers is inference rather than confirmed reporting. What is confirmed is the ranking itself, and the fact that it comes from a 100%-AI-adoption population, meaning this isn’t a story about whether Japanese online game companies use generative AI. It’s entirely a story about which vendor they reach for first.
Claude’s 84% share and Anthropic’s growing enterprise coding push
Claude’s 84% adoption rate among the same surveyed companies is a strong second-place showing, and it lines up with Anthropic’s broader push into coding and enterprise workflows that began in earnest once Claude Code went publicly available in May 2025 and went on to exceed $500 million in run-rate revenue by September 2025, a push that accelerated further when Anthropic released Claude Opus 5.5 in September 2026, priced at $4 per million input tokens and $20 per million output tokens. Claude has built a reputation in software engineering circles for strong performance on long-context reasoning and multi-file code editing, both of which matter directly to game development, where a single feature change can touch dozens of interconnected systems: physics, rendering, save-state logic, and localization strings simultaneously.
The 10-point gap between Gemini’s 94% and Claude’s 84% is narrow enough that it doesn’t read as a decisive win for either vendor, especially with Claude Code’s run-rate revenue surpassing $2.5 billion by February 2026 and Anthropic also shipping Claude Sonnet 5.5 in September 2026 at $2 per million input tokens and $10 per million output tokens, a cheaper mid-tier model that makes high-volume daily coding use more affordable for studios running Claude across large teams. Both tools are clearly embedded in Japanese game studio workflows at a scale that would have been unusual even 18 months ago, when generative AI in games was still framed mostly around concept art and NPC dialogue rather than production coding and pipeline automation.
GitHub Copilot’s 76% and what it means for Microsoft
GitHub Copilot’s 76% adoption rate places it third among the three named tools in the Japanese online game company survey, still a majority position but a meaningfully lower share than either Gemini or Claude in this specific population. That’s a notable data point for Microsoft, given Copilot’s dominant position in Western enterprise software development more broadly, where Microsoft reported more than 15 million Copilot users as of April 2025 and Rudys.ai tracked the tool crossing 20 million cumulative users by July 2025, adding roughly 5 million users in just three months. It suggests that in Japan’s game industry specifically, general-purpose coding assistants aren’t automatically winning the adoption race against multimodal AI platforms that can also handle art direction, narrative drafts, and localization, tasks that sit outside a pure code-completion tool’s core use case.
Game development is unusually cross-disciplinary compared with most enterprise software categories. A single sprint might require AI assistance for shader code, dialogue trees, marketing copy, and QA test-case generation, all in the same week. That breadth may explain why Gemini and Claude, both positioned as general-purpose multimodal assistants, edge out a tool built primarily around in-editor code completion.
From 51% to 85.8%: the one-year adoption curve
Context matters here, because the trajectory is as newsworthy as the snapshot. CESA’s 2025 survey, run between June and July 2025 across 54 Japanese game companies, found that 51% used AI or generative AI in development, with image and visual-asset generation as the most common use case, followed by story and text generation, and programming support ranking third. A year later, the 2026 preview shown at Tokyo Game Show 2026 puts combined generative AI usage at 85.8%, with the majority, 63.0%, already using it daily rather than occasionally.
That reordering of use cases, from art generation leading in 2025 toward daily, workflow-embedded use in 2026, tracks with the tool-level data showing Gemini, Claude, and Copilot all crossing majority adoption. It suggests Japanese studios moved past the experimental phase (using AI for one-off concept art) and into production-integrated usage (AI tools running inside daily coding, writing, and QA pipelines) inside a single 12-month window.
Japan vs Western game industry AI adoption
Direct percentage comparisons against Western game industry AI surveys, such as the GDC State of the Industry report, aren’t available in current coverage of the CESA/TGS data, so any claim of Japan “leading” Western studios numerically would be speculative. What can be said with confidence is qualitative: commentary accompanying the CESA figures, including Chinese-language coverage from NetEase, frames Japan’s jump from 51% to over 85% as unusually fast relative to how AI tool adoption curves have historically played out in enterprise software, Western game studios included.
Western AAA studios have publicly discussed generative AI use over the past two years, largely around concept art, dialogue prototyping, and QA automation, but few have published adoption percentages as high or as precisely quantified as CESA’s 2026 figures. That gap, in available public data rather than necessarily in actual usage, is itself notable: Japan’s game industry, through CESA, has now published two consecutive years of hard adoption numbers, giving outside analysts a rare longitudinal dataset that most Western markets haven’t produced.
Market impact: what this means for Google, Anthropic, and Microsoft
Game development is a specific, high-value enterprise vertical, and adoption data like this functions as a sales asset whether or not it was intended that way. A 94% usage figure for Gemini among Japanese online game companies is the kind of number that shows up in Google Cloud sales decks aimed at other regional game markets, Korea, Southeast Asia, and eventually North America and Europe. The same logic applies to Anthropic’s 84% figure for Claude and Microsoft’s 76% figure for Copilot: each becomes a reference point in competitive enterprise sales conversations well beyond Japan.
