🎮 How is Gen. AI Impacting the Gaming Sector?
Overlooked #168
Hi, it’s Alexandre from Eurazeo. I’m investing in seed & series A European vertical solutions (vSol) which are industry specific solutions aiming to become industry OS and combining dynamics from SaaS, marketplaces and fintechs. Overlooked is a weekly newsletter about venture capital and vSol. Today, I’m sharing a post that is not about vSol but focuses on the potential impact of Generative AI on the gaming sector. As both a gamer and an investor, I’ve dedicated considerable time over the past few months to pondering this topic and discussing it with entrepreneurs, starting with Olivier and Daniel at Homa. I believe it would be beneficial to share some insights and ideas with the broader world.
Today, I’m sharing a collaborative piece that Camille and I wrote with Marie and Roméo at Fly on the impact of Gen. AI on the gaming sector.
We divided this paper into two sections: (i) general insights on the gaming sector, (ii) a mapping of 80+ companies building gen. AI infrastructure for the gaming sector combined with key learnings on the space.
Section 1: General Insights on Solutions for the Gaming Market
What does it take to generate $100m in revenues from gaming companies?
Gaming is a paradoxical industry for infrastructure startups. On the one hand, gaming is the largest form of entertainment, generating more revenues ($216bn) than music ($36bn), over-the-top video ($114bn), and cinema ($46bn) combined.
On the other hand, it’s extremely hard to build a large software company selling to the gaming industry for several reasons: (i) there is a massive long tail of independent gaming studios which don’t have strong buying power, (ii) there is a mental ceiling for gaming studios preventing them from paying more for software than what they would pay for their game engine, and (iii) it’s a project-based industry making it reluctant to pay for a recurring subscription for SaaS.
Historically, only three infrastructure categories have managed to build $100m annual revenue businesses by selling to gaming companies: (i) ad platforms (e.g. Unity, AppLovin, Meta, Google), (ii) game stores (e.g. Roblox, Steam, Apple App Store, Google Play Store), and (iii) game engines (e.g. Unity, Unreal, Roblox). Ad platforms and game stores reached this milestone by indexing their revenues to the gaming industry, as they take a cut of the gaming revenues generated by their customers. Third-party game engines are must-have platforms for producing games and only a few AAA studios are still building games on proprietary game engines.
Overall, it’s extremely hard to build a $100m annual revenue business selling infrastructure to gaming companies. For startups, we view several paths:
Use the gaming sector as a beachhead to build and commercialise your solution before moving into other sectors (e.g. Unreal Engine now being used in construction),
Index your revenues to the gaming sector either by taking a cut of revenues generated by gaming companies or by using your technology to build your own games,
Have a product vision in which you can replace incumbent game engines in the long term, which can make sense with an AI-native game engine and in a market where Unity is under massive pressure from the gaming developers' community,
Tap into professional services budgets (”sell work, not software”) by convincing gaming studios to pay unprecedented ACVs for gaming infrastructure companies because you’re replacing the work that certain functions would usually do (e.g. 2D or 3D artists, voice actors, etc.). Several horizontal applications like Midjourney ($200m in revenues) and ElevenLabs ($1.1bn valuation) are already growing commercially at an exponential path. We would be curious to know how much they make from the gaming vertical.
What workflows in the gaming value chain can be augmented by Gen. AI?
The gaming value chain can be broken down into six distinct phases: ideation, pre-production, production, testing, launch, and live ops, all of which can be accelerated by the adoption of generative AI tools.
Bain estimates that up to 50% of the gaming value chain can be automated with generative AI. This technology helps to both decrease production costs and increase development speed and output quality. For specific production workflows the impact will be even bigger.
On level generation, casual puzzle games startup Cosmic Lounge has integrated multiple AI models into their in-house technology to increase the efficiency its of game level content creation and testing by 5x. Homa releas
On ideation, mobile game publisher Homa released its Game Idea Generator to empower its game creators to generate game ideas with strong potential leveraging mobile gaming market data, social & cultural trends trends as well as its proprietary data-set on games Homa tested and published.
New studios like AI Roguelite and RealSpawn are pushing the limits of integrating generative AI throughout the entire gaming value chain by making it a core component of their gameplay. AI Roguelite is a text-based role-playing-game where everything is AI-generated (location, NPCs and items) while RealSpawn is building a MMORPG where new content is created based on how players interact within the game.
Will gaming companies build or buy AI tooling?
1/ Can a midsize studio pull this off?
Let’s consider Embark Studios and its game The Finals as it’s one of the rare studios that has shipped a live game incorporating AI.
