QUICK INFO BOX

Introduction

Gaming is constrained by development costs. The average AAA game requires $100-300 million and 3-7 years to build—with 200-500+ developers creating:

1. 3D worlds: Modeling terrain, buildings, objects (millions of polygons)
2. Textures: Creating surfaces, materials, lighting (artists spending weeks per asset)
3. Physics: Programming collision detection, gravity, object interactions
4. Rules: Coding game logic, AI behavior, quest systems
5. Testing: QA teams finding bugs, balance issues—iterating for months

Result: $200 billion gaming industry bottlenecked by creation speed. Games launch with fixed content—once players finish (20-100 hours), they move on. The dream of infinite, procedurally generated worlds (Minecraft’s blocky landscapes, No Man’s Sky’s repetitive planets) remains limited by:

Traditional procedural generation:

• Algorithmic: Rules-based (place trees randomly, generate dungeons)—repetitive, lacks creativity
• Pre-authored: Artists creating tiles, assets—procedurally arranged but finite content
• Not reactive: Worlds don’t respond to player actions dynamically

What if AI could generate game worlds in real-time—responding to every player action, creating unlimited content, rendering photorealistic environments instantly? Not pre-generating assets (slow, storage-heavy), but generating each frame on-the-fly as players explore?

Enter Decart (formerly Etched), the AI research lab building real-time world generation models—diffusion models generating interactive 3D environments at 20-40 frames per second, enabling players to explore infinite AI-generated worlds. Founded in 2023 by Dean Leitersdorf (CEO, ex-IDF cyber intelligence officer) and Oded Ovadia (CTO, ex-Intel AI researcher), Decart demonstrated Oasis—the world’s first playable AI-generated game (Minecraft-style world generated entirely by neural network, zero game engine code).

As of February 2026, Decart operates at a $1.6 billion valuation with $133 million in funding from Sequoia Capital, Index Ventures, Benchmark, NVIDIA, Unity Technologies, and Tencent. The company employs 50+ researchers and engineers (February 2026) across Palo Alto and Tel Aviv, training world foundation models on custom infrastructure (2,000+ NVIDIA H100 GPUs). Decart’s Oasis demo (October 2024) attracted 2 million+ players and 100+ game studios are testing world generation API in private beta—with commercial platform planned for late 2026.

What makes Decart revolutionary:

1. Real-time generation: 20-40 fps interactive worlds (not pre-rendered)—AI generating each frame based on player input
2. No game engine: Oasis uses zero traditional code (no Unity, Unreal)—pure neural network generating physics, graphics, logic
3. Photorealistic potential: Current demo is Minecraft-style (proof of concept), roadmap targets photorealistic AAA quality
4. Infinite content: Worlds never repeat—AI creating unlimited terrain, objects, scenarios on-the-fly
5. Reactive environments: Worlds responding to player actions—breaking blocks, building structures, AI adapting dynamically

The market opportunity spans $200+ billion gaming industry, $50+ billion game development tools, $100+ billion metaverse/virtual worlds, and $500+ billion entertainment. Every game studio seeks faster content creation, infinite replayability, lower costs. Decart provides foundation models enabling studios to generate game worlds procedurally—reducing development time from years to months, creating infinite content, enabling new game genres impossible with traditional development.

Decart competes with NVIDIA Omniverse (virtual world platform), Unity ($7B market cap, game engine), Unreal Engine (Epic Games), Roblox ($18B market cap, user-generated worlds), procedural generation tools (Houdini, SpeedTree), and AI research labs (Google GameNGen, MIT generative games). Decart differentiates through real-time generation (20-40 fps, not pre-rendered), pure AI approach (no game engine code), playable demo (Oasis proving viability), Israeli tech talent (Tel Aviv AI hub), and founder pedigree (IDF cyber intelligence + Intel AI expertise).

The founding story reflects technical audacity: Dean Leitersdorf (IDF cyber intelligence, understanding real-time systems) and Oded Ovadia (Intel AI researcher on edge inference) believed diffusion models (typically 5-10 seconds per image) could run at 20-40 fps (50ms per frame)—1000x faster. After discovering novel architecture combining autoregressive models (fast) with diffusion models (high quality), they founded Decart to build real-time AI worlds—proving it with Oasis, the first neural-network-only game.

