Project Genie: Google's AI that wiped $47B off gaming stocks

What is Project Genie and how it works

Let me break this down: imagine describing a world to an AI and, within seconds, being able to walk through it. That's exactly what Project Genie does — Google DeepMind's new experimental project launched on January 29, 2026.

This isn't a pre-recorded video. Project Genie generates interactive 3D worlds in real time, at 24 frames per second and 720p resolution. Type "a medieval forest with a castle in the background" and the AI creates that environment for you to explore with your character. Frame by frame, live.

The system runs on three AI components working together:

• Genie 3: The core engine. An 11-billion-parameter autoregressive model that generates worlds frame by frame. It doesn't use a traditional physics engine — it learns how objects move, how gravity works, and how environments interact through self-supervised learning.
• Nano Banana Pro: Google's image generation system that works as a "world sketcher." Before entering a world, you can preview what it'll look like and tweak details.
• Gemini: Orchestrates everything — camera controls, character movement, and coordination between the other two systems.

Shlomi Fruchter, Research Director at DeepMind, put it this way: "Genie 3 is the first real-time interactive general-purpose world model."

Three ways to use it

Project Genie offers three interaction modes:

1. World Sketching: Create an environment from scratch with a text prompt. Nano Banana Pro shows you a preview before generating the full world.
2. World Exploration: Navigate freely through the generated world. You can walk, jump, and even paraglide.
3. World Remixing: Modify an existing world by adjusting the original prompt.

Demos have shown photorealistic forests, sci-fi landscapes, cities, and fantasy environments — in both first-person and third-person views.

But here's a crucial detail many are overlooking: each generation lasts 60 seconds maximum. The model's memory extends roughly one minute back. There are no complex game mechanics, no objectives, no win conditions. And you can't export what's generated to engines like Unity or Unreal Engine.

The Mario and Zelda scandal

This is where the story gets interesting. Journalist Jay Peters from The Verge demonstrated on Bluesky that Project Genie could generate Nintendo-style worlds with minimal resistance. He posted: "Google's new world AI model tool let me generate a bunch of Nintendo-inspired games. Including one featuring Link with a paraglider!"

Documented replications include:

• The Legend of Zelda: Breath of the Wild (with functional gliding mechanics)
• Super Mario 64 (walking and jumping)
• Metroid Prime 3/4 (combined version)
• Kingdom Hearts (featuring Donald Duck, Sora, and Cloud — though this generation was eventually blocked)

Google's safeguards exist but are inconsistent. When Peters attempted more Super Mario 64 generations, the system eventually blocked them, citing concern for "interests of third-party content providers." But initial generations went through without filters.

Nintendo, known as one of the most aggressive IP protectors in the world, hasn't taken formal legal action yet. But as Nintendo Life pointedly asked: "When will Nintendo's legal ninjas strike?"

The historical parallel is clear. When OpenAI launched Sora 2 in October 2025, copyrighted characters appeared freely until legal threats arrived. Two months later, OpenAI and Disney struck a licensing deal. A similar trajectory is likely for Google.

$47 billion wiped out in hours

The January 29 announcement triggered what Tom's Hardware described as a "gaming market meltdown," briefly erasing approximately $47 billion in market capitalization.

Here are the most significant drops:

But was this reaction rational?

The short answer is no. And it's not just my take.

Jefferies characterized the market reaction as "overdone," arguing these tools "can help spark new ideas and creativity, but they are not advanced enough to replace game engines like Unity."

Andrew Green, CEO of Storygrounds, was blunter: "The market reaction tells you everything about how little Wall Street understands games."

Eric Kress from Gossamer Consulting pointed to Google's gaming track record: Stadia, YouTube Gaming. "Google lacks the organisational DNA for gaming," he concluded.

By Monday, February 3, most stocks had started recovering. Unity was up 3%+, though none had reached pre-announcement levels.

What Project Genie can't do (yet)

Before declaring the gaming industry dead, here are Project Genie's real limitations:

• 60 seconds maximum per generation. A typical game lasts hours.
• 720p at 24 fps. Modern games run at 4K and 60-120 fps.
• No game mechanics. No combat systems, inventories, quests, or progression.
• No export capability. You can't bring generated content to Unity, Unreal, or any other engine.
• Noticeable input lag. The delay between your action and the response is perceptible.
• Can't generate readable text or handle multi-agent interactions.
• Can't replicate real-world locations with geographic accuracy.

