A new Tomb Raider used AI to build its world — and players noticed before they played it

Gaming 4 min 24 sources

A buried AI disclosure on a gorgeous remake set off a backlash, and it points at the real pressure underneath: big games cost more to make while selling fewer copies. This week showed every escape the industry is reaching for — cheaper art, owned franchises, subscriptions — plus a huge market it keeps ignoring.

Key takeaways

A beautiful new Tomb Raider remake disclosed it used generative AI in development, and the backlash hit before launch — a sign of the real pressure underneath: big games cost more to make while selling fewer copies.
PlayStation's own data shows first-party game sales fell from 58.4 million units in 2020 to about 32 million now, the squeeze pushing studios toward cheaper AI-made art, owned franchises like Amazon's new Bond deal, and subscriptions.
The counterpoint: IO Interactive's 007 First Light sold 1.5 million copies in 24 hours by simply being good — while analysts note the industry keeps ignoring a huge, moneyed market of older players (a third of over-80s play games).

The shine came off before anyone played it

Crystal Dynamics has been showing off Tomb Raider: Legacy of Atlantis, a glossy Unreal Engine 5 remake of the 1996 original. It looked great. Then fans found a disclosure buried on its Steam store page: the game was made, in part, with generative AI [8].

The wording is careful. “AI-assisted tools were used during development to support some early exploration and temporary development content,” it reads. “Any AI-assisted assets were either replaced or refined by humans” [2][8]. Generative AI here means software that produces images or text from a prompt rather than an artist drawing each one. The parent company, Embracer, has called the technology “powerful” [8]. Some of the goodwill drained out of the project the moment the disclosure surfaced — before a single person had played it.

The system underneath: art is the most expensive, most labour-heavy part of building a big-budget game — armies of artists modelling every rock and ruin. Generative AI promises to do some of that faster and cheaper. That promise is why studios keep reaching for it, and the friction is what happens when players notice.

Why studios are reaching: the big-budget squeeze

The reaching makes more sense once you see the maths. New data on PlayStation’s own games shows the pressure plainly. Sony’s first-party sales — copies of the games it makes itself — peaked in the financial year ending March 2021 at 58.4 million units, then fell to 28.9 million by the 2024 financial year, recovering to 32.1 million in 2025 [11].

Some of that peak was inflated: lockdown-era demand and late PS4 hits like The Last of Us Part 2 padded the 2020 figure [11]. But the trend is real. Big games now take longer and cost more to build, and Sony’s expensive bet on “live-service” titles — games designed to sell to the same players for years — largely misfired [11]. When the cost of making a game keeps rising and the units sold don’t, the margin gets squeezed from both ends. That squeeze is what pushes a studio toward the cheapest version of every input it can find.

The other escape hatch: lean on what you own

Cutting costs is one response. Leaning on a franchise you already own is another. This week Amazon and MGM — the studio Amazon bought, which controls the James Bond rights — said Amazon Game Studios will publish future Bond games made by IO Interactive [5].

It fits a wider Amazon reset. After years of unclear strategy and a retreat from big MMOs, Amazon has refocused its Luna cloud-gaming service — which streams games over the internet so you don’t need a console — on quick party games you can play using a phone as the controller, including one with an AI-driven Snoop Dogg as a judge [13]. The thread tying it together is leverage: owned IP (Bond), a platform (Luna), and AI to make the content cheaply. When making new things is expensive, you reach for the things you already have.

And a reminder it still works when you build it well

For all the cost-cutting, the week’s clearest hit was a game someone simply made well. IO Interactive’s 007 First Light — a James Bond adventure — sold 1.5 million copies in its first 24 hours, the fastest launch in the studio’s history, beating every one of its Hitman games [21]. It managed that before a planned Nintendo Switch 2 version even arrived, and reviewed strongly [21].

It’s the same studio Amazon and MGM just signed for future Bond titles — which is the point. The craft is the asset everyone is bidding on. Cheaper inputs are a way to survive the squeeze; a game people actually love is still the only thing that reliably ends it.