This comprehensive article explores Decart’s journey from research vision to the $1.6 billion real-time AI world generation platform revolutionizing gaming.

Founding Story & Background

The Generative AI × Gaming Gap (2022-2023)

By 2023, generative AI transformed creative industries:

Images: Stable Diffusion, Midjourney, DALL-E—generating photorealistic images in seconds
Video: Runway Gen-2, Pika—generating 3-second clips in 30-60 seconds
Text: GPT-4, Claude—generating articles, code, conversations instantly

Yet gaming remained untouched—AAA games still requiring years of development, fixed content, traditional engines.

Why? Fundamental constraint: Real-time requirement

Gaming demands:

• 20-60 fps: 16-50ms per frame (vs. 5-10 seconds for single diffusion image)
• Interactivity: Responding to player input immediately (<50ms latency)
• Consistency: Maintaining object permanence (block placed remains placed)
• Physics: Realistic gravity, collisions, object interactions
• Infinite content: Generating as players explore (not pre-generating entire world)

Diffusion models (Stable Diffusion, DALL-E) too slow:

• 5-10 seconds per image (1024×1024)
• Not interactive: Can’t respond to real-time input
• Not consistent: Each generation independent (objects disappear/change)

Procedural generation (Minecraft, No Man’s Sky) too limited:

• Algorithmic: Rules-based (place trees, generate caves)—repetitive
• Low fidelity: Blocky (Minecraft), repetitive assets (No Man’s Sky)
• Fixed rules: Can’t adapt to player creativity

The challenge: Can AI generate game worlds in real-time?

Dean Leitersdorf: IDF Intelligence to AI

Dean Leitersdorf (co-founder, CEO):

2015-2020: IDF Cyber Intelligence (Unit 8200)

• Role: Officer in elite cyber intelligence unit (Israeli equivalent of NSA)
• Skills: Real-time systems, low-latency processing, pattern recognition
• Insight: Real-time systems require algorithmic efficiency (not just model size)

2020-2023: Transition to AI

• Stanford MS: Computer Science (AI/ML focus)
• Research: Real-time inference, model optimization, edge deployment
• Observation: Diffusion models powerful but too slow for real-time applications

2023: The Insight

• Problem: Diffusion models generate beautiful images but take 5-10 seconds
• Hypothesis: Combining autoregressive models (fast, like GPT) with diffusion (high quality) could achieve real-time generation
• Vision: Gaming as first killer app for real-time generative AI

Oded Ovadia: Intel AI Research

Oded Ovadia (co-founder, CTO):

2017-2023: Intel AI Researcher

• Focus: Edge AI, inference optimization, neural network compression
• Contributions: Techniques for running large models on mobile/edge devices (quantization, pruning, distillation)
• Patents: 5+ patents on efficient inference, low-latency neural networks

Research insight (2022):

• Observation: Autoregressive models (GPT) generate tokens at 50-100ms each—fast enough for real-time
• Problem: Autoregressive models generate text (1D sequences)—not images (2D), not 3D worlds
• Breakthrough: Hybrid architecture—autoregressive for speed, diffusion for quality

2023: Founding Decart

In April 2023, Leitersdorf and Ovadia met at AI conference (CVPR) and discovered shared obsession: real-time AI worlds.

Leitersdorf: “IDF intelligence systems process data in real-time—why can’t AI generate worlds in real-time?”

Ovadia: “I’ve built inference systems hitting 20ms latency at Intel—we can make diffusion models fast enough for gaming.”

Founding thesis: Gaming is ideal application for generative AI—but requires novel architecture achieving 1000x speedup (5 seconds → 5ms per frame).

They founded Decart in Palo Alto (May 2023) with mission:

“Build real-time AI world generation enabling infinite, interactive, photorealistic game environments.”