As a PYMNTS analyst noted: "Fears that tools like Genie could quickly disrupt the gaming industry are misplaced."

There's an enormous gap between a generated environment and a game. An environment is a space. A game is a designed experience with rules, narrative, progression, balance, and feedback. Project Genie does the former. It's nowhere close to the latter.

The AGI connection

Here's the twist many are missing. For Google, Project Genie isn't about video games. It's about artificial general intelligence.

Demis Hassabis, DeepMind's CEO, argued on CNBC that large language models alone cannot achieve human-level intelligence. What's missing: "internal world models that capture causality and physical dynamics."

This is where Genie connects to SIMA, another DeepMind project. SIMA is a generalist AI agent for 3D environments. The concept is a closed loop: Genie generates worlds, SIMA explores and learns from them. Whatever SIMA needs to learn, Genie can create on the fly.

Hassabis explained: "Whatever the SIMA agent is trying to learn, Genie can basically create on the fly."

It's an autonomous training system where two AIs feed each other. Genie creates increasingly complex environments; SIMA develops increasingly sophisticated skills to navigate them.

How long until AGI, according to Hassabis? Five to ten years, requiring "one or two more breakthroughs" in model reliability, reasoning, and memory.

Genie's evolution: from 10 seconds to minutes

To understand where we are, it helps to see where we've been:

Genie 1 generated basic 2D environments with no real interaction. Genie 2 improved responsiveness and rendering, but its memory capped at 10-20 seconds — turn the camera and come back, and the world had changed. Genie 3 is the first to maintain consistency for minutes and respond in real time.

TIME included Genie 3 in its "Best Inventions of 2025" list. The key difference from video generators like Sora or Runway is that Genie doesn't produce predetermined sequences. It generates frames auto-regressively based on both the initial prompt and the user's ongoing actions.

Who can use it and what it costs

Project Genie is available exclusively to Google AI Ultra subscribers in the United States, for users 18 and older.

The price: $249.99 per month (or $124.99/month with the introductory discount for the first 3 months). The AI Ultra subscription also includes 30 TB of cloud storage, Veo 3.16 video generation, Deep Think, Gemini Agent, and 25,000 monthly AI credits.

Free users can only interact with pre-built demos.

Is it worth it for indie developers? Realistically, as a rapid prototyping tool, it could be interesting. But as a game engine replacement, not remotely. Indie developer Tomoko Saito put it well: "I can imagine using tooling like this to brainstorm spaces faster, but I'd never ship something straight from GenAI."

Unity CEO Matthew Bromberg also weighed in, calling AI world models a "powerful accelerator" for the industry — not a threat. He anticipated Unity would integrate generative models into existing tools rather than competing against them.

What this means for the future of gaming

Project Genie won't kill the gaming industry. But it does mark an inflection point.

The ability to generate playable environments from text in real time was science fiction two years ago. Today it exists — with serious limitations, but it exists. And if you look at the progression from Genie 1 to Genie 3 (from static 2D environments to navigable 3D worlds in real time), the trajectory is clear.

The analyst consensus points to a hybrid workflow: AI-generated prototypes refined with professional tools and human talent. Not a replacement — an accelerator.

But there's a deeper question Project Genie raises: if AI can learn physics, causality, and spatial dynamics without anyone programming them, what else can it learn on its own?

Hassabis is betting the answer is: almost everything. And that's why $47 billion trembled on Wall Street.

Frequently asked questions

Can Project Genie create complete games?

No. It generates navigable environments lasting 60 seconds maximum, with no game mechanics, objectives, or progression. It's an impressive tech demo, not a game engine.

Is it free?

No. It requires a Google AI Ultra subscription ($249.99/month, or $124.99/month with introductory discount). Only available in the US. Free users can only access pre-built demos.

Can it generate games with Nintendo or Disney characters?

Technically yes, though Google has inconsistent safeguards. Bootleg Zelda and Mario worlds have already been generated. No formal legal action from Nintendo yet.

Have gaming stocks recovered?

Partially. By February 3, Unity was up 3%+ and most stocks were rebounding, but none had reached pre-announcement levels.

Is Google trying to compete with Unity or Unreal Engine?

Not directly. For Google, Project Genie is a step toward artificial general intelligence (AGI), not a game development tool.