Direct answer — Should a game studio build its own AI tooling or buy an existing platform in 2026? Most game studios should buy: commercial AI tools and model APIs already cover coding, research, concept work and asset iteration at a cost that is trivial next to a single salary. Build only when the AI capability ships to players, rests on proprietary data no vendor has, and can be funded for several years, which is the bet Krafton is making with its GPU cluster. Large publishers that need custom workflows without owning model research have a third option, the partnership route EA took with Stability AI.
The AI tooling build vs buy question has moved from the CTO’s backlog to the board agenda. It is no longer about which plugin your artists install. It is a capital-allocation decision: a recurring payroll line, a compute bill, and a multi-year commitment that competes directly with your next game for the same money.
In my 20+ years across Gameloft, SFR, Blacknut and now as a board member at Impulse Media Hub, I have watched the same pattern repeat with every new technology wave, from mobile distribution platforms to cloud gaming: studios overbuild infrastructure that later became a commodity, and underinvest in the integration work that actually changed their output. AI is following the same script, only faster. When I work with a studio as a video game consultant, the AI investment question now comes up in almost every strategy review, and the right answer is rarely the one the engineering team proposes first.
Key Takeaways
- Buy is the default for any AI use case where the model is a commodity and your edge lies in how you apply it.
- Build is justified only when AI is part of the player-facing product, rests on proprietary data, and has multi-year funding.
- Partnering, as EA did with Stability AI, buys custom expertise without owning model research.
- Krafton’s GPU cluster, valued at roughly KRW 100 billion ($70 million), is a strategic bet for a company of Krafton’s size, not a template for a mid-size studio.
- The hidden cost of building is time to first useful tool, not salaries.
Three ways to invest in AI tooling: build, buy, partner
The game studio AI investment decision in 2026 has three distinct shapes, and they carry very different risk profiles.
Buy means subscribing to commercial tools or calling model APIs, then wiring them into your existing pipeline. Build means employing people who train, fine-tune or host models and own the tooling around them. Partner sits between the two: a model provider co-develops tools with your teams, and you keep control of the workflow and the creative direction.
| Dimension | Buy (tools and APIs) | Partner (co-development) | Build (in-house team and compute) |
|---|---|---|---|
| Typical cost shape | Per-seat or usage subscription | Contract plus internal integration team | Payroll plus compute, recurring |
| Time to value | Days to weeks | Quarters | Several quarters before daily use |
| Differentiation | Low, competitors use the same tools | Medium, workflow is yours | High, if the capability ships to players |
| Lock-in risk | Vendor pricing and terms | Partner dependence | Talent retention and model obsolescence |
| Who it fits | Most studios, any size | Large publishers with custom pipelines | Publishers betting the product on AI |
Rough cost orders of magnitude
There is no published benchmark for the cost of an in-house AI team at a game studio, and I would distrust anyone quoting one to the dollar. The figures below are rough planning ranges from my own practice, meant to set the order of magnitude, not to replace your own payroll model.
- Buy: commercial AI subscriptions typically run from roughly $50 to $500 per month per tool or team tier. Even across a whole studio, this rarely exceeds a fraction of one salary.
- Build, minimal team: three to four people (an ML engineer, a pipeline engineer, a data engineer, part of a technical lead) at roughly $420,000 to $590,000 a year fully loaded in Western Europe or North America, before compute.
- Build, infrastructure bet: Krafton’s GPU cluster, reported by Game Developer as “valued at roughly KRW 100 billion ($70 million)”. This is the only figure in this section drawn from a public source.
The ratio matters more than the precise figures. One year of a minimal in-house team buys many years of commercial tooling for an entire studio. Building has to produce something a subscription cannot.
Krafton vs EA: two opposite bets
The two most visible publisher moves of the past year illustrate the AI game development tools decision at its extremes.
Krafton: build, and make AI the product
In the English version of its Q3 2025 fiscal report, Krafton explained it is “accelerating towards becoming an AI First company,” according to Game Developer. The infrastructure behind that statement is substantial: a GPU cluster valued at roughly KRW 100 billion ($70 million), and a partnership with SK Telecom “to develop a proprietary foundation model with ‘five billion parameters’.”
The key point for a CFO is why Krafton builds. Its stated goal is to position itself as a leader in game AI technology and advance Co-Playable Characters (CPCs), with its first CPC, PUBG Ally, planned for PUBG: Battlegrounds Arcade Mode in the first half of 2026. The Krafton AI first strategy is about the product players touch, not about cheaper textures. That is exactly the condition under which building makes sense.
