Anthropic's $6 Billion Bet: Why the Claude Maker is Buying Decart AI

Anthropic is preparing to drop $6 billion on a three-year-old startup, and it has nothing to do with adding more parameters to Claude.

According to reports from Bloomberg and Fortune, the AI research lab is in advanced talks to acquire Israeli AI startup Decart for roughly $6 billion. If finalized, this will be Anthropic’s largest acquisition to date—and its fifth this year alone.

But this isn't just an acqui-hire. The impending Decart acquisition highlights a critical pivot in the artificial intelligence industry: the war for AI dominance has shifted from pure reasoning capabilities to infrastructure efficiency and real-time multimodal generation. As Anthropic gears up for a highly anticipated IPO, the company is making a massive, multi-billion-dollar bet that Decart holds the key to slashing inference costs and unlocking the next generation of "world models."

Here is a deep dive into why the Claude maker is willing to pay a 50% premium on a startup that was valued at $4 billion just three months ago.

The Meteoric Rise of Decart AI

To understand the $6 billion price tag, you have to look at Decart’s unprecedented trajectory. Founded in September 2023 by Israeli engineers—brothers Dean and Orian Leitersdorf, alongside Moshe Shalev—Decart has been one of the quietest yet most formidable players in the AI infrastructure space.

The startup’s valuation history tells the story of an industry desperate for its technology:

• August 2025: Valued at $3.1 billion.
• May 2026: Raised $300 million at a $4 billion valuation. The round was led by Radical Ventures, with heavy-hitting participation from Nvidia, Adobe Ventures, Atreides Management, and Valor Equity Partners.
• August 2026: Now in advanced acquisition talks with Anthropic for $6 billion.

"I really can’t imagine being in any other market," co-founder Dean Leitersdorf remarked during their May funding round. "This market is moving so fast. We get to reinvent every different aspect of the economy in the next 18 to 24 months."

Decart operates on two distinct but highly complementary pillars: hardware-agnostic compute efficiency and real-time world models. For Anthropic, both are existential necessities.

Pillar 1: Solving the Inference Crunch

The open secret in the generative AI space is that while training models is expensive, serving them at scale is a financial black hole. Anthropic has been bleeding cash on compute to keep up with the surging demand for its Claude models.

The company’s compute strategy has been aggressive and highly diversified. Over the past year, Anthropic has:

• Committed to long-term arrangements using Amazon Web Services' custom Trainium chips.
• Announced plans to heavily utilize Google's Tensor Processing Units (TPUs).
• Struck a massive compute-power deal with SpaceX.
• Reportedly entered talks to lease up to $10 billion worth of computing capacity from Meta over the next two years.

But throwing money at hardware isn't enough; the software layer needs to be optimized to squeeze every drop of performance out of that rented and owned silicon. This is where Decart comes in.

Decart has developed proprietary software designed to drastically lower AI training and inference expenses by improving how efficiently chips are utilized. According to insiders, if the deal closes, Decart's engineering team will be directly absorbed into Anthropic's inference and performance organization.

By vertically integrating Decart’s efficiency algorithms, Anthropic hopes to fundamentally alter the unit economics of running Claude. In a landscape where competitors like DeepSeek and xAI are aggressively driving down API costs, Anthropic needs Decart’s tech to maintain its margins without sacrificing the top-tier performance of its frontier models.

Pillar 2: World Models and Real-Time Video

While Decart’s efficiency software is the financial engine of this deal, its "world models" are the product engine.

A world model is an AI system designed to simulate the physics, spatial dynamics, and logic of the physical world. While OpenAI has focused heavily on asynchronous video generation with Sora, Decart has focused on real-time, interactive generation.

Decart has two flagship models that have caught the attention of the industry:

• Oasis: A model built to create highly accurate synthetic environments. Oasis isn't just a video generator; it's a physics engine powered by neural networks. It allows autonomous systems to train in infinitely variable, physically accurate virtual worlds without the constraints of real-world data collection. It is heavily utilized in the development of robotics and self-driving technology.
• Lucy: A consumer-facing model that processes live video feeds to produce real-time, high-resolution modifications.

Lucy, in particular, is a technological marvel. It solves one of the most notoriously difficult problems in computer vision: the realistic, real-time rendering of fabrics. E-commerce giant eBay—which is both an investor and a customer—uses Lucy to allow shoppers to see themselves wearing clothing or accessories via live video feeds.

Furthermore, Lucy is already being deployed by top-tier influencers and VTubers on platforms like Twitch, TikTok, and YouTube (including notable creators like CodeMiko) to generate real-time avatars and environments with near-zero latency.

By acquiring Decart, Anthropic instantly leapfrogs into the real-time video and spatial computing race. Instead of just generating text or analyzing static images, Claude could soon be capable of real-time video manipulation, interactive synthetic environments, and live multimodal streaming.

The IPO Catalyst

Context is everything, and the timing of this $6 billion acquisition is no coincidence. In early June 2026, Anthropic submitted a confidential S-1 filing to the SEC, signaling its preparation for what will likely be the most anticipated tech IPO of the decade.

The S-1 filing set the clock ticking. Public markets have grown increasingly skeptical of the "growth at all costs" mentality that defined the early generative AI boom. Investors are demanding clear paths to profitability and sustainable unit economics.

A $6 billion acquisition right before an IPO is a massive flex, but it’s also a strategic necessity. By bringing Decart in-house, Anthropic is vertically integrating its software stack. If Decart’s algorithms can improve chip utilization by even 15-20%, that translates to billions of dollars in saved inference costs over the next decade. It transforms Anthropic's financial narrative from a capital-intensive research lab into a highly optimized, vertically integrated AI powerhouse. It sends a clear message to future shareholders: We are not at the mercy of cloud providers and chip manufacturers. We own the software that makes AI infrastructure profitable.

The Broader Market Impact

If finalized, Anthropic’s acquisition of Decart will send shockwaves through the AI startup ecosystem. It establishes a clear precedent: the highest premiums are no longer being paid for pure LLM wrappers or niche AI agents. The real money is in infrastructure optimization and interactive world models.

This acquisition also puts immense pressure on competitors. OpenAI, Google, and Meta will likely accelerate their own M&A strategies to capture remaining independent infrastructure startups. The era of foundational model companies relying entirely on third-party cloud providers for efficiency optimizations is ending. We are entering an age of deep vertical integration, where the line between model architecture, inference software, and hardware utilization is completely blurred.

For the rest of the industry, this deal is a wake-up call. The frontier of AI is no longer just about who has the smartest text model. It’s about who can run that model the cheapest, and who can extend that intelligence into the physical, real-time world. With Decart in its arsenal, Anthropic is proving it intends to win on all fronts.