Why “Decart”? Derivative of Descartes (“I think, therefore I am”)—AI creating worlds through thought, questioning reality (is generated world less real than programmed world?).

Dual-location strategy:

• Palo Alto HQ: Fundraising, partnerships with game studios, US presence
• Tel Aviv R&D: Israeli AI talent (Technion, Tel Aviv University graduates), lower costs (50% of Bay Area), proximity to IDF Unit 8200 alumni network

2023: Seed and Real-Time Architecture Breakthrough

Seed (June 2023): $21 Million

• Lead: Index Ventures
• Additional: Sequoia Capital, Benchmark
• Purpose: Core team (12 researchers), GPU infrastructure (300+ A100s), architecture research

Index Ventures’ lead (partnership with Sequoia, Benchmark) signaled:

• Technical conviction: Believing real-time generation solvable
• Gaming opportunity: $200B+ market ready for disruption
• Israeli talent: Index’s deep Israel network (Wix, monday.com investments)

Architecture breakthrough (September 2023):

Problem: Diffusion models require 20-50 denoising steps (5-10 seconds)

Decart’s solution: Autoregressive-Diffusion Hybrid (ADH)

# Simplified Decart architecture

class RealTimeWorldModel:
    def __init__(self):
        # Fast autoregressive model (predicts next latent token)
        self.autoregressive = GPT-like transformer
        
        # High-quality diffusion upsampler (refines latents → pixels)
        self.diffusion = Fast diffusion model (4-8 steps)
    
    def generate_frame(self, previous_frame, player_action):
        """
        Generate next frame in <50ms.
        """
        # Step 1: Autoregressive prediction (fast, <10ms)
        # Predict next frame's latent representation
        latent = self.autoregressive.predict(
            previous_latents=previous_frame.latent,
            action=player_action  # W/A/S/D, mouse movement
        )
        
        # Step 2: Diffusion upsampling (quality, ~40ms)
        # Convert latent → high-quality pixels
        pixels = self.diffusion.upsample(
            latent=latent,
            num_steps=4  # Fast (vs. 20-50 standard)
        )
        
        return pixels
    
    # Total latency: 10ms + 40ms = 50ms (20 fps)
    # With optimization: 25ms (40 fps)

Key innovations:

1. Latent-space autoregressive: Predicting compressed representations (not pixels)—10x faster
2. Fast diffusion: 4-8 steps (vs. 20-50)—using rectified flow (like FLUX.1)
3. Temporal consistency: Conditioning on previous frames—maintaining object permanence
4. Action conditioning: Neural network understanding player input (keyboard, mouse)

Benchmarks (December 2023):

• Latency: 50ms per frame (20 fps)
• Quality: Minecraft-level fidelity (blocky but playable)
• Consistency: 85% object permanence (blocks placed stay placed)

2024: Oasis Demo Goes Viral

January-October 2024: Decart trained world model on:

• Data: 1M+ hours of Minecraft gameplay (videos + player actions)
• Compute: 1,000+ A100 GPUs, 3 months training ($2M+ compute cost)
• Architecture: Scaling ADH to 3B parameters

Oasis launch (October 31, 2024)—Halloween surprise:

Demo: oasis.decart.ai—fully playable Minecraft-style game generated by AI

Features:

• Zero game engine code: No Unity, Unreal, custom C++—pure neural network
• Real-time: 20 fps (50ms per frame) on NVIDIA H100
• Interactive: Players move, break blocks, place blocks—AI responding instantly
• Infinite world: Generating terrain as players explore—no pre-generated map
• Physics: Blocks fall (gravity), water flows, sun moves—learned from data (not programmed)

Viral impact:

• 2 million+ players first week
• 500K+ concurrent peak
• Twitter explosion: 50M+ views, trending #1, Elon Musk tweeting (“The future of gaming”)
• Media coverage: NYT, WSJ, TechCrunch, Verge—“AI-generated Minecraft” headline

Technical achievement: First playable game with zero traditional code—proving real-time AI worlds viable.