Embark has 300 employees but only 6 AI roles (3 ML Engineers + 2 ML Research Engineers + 1 ML Ops) making up 2% of the company. To be fair, they also have 2 Data Scientists and 1 Data Ops, but we assumed these roles are focused on product data science/analytics.
You need different technical experts to build a game. Embark has a numerous Software Engineers (good taste too, Rust engineers), Game Engineers, Designers, UI experts etc. It does not leave a lot of room for developing in house generative AI tech.
On the contrary, Stability AI has 200 employees and 80 of them are AI roles split between Research Scientists, ML Engineers, ML Ops, etc. It means that 40% of the company is dedicated to AI research and engineering.
[Of course, the ability to build AI centric products involves more than just the number of ML researchers you can hire. It’s not just about the numbers.]
2/ What about the really big studios?
The larger gaming corporations are, in the end, a collection of teams working on many games (so if we zoomed in on a given team, we would find a situation similar to that of Embark). However, they have more resources allocated for central functions.
Some of the tier-one studios do have identified R&D teams.
What if we delved deeper into these teams and spent an infinite amount of time on Sales Navigator to identify who among these teams has the right profile to be working on AI tech, count them and track their educational backgrounds? How many relevant profiles would we find?
Given that we’re VCs with infinite time to spare, we did exactly that.
SEED at EA: ~27 relevant profiles
La Forge at Ubisoft: ~15 relevant profiles at La Forge + Ghostwriter team (a bit hard to know because people in China don’t use Linkedin)
NERD at Nintendo: ~21 relevant profiles
SONY AI at Sony: ~33 relevant profiles
These teams support various existing transversal projects, and some have to publish papers…
3/ Sure, we can all agree that companies like Ubisoft and EA may not be the birthplaces for groundbreaking AI models such as Stable Diffusion. But, could they have developed something similar to Inworld AI?
Inworld has 80 people and 15% AI profiles (including 2 from MIT, 2 from Carnegie Melon, 1 from Stanford) making up 19% of the company. So the answer is: probably not.
To achieve this, they would need to allocate 50-100% of their top R&D talent for 1-2 years to a project of Inworld's caliber. Even if they possess the necessary skills in-house, it would be very risky.
To achieve this, they would have had to dedicate 50 to 100% of their top R&D talent for 1-2 years for a project like Inworld. Assuming they have the exact skills in-house, it would be very risky.
Is hiring a new team an option? Maybe.
Yet, what would be the cost? ChatGPT was kind enough to calculate that for me: about $2.4m for a year of work, which obviously includes risk of technical failure and delay.
An interesting example is how much noise Ubisoft makes with Ghostwriter, which has a very limited mission statement (it produces… AI generated barks). It does do the job to prove that they can pull off something of ~1/10th of the complexity of an Inworld.
4/ Exceptions to the rules:
If the traditional gaming powerhouses fall short, we would keep a close eye on the following companies:
King (the people who did Candy Crush): arguably the company with the densest concentration of Data Scientists per square meter in the gaming industry. If there’s a new way to make more money out of the Candy Crush Saga, they’ll make it happen.
Epic (they’re cheating as they are already a technology company): home of Unreal Engine, they acquired Sketchfab in 2021, pretty much the only fully labeled database of 3D models out there. See where we are going with this?
”Creating 3D barrel rolls, carts and similar elements for the game background is a nightmare. It's not the main focus of your game, so you don't want to spend time on it, but today it takes a lot of time to do. There are probably random corridors in Star Wars Jedi: Survivor that might have cost more than $10k to make. However, the size of available databases for 3D assets is very small compared to image databases.” - Rémi Malet (founder at Irregular Shapes)
Roblox (also a technology company) has a seriously strong team with at least 300 ML profiles [we kind of stopped counting after 300].
Microsoft/Xbox - you know the people who casually invested $10bn in OpenAI.
How gaming incumbents are integrating AI into their operations?
Section 2: Mapping Gen. AI Startups Going After the Gaming Industry
Introducing our Mapping of Gaming Gen. AI Startups
We have categorized promising companies into three main areas:
At the infrastructure level, Gen. AI has the potential to enhance various steps of the gaming value chain. During the pre-production phase, it can facilitate game ideation and prototyping. Subsequently, technologies like code generation are poised to improve efficiency in production, testing, and liveops. These technologies can also introduce new gameplay dynamics, such on-the-spot customisation.
Asset generation is arguably the area where we see most startup creations. Every asset involved in game production (from marketing materials to in-game landscapes, character voices, etc.) is concerned but with varying level of tech-readiness. While 2D, voice, dialogue, NPC and texture generation are already used, 3D, animations, avatar, and music asset generation remain in the experimental stage.