The players nobody is building for

End on the market hiding in plain sight. At a recent industry conference, analysts made an awkward observation: games are overwhelmingly made by and for the young, while older players go ignored. “You have more and more players that are older… maybe 40, 50, 60 years old,” said Newzoo’s Emmanuel Rosier. “And if they are retired, they have even more money — but nobody’s making games for the retired people” [12].

The numbers back him up. An industry survey found 212 million Americans play video games — including roughly a third of those over 80 [24]. Populations in the US, Japan and much of Europe are ageing, and the first generation that grew up with games is now heading toward retirement with time, money, and a lifelong habit [12]. An industry straining to wring savings out of the games it already makes is, meanwhile, barely trying to sell to one of the largest audiences it has. The squeeze is real — but so is the room it keeps overlooking.

02 · Lesson · why it matters

The cost was carrying something the spreadsheet couldn't see

Every expense is quietly buying you a second thing you never put on the ledger — and when you cut the cost, you cut that too.

A backlash before a single play

Here’s the oddest part of the Tomb Raider story: people turned against the game before they played it.

Nothing about how it looks changed. The trailers were still gorgeous. What changed was a sentence on a store page admitting that some of the art had been made with AI. That one disclosure spent something — goodwill, trust, the benefit of the doubt — and it spent it instantly, on a game nobody had touched yet.

The studio was trying to save money on art. It did. But it lost something at the same time, and that something never appeared in the budget it was trying to cut.

Two jobs, one line item

On a spreadsheet, the artists are a cost. A big one — modelling every surface of a game world by hand is slow and expensive. So when you’re squeezed, that line glows red and begs to be trimmed.

But the hand-crafted art was doing two jobs, not one. The first job was visible: making the game look good. The second was invisible: being proof. Proof that humans cared about every corner of this world, that the studio respected the craft, that the thing was made and not generated. Players read that proof, mostly without noticing, and it’s part of why they trust a game enough to buy it.

The cut removed the cost. It also removed the proof. The spreadsheet only showed the first one leaving.

The ledger lies by leaving things out

This is the trap, and it’s everywhere: a spreadsheet puts costs in one column and value in another, as if they were separate things. Often they’re the same thing wearing two labels.

The expensive senior engineer is also the person who quietly mentors three juniors and remembers why the system was built that way. The “slow” testing team is also your reputation for things that work. The maintenance you keep deferring is also the reason nothing has broken. The costly customer-service line is also the loyalty of the people who called it and felt heard. In every case the cost is real — and in every case it’s holding up a second thing that the accounting never names.

Cut the line, and the cost does fall, exactly as predicted. The second thing falls too, quietly, somewhere the numbers weren’t looking.

Why the cut always looks free

There’s a reason this mistake is so easy to make over and over. The two halves arrive on different clocks.

The saving is immediate, visible, and easy to count. You cut the cost today and the budget improves today; you can put it in a report and take credit for it. The loss is delayed, diffuse, and hard to measure. Trust erodes slowly. The mentor’s absence shows up months later in juniors who never grew. The deferred maintenance fails on some random Tuesday you can’t trace back to the decision.

So on the day you make the cut, it looks free — pure savings, no downside. The bill is real, but it’s posted to a different date and a different column, and by the time it lands, almost no one connects it back.

Ask what else it was holding up

You can’t keep every cost, and you shouldn’t. Some expenses really are just expenses. The point isn’t to never cut — it’s to cut with your eyes open.

Before you trim a cost, ask one question: what else is this quietly holding up? What’s the second job this line item is doing that nobody wrote down? If you can’t see a second job, that isn’t proof there isn’t one — it’s usually just proof it’s invisible, which is exactly how these things hide.

The studio saw the price of the art. It didn’t see that the art was also the trust. The most expensive things you pay for are often expensive precisely because they’re carrying more than one thing at once — and the surest way to find out what a cost was buying you is to cut it, and watch what leaves with it.

03 · Lab · your turn

The Hidden Bill

Rehearse cutting costs by hand-vs-AI on a game, watching the visible savings climb while the player-trust bill stays sealed until launch.