2024: Series A and Studio Partnerships

Series A (November 2024): $112 Million

• Lead: Sequoia Capital
• Additional: Index Ventures, Benchmark, Conviction Capital, NVIDIA, Unity Technologies, Tencent
• Valuation: $1.6 Billion (unicorn status, 18 months after founding)
• Purpose: Photorealistic models, 2,000+ H100 GPUs, team expansion (12 → 40), game studio partnerships

NVIDIA’s investment provided:

• GPU allocation: Priority H100 access (scarce)
• Technical collaboration: Optimizing CUDA kernels, inference latency
• Go-to-market: Joint customers, NVIDIA Omniverse integration

Unity Technologies investment signaled:

• Strategic validation: Unity (game engine leader) recognizing AI world generation as future
• Integration: Unity exploring Decart integration (AI-generated levels in Unity games)
• Existential hedge: Unity hedging against AI replacing traditional engines

Tencent (Chinese gaming giant) provided:

• Distribution: Access to 600M+ Chinese gamers
• Use cases: Mobile gaming (Decart targeting mobile deployment)
• Strategic: China seeking sovereign AI gaming capabilities

Studio partnerships (December 2024 – Present):

• 100+ game studios in private beta (indie, mid-size, AAA)
• Use cases: AI-generated levels (unlimited content), prototype/concept art (rapid iteration), NPC behavior (AI-driven characters)

By February 2026:

• 50+ employees (30 research/engineering, 15 product/partnerships, 5 ops)
• Oasis v2: 30 fps, higher resolution (512×512 → 768×768)
• Private API: 100+ studios generating worlds, testing game mechanics

Founders & Key Team

Dean Leitersdorf (CEO) brings real-time systems expertise from IDF cyber intelligence—understanding low-latency requirements, scalable architectures. His Stanford AI research provides technical depth, while IDF leadership experience informs go-to-market strategy.

Oded Ovadia (CTO) provides inference optimization mastery from Intel—techniques for running large models efficiently (quantization, pruning, distillation). His edge AI work directly enables Decart’s real-time generation (50ms latency impossible without aggressive optimization).

Assaf Shocher (VP Research) brings 3D vision expertise from Google Research—NeRF (neural radiance fields), 3D reconstruction, novel view synthesis. His research on representing 3D worlds as neural networks informs Decart’s architecture.

Daniel Cohen-Or (Chief Scientist) is computer graphics legend—Tel Aviv University professor, 500+ papers on 3D geometry, shape analysis, graphics. His involvement provides academic credibility, theoretical foundations, PhD student recruiting pipeline.

Funding & Investors

Seed (June 2023): $21 Million

• Lead Investor: Index Ventures
• Additional Investors: Sequoia Capital, Benchmark
• Purpose: Core team, GPU infrastructure (300+ A100s), real-time architecture research

Series A (November 2024): $112 Million

• Lead Investor: Sequoia Capital
• Additional Investors: Index Ventures, Benchmark, Conviction Capital, NVIDIA, Unity Technologies, Tencent
• Valuation: $1.6 Billion (unicorn status)
• Purpose: Photorealistic world models, 2,000+ H100 GPUs, team expansion (12 → 50), game studio partnerships, commercial platform

Total Funding Raised: $133 Million

Decart deployed capital across:

• Compute infrastructure: $40-60M in H100/A100 GPUs (2,000+ GPUs for training, 500+ for inference)
• Research talent: $30-50M in compensation (researchers from Google, NVIDIA, Intel, IDF Unit 8200—competitive Israeli + Bay Area salaries)
• Training data: $10-20M licensing gameplay data (Minecraft, AAA games, synthetic generation)
• Engineering: $15-25M building inference infrastructure, game studio API, partnerships
• Operations: $10-20M dual offices (Palo Alto, Tel Aviv), Israeli R&D operations

Product & Technology Journey

A. Oasis (Playable Demo)

World’s first AI-generated game (October 2024):

Technical specs:

• Resolution: 360×640 (mobile-friendly)
• Frame rate: 20 fps (50ms latency)
• Model size: 3B parameters
• Controls: WASD (movement), mouse (look), left-click (break), right-click (place)
• World: Infinite terrain generation (forests, caves, mountains, water)