EA: partner, and keep artists in charge
EA took the opposite route. In October 2025 it announced a partnership with Stability AI to co-develop “transformative AI models, tools, and workflows.” The first targets are practical: faster creation of Physically Based Rendering (PBR) materials, and systems that can “pre-visualize entire 3D environments from a series of intentional prompts.”
The structure of the EA Stability AI partnership is the instructive part. Stability AI CEO Prem Akkaraju described “embedding our 3D research team directly with EA’s artists and developers.” EA does not need to hire and retain a model research team; it borrows one, and keeps ownership of the workflow and the creative standards. EA framed the intent as making AI “a trusted ally: supporting faster iteration, expanding creative possibilities, accelerating workflows,” and its Head of Technical Art for EA SPORTS, Steve Kestell, called the tools “smarter paintbrushes.”
Neither company is wrong. They are solving different problems. Krafton wants AI in the game. EA wants AI in the pipeline. Your decision starts with knowing which of those two problems you have. For the wider strategic picture of where AI agents are taking the industry, see my analysis of AI agents in gaming and what they mean for studios.
Weighing an AI budget line against your next production? Book a Strategy Call and we will test the business case before you sign the first hire.
A decision framework for the build vs buy question
I use five questions with studio leadership teams. If you answer “no” to the first two, the decision is made: buy.
- Does the AI capability ship to players? AI-driven characters, systemic content, personalization inside the game. If it only touches the back office, buy. Player-facing AI also brings disclosure and ownership obligations, covered in our Steam AI disclosure policy and copyright guide.
- Do you own data no vendor has? Years of telemetry, a large proprietary asset library, a distinctive art style codified at scale. Without proprietary data, your custom model will trail commercial ones.
- Can you fund it for three years? A team that is cut after twelve months has produced cost, not capability.
- Can you hire and keep the talent? ML engineers compare your offer with AI labs, not with other studios.
- What does the integration layer need? Even when you buy, someone must connect tools to your pipeline, asset standards and review steps. This is where internal engineering time pays off.
What adoption tells you about the market
The GDC 2026 State of the Game Industry survey, based on more than 2,300 game industry professionals, found that “over one-third (36%) of game industry professionals are using generative AI tools.” At game studios, “30% of respondents at game studios reported using AI tools,” against 58% at publishing companies, support teams and marketing/PR firms. The same survey found that “over half (52%) of game industry professionals think generative AI is having a negative impact.”
Two consequences for the build vs buy decision. First, adoption inside studios is still partial, so any tool you build or buy needs a change-management plan, not just a budget. Second, sentiment is a real constraint: a team that distrusts generative AI will not use an internal tool simply because you paid for it. Starting with bought tools on low-stakes tasks is the cheapest way to learn what your people will actually adopt.
Where studios go wrong
The mistakes I see most often in AI game engine adoption and tooling projects are not technical.
- Building what is already sold. An internal code assistant or asset tagger, built over two quarters, that ends up worse than a commercial tool the team could have used on day one.
- Hiring before scoping. A head of AI is recruited, then asked to find a use case. Define the three workflows to change first, then decide who is needed.
- Ignoring time to value. In-house AI teams usually need several quarters before production staff rely on their output daily. Budget for that gap explicitly.
- Treating build vs buy as permanent. Buy first, measure where commercial tools fall short, then build narrowly where the gap is proven. This is the same logic I recommend in the lean studio model: keep fixed costs low until a capability earns its place.
The same trade-off applies to people as well as tools. Deciding when external expertise beats a permanent hire is a question I cover in external growth consultant vs in-house team.
What we could not verify
Several figures circulate on this topic without a primary source behind them. I could not find a published, dated benchmark for the cost of an in-house AI team at game studios specifically, so the team range above is presented as my own planning estimate. I also found no public figure for the financial value of the EA and Stability AI partnership or for its duration, and none is stated here. Finally, no source I checked quantifies productivity gains from AI tooling in game production with a stated methodology, so this article makes no percentage claim on time saved.
Conclusion
For most studios in 2026, the right AI investment is a bought toolset, a small internal integration effort, and a clear list of workflows to improve. Build when AI is the product, the data is yours, and the funding horizon is long. Partner when you need custom workflows at publisher scale. Krafton and EA show that both extremes can be rational; what is never rational is building a commodity.
Deciding whether to build, buy or partner on AI? Book a Strategy Call with Game Growth Advisor and we will map your use cases, costs and risks into a decision your board can approve.