Shaga Labs, Inc.

Building the DePIN network where today's gameplay powers tomorrow's AI-generated worlds

[Shaga Labs, Inc. @ShagaLabs]

World models need authentic gameplay to train.

Enterprise supply doesn’t exist.

Our peer-to-peer cloud gaming infrastructure produces this data at scale.

Every move and decision trains world models to generate tomorrow's games at the speed of a prompt.

Shaga is the source.

[Shaga Labs, Inc. @ShagaLabs]

AI learned from libraries.
Now it's learning from lobbies.

[Shaga Labs, Inc. @ShagaLabs]

Kill → Panic fire → Weapon switch → Miss → Close the distance → Try again.

Synthetic gameplay can mimic the sequence. Not the uncertainty behind it.

That's what we're capturing.

[Shaga Labs, Inc. @ShagaLabs]

Games began as code. Then culture.

Next, they'll be imagined by you.

[Shaga Foundation @ShagaFDN]

Shaga Arcade just hit 1 month.

The first GAMENET REWARDS are going out.

To every host who kept their rig running.

Every gamer who showed up.

This is what you're building. ↓

[Shaga Labs, Inc. @ShagaLabs]

We’re building around this shift:
- Cloud gaming for players
- Authenticated gameplay capture
- Node operators can earn
- AI labs access high‑signal, compliance‑ready data
- Players just play

[Shaga Labs, Inc. @ShagaLabs]

Most of today’s AI is optimized for understanding stories.

World models will demand we understand behavior.

[Shaga Labs, Inc. @ShagaLabs]

In matches like this, the real game is reading other players: position, intent, risk.

How players respond to each other is what makes this kind of play valuable.

It’s the difference between AI that sees a level and AI that understands a match.

[Shaga 🌐🎮 @joinplayroll]

Those Battlefield 6 players: “Where did he come from??”

Me: from the clouds, brother.

Literally playing on Shaga with 60 FPS, 2ms latency and 0% network drop.

Two in a row with the bazooka.
They never stood a chance. 😂

PC gaming on mobile gaming is officially inevitable.

[Shaga Labs, Inc. @ShagaLabs]

Prediction is moving up a layer.

Not pixels, but conditions, actions, and what happens next.

Robotics is proving it.

Gameplay has the same structure underneath - and orders of magnitude more of it.

[RoboPapers @RoboPapers]

Reasoning over long horizons would allow robots to generalize better to unseen environments and settings zero-shot. One mechanism for this kind of reasoning would be world models, but traditional video world models still tend to struggle with long horizons, and are very data intensive to train. But what if instead of predicting images about the future, we predicted just the symbolic information necessary for reasoning?

@nishanthkumar23 tells us about Pixels to Predicates, a method for symbol grounding which allows a VLM to plan sequences of robot skills to achieve unseen goals in previously unseen settings.

To find out more, watch episode #44 of RoboPapers with @micoolcho and @chris_j_paxton now!

[Shaga Labs, Inc. @ShagaLabs]

Why does input synchronization matter for world model training?

Human decisions are temporal, not just visual: when they moved, why they waited, what led to each choice.

World models won’t be built on pixels alone, but on this layer of timed, human decisions.

[Shaga Labs, Inc. @ShagaLabs]

Every serious AI company eventually builds its own data rail + world‑model stack.

Tesla is doing it with robots and factory video.

Interactive agents will need the same thing for gameplay: authenticated trajectories of how humans actually move, explore, and decide in complex environments.

[TeslaZoa @TeslaZoa]

Tesla’s humanoid robots are being trained on Tesla’s own neural network, world model, and generative model.

[Shaga Labs, Inc. @ShagaLabs]

Fei-Fei Li drops a manifesto on spatial intelligence.

World Labs ships Marble a week later.

Yann LeCun (godfather of deep learning) reportedly leaves Meta to chase world models.

Gemini 3 launches with "spatial reasoning" as the headline feature.

November 2025 might be the month we look back on as the pivot point.

[Shaga Labs, Inc. @ShagaLabs]

"map-readers vs explorers" is the right frame.

World models let AI explore reality instead of just reading about it.

Generating the worlds is getting solved. Capturing how humans behave in them is still the bottleneck.

[David Pantera @davidpantera_]

We're not talking about world models enough. We're obsessed LLMs, but LLMs understand text, not reality. They are map-readers, not explorers.