Architecture:

# Oasis model (simplified)

class OasisWorldModel:
    def __init__(self):
        # Vision encoder: Current frame → latent representation
        self.encoder = CNN (ResNet-style)
        
        # Autoregressive dynamics: Predict next latent from current + action
        self.dynamics = Transformer (GPT-style, 3B parameters)
        
        # Diffusion decoder: Latent → pixels
        self.decoder = Fast diffusion (4 steps, rectified flow)
    
    def step(self, current_frame, player_action):
        """
        Generate next frame (50ms total).
        """
        # Encode current frame (5ms)
        current_latent = self.encoder(current_frame)
        
        # Predict next latent using action (15ms)
        next_latent = self.dynamics.forward(
            latent=current_latent,
            action=player_action,  # e.g., {"forward": 1, "mouse_x": 0.1}
            previous_latents=self.latent_history[-16:]  # Temporal context
        )
        
        # Decode to pixels (30ms)
        next_frame = self.decoder.generate(
            latent=next_latent,
            steps=4
        )
        
        self.latent_history.append(next_latent)
        return next_frame

What Oasis learned from data (not programmed):

• Physics: Blocks fall when unsupported (gravity)
• Water: Flows downhill, spreads horizontally
• Day/night cycle: Sun rises, sets, moon appears
• Inventory: Breaking blocks adds to inventory (though imperfect—sometimes glitches)
• Crafting: Placing blocks constructs structures

Limitations (v1, October 2024):

• Consistency: 15% of time, objects flicker/disappear (temporal consistency imperfect)
• Inventory bugs: Sometimes blocks placed don’t match inventory
• Resolution: 360×640 (low by modern standards)
• Complexity: Minecraft-style blocks only (not photorealistic)

B. Oasis v2 (February 2026)

Improvements:

• 30 fps: 33ms latency (optimized inference)
• 768×768 resolution: 4x more pixels
• Better consistency: 95%+ object permanence
• Richer worlds: More biomes (desert, snow, jungle), weather effects
• Physics: More realistic (proper collision detection, momentum)

Still limitations:

• Not photorealistic: Minecraft-style graphics (proving real-time viability, not yet AAA quality)
• Single-player: No multiplayer (synchronizing multiple players’ AI worlds unsolved)
• No complex mechanics: No redstone (Minecraft logic circuits), no entities (animals, mobs)

C. World Foundation Models API (Private Beta)

For game studios:

# Decart World API (conceptual)

import decart

# Initialize world generator
world = decart.World(
    style="fantasy_forest",  # or "cyberpunk_city", "desert_wasteland"
    resolution=(1024, 1024),
    fps=30
)

# Generate initial frame
frame = world.generate_start()

# Game loop
while True:
    # Get player input
    action = get_player_input()  # WASD, mouse, etc.
    
    # Generate next frame
    frame = world.step(action)
    
    # Render to screen
    display(frame)

Use cases (studios testing):

1. Procedural levels: Generating infinite dungeons, open worlds—replacing manual level design
2. Rapid prototyping: Designers describing levels (“dark forest with ruins”), AI generating instantly—faster iteration
3. Dynamic events: Worlds responding to player actions (burning forest spreads, floods submerge areas)
4. Content expansion: Existing games adding AI-generated DLC (new areas, quests)

Pricing (planned):

• Compute-based: $0.001-0.01 per frame (20fps game = $1.20-12/hour gameplay)
• Subscription: $1K-10K/month (unlimited generation for development/testing)
• Licensing: $100K-1M/year (shipping games with Decart-generated worlds)

D. Roadmap to Photorealism

Current (February 2026): Minecraft-style blocks (proof of concept)

2026 goals: Semi-realistic graphics (Fortnite-level quality)

• Training data: Unreal Engine 5 game footage (licensed from studios)
• Resolution: 1080p (1920×1080)
• Physics: Advanced (cloth simulation, particle effects, destructible environments)

2027+ vision: Photorealistic (The Last of Us, Cyberpunk 2077 quality)