For Google specifically, a leading position in Japan’s game industry matters because Japan remains one of the largest mobile and console game markets globally, home to publishers including Capcom, Sega, Level-5, and Konami, all named by CESA as part of its member-company survey population. Winning developer mindshare inside that ecosystem gives Gemini a foothold that extends into console tooling, cloud backend contracts, and potentially platform-level partnerships as studios standardize their AI stack around whichever vendor proves stickiest in daily production use.
Competitive comparison: Gemini vs Claude vs GitHub Copilot in game development
Each tool brings a different strength profile to game development workflows, based on their public capabilities and the reported adoption pattern (see how these models stack up more broadly in our GPT-6 Astra vs Claude Opus 5 vs Gemini 3.8 Flash comparison and our look at hallucination rates across Grok, ChatGPT, and Gemini). Gemini’s multimodal design, handling text, image, and code within a single interface, maps naturally onto a discipline that constantly moves between visual assets, narrative text, and implementation code, and Google’s newer Gemini 3.8 Flash model backs that up on agentic coding tasks specifically, scoring 19.1% on Terminal-bench 4.0 in September 2026 versus 11.2% for the prior-generation Gemini 3.7 Flash. Claude’s strength in long-context, multi-file reasoning fits large codebases with complex interdependencies, a common feature of game engines built over years by rotating teams. GitHub Copilot’s tight IDE integration remains the fastest path for line-by-line code completion inside established engineering workflows, even if it doesn’t extend as naturally into art or narrative tasks.
| Tool | Vendor | Adoption Among Japanese Online Game Companies | Core Strength in Game Dev |
|---|---|---|---|
| Gemini | 94% | Multimodal: text, image, and code in one workflow | |
| Claude | Anthropic | 84% | Long-context, multi-file code reasoning |
| GitHub Copilot | Microsoft/GitHub | 76% | In-IDE code completion for established pipelines |
Source: AUTOMATON West, July 2026 survey of Japanese online game companies, cited in coverage of the broader CESA Tokyo Game Show 2026 generative AI figures.
Year-over-year adoption: CESA 2025 vs CESA 2026
| Metric | 2025 Survey | 2026 Survey (TGS Preview) |
|---|---|---|
| Overall AI/gen-AI usage | 51% | 85.8% |
| Sample | 54 Japanese game companies | Game developers + CESA member companies |
| Survey window | June–July 2025 | Previewed at TGS 2026, announced Sept. 17, 2026 |
| Daily use among AI users | Not separately reported | 63.0% |
| Occasional use among AI users | Not separately reported | 22.8% |
| Leading 2025 use case | Image/visual asset generation | Not broken out by category in 2026 preview |
Source: CESA surveys as reported by Denfaminicogamer, PC Guide, Tech4Gamers, and AUTOMATON West.
Historical context: from concept art to production pipelines
Generative AI’s entry into game development followed a fairly predictable path across most markets. It started around 2022 and 2023 with concept art generation, tools like Midjourney and Stable Diffusion used mostly for early ideation rather than shipped assets. By 2024, text generation tools began appearing in narrative drafting and localization support. CESA’s own 2025 data confirms that pattern for Japan specifically: image generation led, story and text generation followed, and programming support trailed as the least common use case among AI-using companies that year.
What the 2026 data shows is a reversal of sequencing, not necessarily of proportions: coding-focused tools, Gemini, Claude, and Copilot, are now the names driving adoption headlines, even though none of the current reporting explicitly re-ranks use cases by discipline for 2026. The shift in which tools get named in coverage, from creative-generation platforms to code-assistant platforms, mirrors a broader industry trend where AI moved from being a pre-production aid to being embedded in daily engineering and production work.
A sample AI-assisted CI step Japanese studios are likely automating
While the CESA and AUTOMATON West reporting doesn’t publish specific pipeline configurations, the daily-use figure (63.0%) is consistent with studios wiring AI code review or test-generation directly into their build pipelines rather than using tools ad hoc through a chat window. A typical integration pattern looks like this, using GitHub Copilot’s CLI inside a CI step to flag risky diffs before a build kicks off:
name: AI-assisted PR check
on: pull_request
jobs:
copilot-review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Run Copilot code review
run: gh copilot review --diff-context 500
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}This kind of automated review gate, whether built on Copilot, Claude’s API, or Gemini’s coding endpoints, is the kind of daily-use integration that separates the 63.0% of “daily” users in CESA’s 2026 figure from the 22.8% who use generative AI only occasionally.
Where labor and union reaction stands
Reporting on the CESA 2026 preview and the AUTOMATON West tool survey does not include on-record statements from Japanese game worker unions, artist collectives, or the International Game Developers Association responding specifically to these figures. That absence is worth flagging rather than filling in: it means the adoption story, as currently reported, is a management and vendor-side data point rather than one that yet includes a documented labor response. Denfaminicogamer’s coverage does note that a majority of respondents describe using AI output with human checking, correction, and supervision, a detail that frames the tools as assistive rather than fully autonomous in current Japanese studio practice.
