Anthropic's Biggest Acquisition Ever Has Nothing to Do With Chatbots

Anthropic has spent the last two years building its reputation on Claude: a chatbot and coding assistant that reads and writes text better than almost anything else on the market. So it's a little jarring that the biggest acquisition in the company's history, according to a Bloomberg report cited across multiple outlets on August 12 and 13, has almost nothing to do with language at all.

The target is Decart AI, a three-year-old Israeli startup that builds "world models" — AI systems trained not on text, but on real-world video, so they can understand and generate physics, motion, and space in real time. The reported price: roughly $6 billion. Neither company has confirmed the talks, and multiple outlets note the deal isn't finalized and could still fall apart. But if it closes, it would be Anthropic's largest acquisition to date, arriving just as the company gears up for what's expected to be one of the biggest IPOs in tech history.

From a Minecraft Demo to a $6 Billion Target

Decart was founded in 2023 by brothers Dean and Orian Leitersdorf and fellow engineer Moshe Shalev — Dean serves as CEO. The company's breakout moment came in October 2024, when it released Oasis, a demo built with chip startup Etched that generated a playable, Minecraft-style world in real time, frame by frame, with no traditional game engine underneath. It was a viral proof of concept: an AI model that could simulate a physical space you could walk through, purely by predicting what the next video frame should look like.

That underlying capability has since split into a small product line. Oasis has evolved into a world model aimed at robotics and autonomous-vehicle training, where simulating physics cheaply and at scale is a real bottleneck. Lucy, Decart's other flagship model, generates real-time video overlays — the company's pitch use case is virtual try-on for fashion e-commerce, letting a live video feed show a person wearing clothes or accessories that aren't actually there. And underneath both sits DOS, the Decart Optimization Stack, a set of tools for squeezing more inference and training performance out of existing chips — a skill that, in an industry burning through GPU budgets, is worth almost as much as the models themselves.

The Valuation Curve Is the Real Story

What makes the numbers striking isn't just the acquisition price — it's how fast Decart got there. The company was valued at roughly $3.1 billion in August 2025. By May 2026, a $300 million round led by Radical Ventures — with participation from Nvidia, Sequoia Capital, Benchmark, Adobe Ventures, Atreides Management, Valor Equity Partners, and Zeev Ventures — pushed that to roughly $4 billion. The reported $6 billion acquisition price, just three months later, represents another 50% jump on top of that.

For a company that was worth a third as much a year ago, that's not normal AI-market froth — it's a sign that multimodal "world model" technology has quickly become one of the most contested pieces of real estate in AI, alongside frontier language models and inference chips.

Why Anthropic, and Why Now

Anthropic's rationale looks like it runs in two directions at once. The obvious one is capability: Claude is a text-and-code specialist, and Decart's video and world-model work would give Anthropic a real answer to Google's Gemini and OpenAI's video efforts, both of which have been pushing hard into multimodal territory. The less obvious one is efficiency. DOS, Decart's chip-optimization stack, addresses a problem Anthropic has been throwing money at directly — just last week, the company signed a six-year, $10 billion compute deal with startup Volta Infra Holdings for Nvidia Vera Rubin capacity in Norway. Buying a team that specializes in getting more out of the chips you already have is the cheaper half of the same problem that $10 billion deal is trying to solve on the supply side.

Then there's the timing. Anthropic has reportedly filed confidentially for an IPO and has been on an acquisition run through 2026, including the developer-tooling startup Stainless. Bulking up on differentiated technology — rather than just raw compute — right before going public is a familiar pre-IPO playbook: it gives bankers and future shareholders a growth story that isn't purely "we bought more GPUs."

The Bigger Pattern

Zoom out, and this deal fits a trend that's been building across the industry for months. Nvidia shipped Cosmos, an open physical-AI world model, specifically to seed exactly this kind of robotics and simulation work across the ecosystem. Researchers have been framing 2026 as the year "world models grew up" — the point where AI stopped just describing scenes and started predicting how they change. If Anthropic closes this deal, it will be spending nine figures a month, effectively, to make sure it isn't left out of that shift.

For now, the deal is still just talk — sourced to people familiar with the negotiations, not confirmed by either company. But the fact that Anthropic is willing to spend more on this than on any acquisition in its history says something about where the company thinks the next competitive line is being drawn: not who writes the best chatbot reply, but who best understands — and can cheaply simulate — the physical world the rest of us actually live in.

Sources

Fortune: https://fortune.com/2026/08/13/anthropic-said-in-talks-to-buy-startup-decart-for-6-billion/

PYMNTS: https://www.pymnts.com/news/artificial-intelligence/2026/anthropic-pursues-6-billion-decart-deal-to-cut-ai-costs/

BigGo Finance: https://finance.biggo.com/news/dac44a03-cf1b-4b6d-b342-e7c22852a593

TradingKey: https://www.tradingkey.com/analysis/stocks/us-stocks/262101232-anthropic-acquires-decart-ai-6-billion-video-generation-chip-optimization-tradingkey