Leveraging the advancements in the two previous categories, companies are forging new gaming experiences. AI-Assisted studios are using AI to drastically lower the cost development time for games that are traditionally expensive to produce. These companies are distinct from AI-first studios, which concentrate on gameplay innovation. Moreover, Generative AI is catalyzing a new wave of prosumer platforms that democratize game creation for a broader audience."
Key Learnings
Learning n°1 - Of Course, the Gaming Industry is Poised to Adopt Gen. AI, Leading to a Significant Increase in Productivity that will Make Game Development Cheaper and Faster…
This is the basic conclusion likely to be drawn by any consultant (hi Bain! 👋), equity researcher or venture capitalist. However, when we sought feedback from industry insiders, they told us that this conclusion is rather simplistic…
“Game content creation processes become more automated over time, reducing the cost and complexity of making games. Higher quality and more tailored content boosts consumer engagement and monetization.” “Our bull case calculations on AI show that AAA game development processes ($250m budget, 4y development, 1y game testing & scaling) could see cost savings ranging from 4% to as much as 15% of their estimated budgets today.” - Morgan Stanley
“It's always felt too linear to extrapolate that AI will lead to building cheaper AAA games at a faster pace. The main impact is that the definition of what a 'game' is will evolve, as it has always been in the history of games when a new tech or business model innovation came to the market. - Levi Fussel (co-founder at Martian Lawyers Club)
”Thinking that generative AI unlocks cheaper AAA games is the same as saying that 10 years ago, 'digital transformation' was driven by blue chip corporations. We will always go for bigger games but not cheaper games.” - Charles Thomas (co-founder at Nocturne Games)
Learning n°2 - … But the Biggest Impact of Gen. AI on Gaming will be the Creation of Unseen Gaming Experiences
Gen. AI is set to unlock a new era of innovative and engaging games, with a more pronounced impact on consumer experiences than on studio productivity.
Independent and challenger studios are poised to lead this evolution, as they face a lower risk of social disruption associated with the adoption of Gen. AI, a point underscored in numerous interviews and articles.
Learning n°3 - The Adoption Curve Will Vary Based on the Tech-Readiness of the Different Use Cases
Although companies are developing generative AI for a variety of use cases, the readiness of these applications for gaming varies greatly. Generally, the solutions become less mature as they move deeper into game production.
Many studios are currently testing generative AI on use cases adjacent to core game development such as marketing, concept testing, etc. Taking asset generation as an example, many studios have begun exploring how they can leverage it to create marketing assets and to test concepts quickly. However, using it to create in-game assets still one step away for most of them.
Learning n°4 - The Social Unrest is Real and is an Understated Problem for all Studios
Some people refuse to adopt AI technologies entirely. Others are very vocal against it on social media and within their own companies. There are powerful professional associations such as SAG-AFTRA in the US, initiating strikes over the use of AI. This issue should not be brushed off, as it causes internal mini or maxi crisis in all studios. Management must take a stance, which can lead to either everyone aligning exceptionally well with increased motivation and productivity after a team-wide consultation, or having to fire the black ship elements in order to be able to move forward.
Learning n°5 - Medium-Term Impact on Low-Level Artistic Jobs
Quite quickly, low-level artistic jobs will no longer exist, mirroring the impact of Microsoft Office on administrative assistants, CAD on industrial drafters, and 3D animation on traditional animators. It's understandable that people are concerned; they will need to re-skill.
There was this fun journalistic moment where the narrative was: "OMG, AI disruption is actually affecting 'white-collar' jobs". However, this isn't entirely accurate. Most graphic design roles are more akin to 'blue-collar' work. Those who can produce exceptional IP have never been safer. Conversely, individuals capable of producing only passable copy are now obsolete and will need to increase their output fivefold by utilizing new AI-assisted tools to remain relevant.
Large studios will likely continue to produce the same games as before, incorporating minor innovations and increasing their profit margin without fully integrating AI in gaming. Their shareholders will be happy for a while and then, they'll be commoditized and disrupted. Their attempts at doing in-game AI will be as underwhelming as current efforts by platforms like Snapchat, Facebook, Netflix, and YouTube to integrate gaming into their interface.
Learning n°6 - Gen. AI will Concentrate the Gaming Power Law
The entertainment business is driven by a power-law in which a small portion of productions capture most of the revenues. This pattern is true for the film and the music industry. Gaming is no an exception. In 2023, 14k games were released on Steam but only 930 made more than $1m in revenues, 157 more than $10m in revenues, 20 more than $50m and the top 10 accounted for 61% of the total revenue generated by new releases. We believe that Gen. AI will accentuate this phenomenon by increasing the quality and distribution gap between top games and other games.
