Anthropic Nears a $7 Billion Deal for Decart — Why AI's Biggest Prize Right Now Is Efficiency

Introduction: The Deal That Moved From $6B to $7B in Three Days

On August 16, 2026, Calcalist reported that Anthropic is closing in on a deal to acquire Israeli AI startup Decart for roughly $7 billion — after beating Nvidia to the table. The price is the headline, but the velocity is the story: talks were first reported at around $6 billion on August 13 by Reuters and Bloomberg, and the number has climbed as competing interest pushed the startup's value higher in a matter of days.

If it closes, this would be Anthropic's largest acquisition ever and one of the biggest AI deals of 2026 — a pre-IPO company buying a three-year-old startup for more than many AI companies raised in entire funding lifecycles. And unlike most AI acquisitions that chase models or talent, this one is fundamentally about something less glamorous: making expensive AI dramatically cheaper to run. We break down what Anthropic is buying, how Nvidia lost a bidding war for its own portfolio company, and what the deal tells you about picking AI tools in 2026.

To see how Decart fits the broader landscape, browse the AI Video Creation tools and AI Design Tools on aitrove.ai.

Who Decart Is — From a Playable Minecraft to a $4B Valuation

Decart, co-founded by CEO Dean Leitersdorf, first turned heads in late 2024 with an AI model that simulated a real-time, playable version of Minecraft — no game engine, just a neural network generating every frame as you moved. It was a viral demo of something serious: neural networks fast enough to generate interactive video rather than static clips.

The company raised $32 million at a $500 million-plus valuation in December 2024, and by mid-2026 it had closed a $300 million round at a valuation approaching $4 billion, with strategic backers including Toyota, Adobe, eBay, and Nvidia. Its customer base leans heavily into e-commerce and livestreaming, and its developer community has grown past 100,000 builders — a distribution asset that is nearly as valuable as the models themselves.

What Anthropic Is Actually Buying: Lucy, Oasis 3, and DOS

Three assets anchor the deal:

• Lucy — Decart's real-time video foundation model, the engine behind its e-commerce and livestreaming business and the platform its developer community builds on.
• Oasis 3 — its interactive world model, unveiled in June 2026, which generates photorealistic driving environments in real time and is aimed at autonomous-vehicle companies that need to simulate rare road scenarios at scale, with robotics and physical AI on the roadmap. It sells via API at $0.02 per second.
• DOS, the Decart Optimization Stack — the quiet crown jewel. DOS is software that optimizes models to run efficiently on Nvidia, Amazon, and Google hardware. Leitersdorf has claimed it makes Decart's models more than an order of magnitude cheaper to run than competitors' — the kind of vertical integration, down to the hardware, that frontier labs normally spend years rebuilding internally.

For Anthropic — a company whose costs are dominated by inference compute across its Claude family — that optimization expertise is the strategic logic. Reports around the deal consistently point to compute efficiency, video generation, and chip-optimization capabilities as what Anthropic is after, not just another frontier model.

How Nvidia Lost Its Own Portfolio Company

The most striking detail in Calcalist's reporting is that Anthropic beat Nvidia to the table — for a startup Nvidia had invested in. Earlier in August, reports had Elon Musk's SpaceX circling Decart as well, and Amazon was named as a potential bidder that could gain from leading the process. A competitive field like that explains how the price climbed from roughly $6 billion to $7 billion in seventy-two hours.

The subtext matters for anyone watching the AI stack: chipmakers, cloud providers, and model labs are all bidding for the same scarce thing — people who can make AI inference radically cheaper. When a GPU vendor, a model lab, and a rocket company all want the same optimization talent, efficiency has stopped being a cost center and become the product.

The Real Prize: Efficiency, Margins, and the Road to IPO

Anthropic is reportedly ramping up pre-IPO dealmaking, and reporting around the Decart talks suggests the company is targeting gross margins around 77% before going public. You cannot get there on raw GPU spending alone. Every point of inference efficiency drops straight to that line, which is why a "boring" optimization stack can justify a $7 billion price tag.

The timing also lands in the middle of the most inference-cost-obsessed week of 2026: DeepSeek rewrote its API pricing with peak and off-peak rates, raising some prices more than 10x, while Google halved Gemini 3.7 Flash's price and OpenAI previewed a premium Ultrafast speed tier. The entire market — from budget APIs to $7 billion acquisitions — is now repricing around one question: what does intelligence cost per second to serve?

What It Means for the AI Tools You Pick

If you are evaluating AI tools right now, the deal carries four practical signals:

• Watch independent real-time video tools closely. If Decart's Lucy and Oasis fold into Anthropic, third-party access could change — pricing, terms, or availability. If your workflow depends on them, confirm roadmap commitments now.
• Simulation is a real product category now. World models like Oasis 3 (and rivals Google Genie 3 and World Labs' Marble) turn "AI video" into a testing ground for AV, robotics, and physical AI. If you build in those spaces, this is a category to track.
• Efficiency is becoming a buying criterion. A tool backed by a strong inference stack can sustain lower prices longer. Vendors that whisper about "rising compute costs" may be telling you about their architecture, not the market.
• Consolidation is accelerating. The standalone champions of 2024–25 are being absorbed into platforms. Diversify your stack so no single acquisition can break your workflow.

The Bottom Line

Anthropic paying roughly $7 billion for Decart — outbidding Nvidia and SpaceX — is 2026's clearest signal that the AI race has shifted from who has the best model to who can serve intelligence cheapest. The smartest models on earth are only as good as the economics behind them, and Decart's optimization stack attacks that math directly. For AI tool buyers, the takeaway is simple: efficiency, sustainability of pricing, and vendor independence are no longer background details. They are the evaluation criteria.

Frequently Asked Questions