• Resolution: 4K (3840×2160)
• Ray tracing: Realistic lighting, reflections, shadows
• NPC AI: Characters with natural behavior, conversations
• Multiplayer: Synchronizing worlds across players

Challenges:

• Compute cost: Photorealistic 4K at 60fps = 100x more compute than current Oasis
• Training data: Need 100M+ hours of AAA game footage (expensive licensing)
• Consistency: Harder to maintain at higher fidelity (more details = more can go wrong)

Business Model & Revenue

Revenue Model (Future)

Not yet monetizing (February 2026). Planned model:

Target launch: Q4 2026 (API general availability)

Target Customers

1. Indie game studios (40%): 5-20 person teams, rapid prototyping, procedural content
2. Mid-size studios (35%): 50-200 person teams, content expansion, DLC generation
3. AAA studios (15%): 500-2,000 person teams, prototype/concept art, experimental titles
4. Metaverse/UGC (10%): Roblox, Fortnite Creative—user-generated AI worlds

Estimated Economics (Post-Launch)

• CAC: $5K-50K (enterprise sales, 6-12 month cycles, partnerships with Unity/Unreal)
• LTV: $500K-10M+ (multi-year contracts, expanding usage)
• Gross Margin: 40-50% (high compute costs—H100 inference expensive)
• Projected 2027 ARR: $50-100M (100-500 studio customers, 10-50 major licenses)

Competitive Landscape

NVIDIA Omniverse ($40B+ NVIDIA market cap): Virtual world platform, physics simulation, collaboration—not real-time AI generation
Unity ($7B market cap): Game engine leader—exploring AI tools, potential integration
Unreal Engine (Epic Games): AAA engine standard—procedural generation tools (PCG framework)
Roblox ($18B market cap): User-generated content—not AI-generated
Procedural tools (Houdini, SpeedTree, World Creator): Artist tools—not AI, not real-time
Google GameNGen (research): AI-generated Doom—10fps, research only

Decart Differentiation:

1. Real-time: 20-40 fps interactive generation (competitors pre-render or research demos)
2. Playable demo: Oasis proven with 2M+ players (not just paper/video)
3. Pure AI: Zero game engine code (competitors hybrid approaches)
4. Israeli talent: Tel Aviv R&D (Unit 8200 network, Technion/TAU graduates)
5. Founder pedigree: IDF cyber + Intel AI (real-time systems expertise)

Impact & Success Stories

Indie Studio

Rogue-like developer: Using Decart API to generate infinite dungeons—each playthrough unique. Result: 10x content variety (vs. hand-crafted levels), 3x player retention (never same twice), shipped game 6 months faster (reduced level design from 12 months to 2 months prototyping + AI generation).

AAA Studio (Confidential)

Open-world game: Using Decart for rapid prototyping—designers describing areas (“haunted swamp with abandoned village”), Decart generating within minutes. Result: 50x faster iteration (vs. weeks for artists to build), 20 concept environments tested (vs. 3-5 traditionally), team alignment on world design 4 months earlier.

Metaverse Platform

VR world builder: Integrating Decart so users can describe worlds (“tropical island with volcano”), platform generates instantly. Result: 100x more user-created worlds (vs. manual building tools), 5x engagement (users spending time exploring vs. building).

Future Outlook

Product Roadmap

2026: API general availability (Q4), Fortnite-quality graphics, 60 fps, 1080p resolution
2027: Photorealistic graphics (The Last of Us quality), multiplayer synchronization, 4K resolution
2028: Full game generation (AI creating mechanics, quests, stories—not just worlds), mobile deployment

Growth Strategy

Game engine partnerships: Unity, Unreal integrations—AI levels in traditional pipelines
Platform plays: Roblox, Fortnite Creative, VRChat—enabling users to generate AI worlds
Consumer product: Direct-to-player platform (play AI-generated games, no studio needed)

Long-term Vision

Decart aims to democratize game creation—enabling anyone to generate AAA-quality games through text descriptions (“Create cyberpunk open-world RPG with heist mechanics”). With $133M funding, $1.6B valuation, viral Oasis demo (2M+ players), and 100+ studios testing API, Decart positioned for IPO ($10B-20B+ valuation) or strategic acquisition (Unity, Epic, Microsoft) within 5-10 years as AI-generated games become standard—potentially disrupting $200B gaming industry and $50B game development tools market.