“There will be fewer games, but people will play them longer. AAA games won’t be less expensive to produce by any means, but they will have a huge durability (let’s see what happens with GTA 6). So in my opinion, there will be fewer [successful] games, that people play longer because you are more attached to them. It’s like TV, we switched from movies to super long TV series like Game of Thrones and big multi-modal franchises (and reboots), because people are attached to certain IPs and want to experience them for a longer period”. - Rob Auten (writer and founder at Hexagram)
“In a world where Gen. AI enables everyone to create a video game, distribution will become more crucial than ever to stand from the crowd. This phenomenon mirrors what occurred in the music industry with the democratisation of music creation.” - Daniel Nathan (CEO & cofounder at Homa)
Learning n°7 - Building the next Unity or Unreal Engine
“It's a very important moment to be completely upfront here. At Unity, we have basically concluded that the developments are going so fast right now, we can't be the builders of all this. We need to bring the builders to our customers, the game developers, and that's what we are doing with this marketplace.” Danny Lange who is SVP of AI at Unity in Cross Validated
Both Unity and Unreal are encountering an Innovator’s Dilemma with the advent of AI. Both companies are suffering from a post-covid and post-ZIRP hangover pushing them to lay-off employees (Unity, Epic), to refocus on their core business or to make questionable pricing changes. Their current focus on improving profitability may deter them from fully embracing the AI revolution aggressively.
It opens a door for new entrants capable of developing an AI-native game engine that could counter position Unity and Unreal. You can imagine a future in which you could go from an idea to a fully functioning game by giving successive prompts to an AI native game engine. You would not need to code to build a game opening the gaming creation process to many more creators. We’ve seen several AI infrastructure startups pitching us the long term vision of building the next game engine based on the AI paradigm.
Learning n°8 - Crossing the Last Mile
For many use cases, projects are not yet production ready especially as we delve deeper into production elements like 3D assets and game mechanisms such as dynamic & on-spot customization. Numerous gaming studios have experimented with new tools but often find themselves needing to rework the results. Thus, they revert to using foundational horizontal tools that most projects rely on, like GPT and Stable Diffusion, in combination with traditional editing tools. To drive adoption, gaming-specific companies must achieve a higher degree of product completion and offer significant added value beyond what these horizontal tools provide.
More specifically, AI gaming infrastructure companies should:
Leverage customers’ data to fine-tune their models,
Be integrated into workflows and tools used by gaming companies (e.g. exports in game engines or ad platforms),
Offer editing capabilities to refine AI outputs based on human feedbacks.
Conclusion
More than a conclusion, we offer a prediction! Fueled by Gen. AI, new challengers gaming studios will emerge and have an opportunity to become $10bn+ companies, reminiscent of the rise of Roblox and Epic Games during the last industry shift around social gaming platforms.
"The next awesome game (i.e. the next Fortnite or Among Us like phenomenon) with AI at the core of the gameplay will come from someone you've never heard of, and it will be huge. It won’t come from a large studio for several reasons including social issues, low willingness to take risks and lack of AI talents. On that latter point, you need very new skills in AI to build anything truly different, and unfortunately, large studios have very little talent in AI at the moment”. - Travis Boudreau (CTO of Azra Games)
(cringe but bear with me:) these studios will in the end create the metaverse - unless there is still a key technology we still don't have. (This prediction will abide by the Rule 34 of the internet)
These challenger studio will be tech companies characterized by:
A high concentration of AI talents (>20% of the company). It means that at least one founder has to be an AI person because you can't hire good AI people out of nowhere,
Visionary creative directors capable of creating new emotional experiences for their users,
A commitment to addressing significant ideas that have an impact on their users beyond the gaming experience itself,
Robust gaming operations (e.g. capable of very fast & effective creative iterations).
Thanks to Julia (🦒), Marie, Roméo and Daniel for the feedback! Thanks for reading! See you next week for another issue! 👋
Impressive!
Just a little reflection regarding learning no.6:
In my opinion, the success of Palworld has approved that Gaming Power Law will be less valid in the short term, or at least there will be a window period, because AI makes basic art contents very cheap, so that mid team can also make great games with amazing concepts and average game quality. The industrialized pipelines of big companys are no longer an unchallengeable barrier.
However, in the long term, when the big players have integrated AI into pipelines, the window will close.
Great article, Alexandre! Wonderfully researched and thoughtful. The future for AI in Gaming will be very, very interesting.