Decart
Decart enables real-time video generation and inference at 10x lower cost.
Decart is an AI research lab building the fastest and most efficient generative AI models for real-time video generation and manipulation. Their flagship products—Oasis (real-time AI game world) and Mirage (live video editing via text prompts)—achieve 10x better efficiency than competitors like OpenAI's Sora, reducing inference costs from $100/hour to $0.25/hour. The company combines a proprietary LLM inference engine (built from scratch in C++ and CUDA) with advanced video generation capabilities, serving developers, enterprises, and streaming/gaming platforms.
Problem solved
AI model inference and video generation are prohibitively expensive and slow, limiting real-time applications in gaming, streaming, and interactive experiences.
Target customer
AI developers, enterprise applications requiring real-time video processing, streaming platforms, gaming companies, and infrastructure teams optimizing GPU workloads.
Founders
D
Dean Leitersdorf
CEO & Co-Founder
Youngest PhD recipient from Technion at 23 (in computer science), completed high school in 2 years, Unit 8200 veteran (Israel's elite military intelligence unit), postdoc in Singapore.
M
Moshe Shalev
CPO & Co-Founder
Former cyber operator and Unit 8200 veteran from Israel's ultra-Orthodox community; brings grounded business mindset and operational expertise to complement Leitersdorf's technical vision.
U
Undisclosed Third Co-Founder
Co-Founder
Equally impressive background but name withheld due to existing commitments.
Funding history
Seed
$21M
October 31, 2024
Led by Sequoia Capital
· Zeev Ventures
Series A
$32M
December 2024
Led by Benchmark
· Sequoia Capital, Zeev Ventures
Series B
$100M
August 7, 2025
Led by Sequoia Capital, Benchmark, Zeev Ventures
· Aleph VC
Total raised:
$153M
Pricing
Free API tokens for developers to test real-time video generation; usage-based pricing that scales with demand; enterprise-tier solutions available via direct contact. Partnership and subscription models offered for Oasis with tiered pricing based on implementation scale.
Notable customers
Not publicly disclosed; company reports revenue in millions and profitability at launch. Notable partners: Cerebrium, Comcast, ElevenLabs, Backblaze.
Integrations
Cerebrium (LLM inference), Comcast (edge deployment), ElevenLabs (avatar lip-sync), Backblaze B2 Overdrive (storage), Nvidia GPUs
Tech stack
DoubleClick Floodlight (Advertising)
Google Tag Manager (Tag managers)
Vercel (PaaS)
CookieYes (Cookie compliance)
Website
Competitors
OpenAI Sora
Decart achieves 10x better efficiency and real-time performance; Sora focuses on high-quality offline video generation.
Stability AI
Decart specializes in real-time video generation with proprietary inference optimization; Stability AI has broader generative AI focus.
LangChain
Decart is infrastructure-focused on efficient inference and video; LangChain emphasizes LLM application frameworks.
Fetch.ai
Decart targets real-time video and inference efficiency; Fetch.ai focuses on autonomous agent infrastructure.
Why this matters: Decart represents a rare combination of foundational AI research rigor and immediate commercial traction—emerging from stealth with $21M in seed funding, 1M users in 72 hours, and profitability at launch. The team's obsessive focus on efficiency (10x cost reduction, sub-100ms latency) mirrors early Google's approach and unlocks entirely new use cases (real-time game worlds, live video editing) previously impossible at scale.
Best for: Companies building real-time interactive experiences—gaming, streaming, live video editing, and digital avatars—that need sub-100ms latency and drastically reduced inference costs.
Use cases
Real-Time Game World Generation
Decart's Oasis enables fully playable AI-generated game worlds that render in real time, reaching 1M users in 72 hours at launch. Developers can create persistent, interactive worlds without pre-rendering massive assets, reducing development time and server costs.
Live Video Stream Editing
Mirage and LSD allow streamers and content creators to modify live video feeds with text prompts in sub-100ms latency. A streamer can change virtual backgrounds, apply effects, or alter avatars on the fly without disrupting broadcast, enabling new creative possibilities for gaming and entertainment.
Cost-Optimized Inference at Scale
Decart's proprietary inference engine reduces LLM inference costs from $100/hour to $0.25/hour. Enterprises running high-volume AI workloads (e.g., content moderation, document processing) can deploy models on commodity hardware with 10x efficiency gains, significantly improving unit economics.
Edge-Deployed Generative AI
Through partnerships like Comcast, Decart enables generative AI inference at the network edge with sub-35ms latency. ISPs and telecom operators can offer AI-powered services (video enhancement, real-time personalization) to millions of users without centralized cloud bottlenecks.
Surgical Video Editing
Lucy brings frame-by-frame video editing control to medical/surgical contexts with real-time performance. Surgeons and medical teams can enhance or annotate live surgical feeds during procedures or training without latency, improving visibility and education.
Alternatives
OpenAI Sora
Sora offers superior visual quality and longer-form generation but lacks real-time capabilities and costs significantly more per inference; choose Sora if video quality is paramount over latency and cost.
Runway Gen-3
Runway focuses on creator-friendly video editing with broader model flexibility but doesn't optimize for real-time inference or extreme cost reduction; choose Runway for polished, interactive creative workflows.
Stability AI Video
Stability AI's video models are accessible but less optimized for real-time use cases and edge deployment; choose Stability for cost-conscious teams without hard latency requirements.
FAQ
What does Decart do? +
Decart is an AI research lab building ultra-efficient generative AI models for real-time video generation and inference. Its flagship products—Oasis (AI game worlds), Mirage (live video editing), and Lucy (surgical video editing)—deliver sub-100ms latency and 10x better efficiency than competitors, powered by a proprietary C++ and CUDA inference engine that reduces costs dramatically.
How much does Decart cost? +
Decart offers free API tokens for developers to test, with usage-based pricing that scales with demand. Enterprise solutions require direct contact. Partnership models with tiered pricing are available for Oasis deployment based on implementation scale. Typical inference costs run at $0.25/hour versus $100/hour for traditional approaches.
What are alternatives to Decart? +
OpenAI Sora (higher quality, offline, expensive), Runway Gen-3 (creator-focused, flexible models), Stability AI Video (cost-accessible, less optimized for real-time). Choose alternatives if you prioritize video quality over latency or need broader creative editing tools rather than infrastructure efficiency.
Who uses Decart? +
Target customers include AI developers, gaming studios, streaming platforms, enterprises optimizing inference costs, and medical/surgical teams. Specific customer names are undisclosed, but the company reports multi-million revenue and profitability at launch, with notable partnerships including Comcast, ElevenLabs, and Cerebrium.
How does Decart compare to OpenAI Sora? +
Decart achieves 10x better efficiency and real-time performance (sub-100ms latency) compared to Sora's offline approach, making it ideal for interactive and live applications. Sora excels at high-quality, long-form video generation offline. Choose Decart for speed and cost; choose Sora for visual fidelity and longer sequences.
What makes Decart different? +
Decart built a proprietary inference engine from scratch in C++ and CUDA that outperforms existing engines (vLLM, TGI) by orders of magnitude. The founding team combines elite technical talent (youngest Technion PhD recipients, Unit 8200 veterans) with obsessive focus on efficiency, reducing real-time video generation to commodity-hardware feasibility.
Tags
real-time video generation
inference optimization
generative AI
GPU efficiency
edge computing
video synthesis
low-latency AI