FAQs

What is Decart?

Decart builds real-time AI world generation models—diffusion models generating interactive 3D game environments at 20-40 fps. Creator of Oasis, world’s first playable AI-generated game (Minecraft-style world with zero game engine code).

How much funding has Decart raised?

$133 million total across Seed ($21M, led by Index Ventures) and Series A ($112M, led by Sequoia with NVIDIA, Unity, Tencent), achieving $1.6 billion valuation (November 2024)—unicorn within 18 months.

Who founded Decart?

Dean Leitersdorf (ex-IDF Unit 8200 cyber intelligence, Stanford CS) and Oded Ovadia (ex-Intel AI researcher, inference optimization expert), founded 2023 in Palo Alto with R&D in Tel Aviv, Israel.

What is Oasis?

Oasis is world’s first playable AI-generated game—Minecraft-style world running at 20 fps with zero traditional game engine code. Pure neural network generates each frame based on player input. 2 million+ players tried demo (October 2024).

How does real-time generation work?

Decart uses autoregressive-diffusion hybrid (ADH)—fast autoregressive model predicts next frame’s latent representation (<10ms), diffusion model upsamples to pixels (~40ms), total 50ms (20 fps). Optimizations achieve 33ms (30 fps).

Conclusion

Decart has established itself as pioneering real-time AI world generation platform, achieving $1.6 billion valuation, $133 million funding from Sequoia/NVIDIA/Unity/Tencent, and 2 million+ players experiencing Oasis—world’s first playable AI-generated game proving real-time generation viable. With novel autoregressive-diffusion hybrid architecture achieving 20-40 fps interactive worlds (1000x speedup from typical diffusion models), zero game engine code approach, and 100+ game studios testing world generation API, Decart demonstrates that AI can generate infinite game content on-the-fly—not pre-rendered assets.

As gaming industry faces rising development costs ($100-300M per AAA game, 3-7 years), demand for procedural content generation grows exponentially—studios seeking infinite replayability, faster iteration, lower costs, new game genres impossible with traditional development. Decart’s real-time generation (enabling infinite worlds), pure AI approach (no engine code bottleneck), playable demo validation (2M+ players), Israeli technical talent (IDF Unit 8200, Technion alumni), and strategic partnerships (Unity, Tencent distribution) position it as transformative gaming infrastructure. With roadmap toward photorealistic 4K generation, multiplayer synchronization, and full game generation (mechanics, quests, stories), Decart is positioned as compelling IPO candidate ($10B-20B+ valuation) or strategic acquisition target within 5-10 years as AI-generated games transition from research curiosity to mainstream entertainment, potentially revolutionizing $200B gaming industry and enabling new era of infinite, personalized, dynamically generated interactive experiences.

Related Article:

• https://eboona.com/ai-unicorn/6sense/
• https://eboona.com/ai-unicorn/abnormal-security/
• https://eboona.com/ai-unicorn/abridge/
• https://eboona.com/ai-unicorn/adept-ai/
• https://eboona.com/ai-unicorn/anduril-industries/
• https://eboona.com/ai-unicorn/anthropic/
• https://eboona.com/ai-unicorn/anysphere/
• https://eboona.com/ai-unicorn/applied-intuition/
• https://eboona.com/ai-unicorn/attentive/
• https://eboona.com/ai-unicorn/automation-anywhere/
• https://eboona.com/ai-unicorn/biosplice/
• https://eboona.com/ai-unicorn/black-forest-labs/
• https://eboona.com/ai-unicorn/brex/
• https://eboona.com/ai-unicorn/bytedance/
• https://eboona.com/ai-unicorn/canva/
• https://eboona.com/ai-unicorn/celonis/
• https://eboona.com/ai-unicorn/cerebras-systems/