A robot arm doesn't have to break 1,000 actual cups to learn how to hold one. It simulates the grip, detects the slip, and adjusts its force in its internal world model first. It learns in milliseconds, safely.

We don't just ask an AI to hypothesize based on what its been trained on; we use a world model to simulate how a theoretical drug molecule binds to a virus, accelerating discovery from years to minutes.

This video is some clips of my favorite worlds I've been making in @theworldlabs , who just opened up their private beta to the public this week. In the private beta, I spent hours spinning up worlds either from prompts or via uploading images (like this one of Roble gym @ Stanford).

Some of the worlds here are public creations from other users, since you can post your worlds to a feed for others to explore in (it's a pretty fun social product, walking through worlds from other ppl's imaginations).

[Shaga Labs, Inc. @ShagaLabs]

https://t.co/pmSzbVvDsK

[Shaga Labs, Inc. @ShagaLabs]

Enterprise procurement cycles: 6-12 months

Infrastructure development precedes buyers

Google Genie 3: August 2025
NVIDIA Cosmos: January 2025

The platforms building supply now position ahead of enterprise contracting 👀

[Shaga Labs, Inc. @ShagaLabs]

https://t.co/5N1venPzRt

[Shaga Labs, Inc. @ShagaLabs]

World models need tens of thousands of hours of authenticated gameplay.

Largest public dataset: 500 hours.

Synthetic data fails on real human unpredictability.

Major platforms can't retrofit data capture without breaking user trust.

Shaga built for this.

[Shaga Labs, Inc. @ShagaLabs]

Reddit sold text data for $130M annually. News Corp got $250M for news.

Authenticated gameplay data? Zero openly licensed supply at enterprise scale.

AI labs need it, but:

Enterprise budgets exist
Infrastructure doesn't

Shaga is building supply before enterprise buyers arrive.

[Shaga Labs, Inc. @ShagaLabs]

New system adapts gameplay to hardware capabilities across the network.

By benchmarking and classifying nodes into performance tiers, we optimize for both accessibility and data quality. Broader hardware participation without sacrificing the fidelity AI training applications require.

The challenge most distributed networks face: scale vs quality.

Optimize for scale? You get more nodes but inconsistent output.
Optimize for quality? You exclude hardware that could participate.

Shaga's dynamic optimization solves both:

Hardware diversity scales the network
Performance tiers maintain data value

A gaming PC in rural Montana and a high-end rig in Manhattan can both contribute, each validated for different workload types.

This is how you build edge infrastructure that serves both gamers and enterprise AI buyers simultaneously.

[Shaga Labs, Inc. @ShagaLabs]

Infrastructure update: Shaga can capture and validate spatio-temporal gameplay data across the distributed network.

World models and embodied AI systems need large-scale interaction data that captures not just what happens, but also how events unfold across space and time.

The bottleneck has been data scarcity at enterprise scale.

The validation layer we built ensures data integrity across nodes, proving that gameplay data meets quality requirements for training advanced AI systems.

This creates a new data asset class at scale:

Authenticated, validated gameplay data with cryptographic proof of origin.

Most platforms can only collect gameplay video.

Shaga can license it for enterprise AI buyers.

The difference matters when AI labs need provenance guarantees for training datasets worth millions.

[Shaga Labs, Inc. @ShagaLabs]

Three documented market gaps are converging:

Enterprise AI labs need authenticated gameplay data.

No openly licensed supply exists at scale as of Q3 2025.

DePIN networks evolved from infrastructure-first to revenue-first models.
Single-revenue platforms face sustainability challenges.

AI infrastructure constraints push labs toward distributed systems.

McKinsey reports only 19% of executives see 5%+ revenue gains from AI
BCG shows 74% struggle to scale due to infrastructure bottlenecks

Shaga's positioning addresses all three simultaneously:

Dual revenue (B2C gaming + B2B data licensing) creates sustainable unit economics
Community hardware provides capital-efficient edge compute
Authenticated data capture serves enterprise AI demand

The platform serves gamers today while building infrastructure for the AI economy tomorrow.

Which gap represents the biggest market opportunity?

[Shaga Labs, Inc. @ShagaLabs]

DePIN specialists proved individual layers work:

• AI training (Bittensor)
• Edge compute (Theta)
• Data generation (IoTeX)

Shaga integrates all three with consumer revenue funding the infrastructure.

Specialists excel at one layer. Few can bridge the full stack.