Predictions: where this goes next
Based on the trajectory shown across the 2025 and 2026 CESA figures and the tool-adoption data, a few directional calls look reasonable heading into 2027:
- Expect Google, Anthropic, and Microsoft to each publicize Japan-specific adoption figures in future enterprise sales material aimed at Korean, Southeast Asian, and eventually Western game publishers.
- Expect CESA to publish a third consecutive annual survey in 2027, likely including, for the first time, a discipline-level breakdown (art, code, narrative, QA) given how much attention the 2026 aggregate figure has drawn.
- Expect at least one major Western studio or publisher to publish its own adoption percentage in 2027, following the transparency precedent CESA has now set two years running.
- Expect the gap between Gemini’s 94% and Copilot’s 76% to narrow rather than widen, as Microsoft pushes Copilot’s multimodal and agentic features further into creative and narrative tasks beyond pure code completion.
- Expect renewed attention from labor organizations once a discipline-level breakdown makes clear how much of the 85.8% adoption figure touches art and narrative roles specifically, rather than backend engineering.
What Western studios should take from Japan’s numbers
For US and European studios watching this from outside Japan’s market (a scene we track across our broader gaming industry coverage), the practical takeaway isn’t the specific percentages, it’s the sequencing. Japanese studios moved from experimental, single-use-case AI adoption (largely art generation in 2025) to daily, multi-tool, production-embedded use (2026) inside one calendar year. Studios elsewhere evaluating whether to formalize AI tool policy, rather than leaving adoption to individual developers, now have a concrete precedent showing how quickly that shift can happen once a critical mass of daily users forms inside an organization.
It’s also a reminder that tool choice, not just AI adoption in the abstract, is becoming a competitive and strategic decision, one that’s getting more complicated as vendors blur their own lines: on October 13, 2026, GitHub added Google’s Gemini 4 Argon directly into GitHub Copilot at $2 per million input tokens and $10 per million output tokens, meaning a studio standardized on Copilot can now run Gemini underneath it without touching Google’s own interface. A studio standardizing on Gemini today is making a different bet on future feature roadmaps, API pricing, and platform lock-in than one standardizing on Claude or Copilot, and the numbers behind that bet are now public: Google has priced Gemini 3.7 Flash at $3.75 per million output tokens through December 2026, a rate studios will be weighing against Anthropic’s newer Opus 5.5 and Sonnet 5.5 pricing tiers when they set next year’s AI tooling budgets. Given how close the three tools’ adoption figures are among Japanese online game companies (94%, 84%, and 76%), most studios in that market appear to be hedging by running more than one tool rather than committing to a single vendor, at least for now.
Frequently asked questions
What percentage of Japanese game developers use generative AI in 2026?
According to CESA’s survey previewed at Tokyo Game Show 2026 and announced September 17, 2026, 85.8% of surveyed Japanese game developers and CESA member companies use generative AI in their work, with 63.0% using it daily and 22.8% occasionally.
Which AI tool is most used by Japanese game studios: Gemini, Claude, or GitHub Copilot?
Among Japanese online game companies surveyed by AUTOMATON West in July 2026, Gemini led at 94% adoption, followed by Claude at 84% and GitHub Copilot at 76%. These figures overlap, since many studios use more than one tool.
How does 2026 AI adoption compare to 2025 in Japan’s game industry?
CESA’s 2025 survey, covering 54 Japanese game companies polled between June and July 2025, found 51% AI or generative AI usage. The 2026 figure of 85.8% represents an increase of roughly 35 percentage points in one year.
What was the most common generative AI use case in Japanese game development?
In CESA’s 2025 survey, image and visual-asset generation was the most common use case among AI-using companies, followed by story and text generation, with programming support ranking third. The 2026 preview does not publicly break down use cases by category.
Have Japanese game worker unions responded to the AI adoption figures?
Current public reporting on the CESA 2026 preview and the AUTOMATON West tool survey does not include on-record statements from Japanese game worker unions, artist collectives, or IGDA specifically addressing these figures.
Which major Japanese publishers were included in the CESA survey?
CESA’s member-company sample includes major firms such as Capcom, Sega, Level-5, and Konami, among other CESA members surveyed alongside individual game developers.
Does this survey data apply to Western game studios too?
No. The 85.8% and tool-adoption figures are specific to CESA’s survey of Japanese game developers and CESA member companies, and to AUTOMATON West’s July 2026 survey of Japanese online game companies. No comparable percentage-based figures for Western studios, such as GDC’s State of the Industry report, appear in the current coverage of this data.
When was the 2026 CESA generative AI figure announced?
Denfaminicogamer and other outlets report that CESA announced the 85.8% figure on September 17, 2026, as part of a preview of the CESA Video Game Industry Report shown at Tokyo Game Show 2026.
