Anthropic is reportedly in talks to acquire AI infrastructure startup Decart AI for about $6 billion. Decart has built software designed to make AI training and inference more efficient across several types of processors, while also developing real-time generative AI systems such as Lucy and Oasis. Under a deal, Decart’s team may join Anthropic’s inference and performance organization.
About Decart AI
Decart was founded in 2023 by Israeli engineers Dean Leitersdorf, Orian Leitersdorf and Moshe Shalev. Dean Leitersdorf and Moshe Shalev are among the company’s leadership.
The company’s early work concentrated on AI systems optimization.
Before Decart became known for realtime generative experiences, it was building software to lower the computational cost of training and running AI systems. Decart’s first enterprise product optimized GPU use and was already generating millions of dollars in revenue.
DOS: the infrastructure layer:
Decart describes DOS as an inference and training stack focused on high-performance, low-latency AI systems. The company said when announcing its May 2026 funding that DOS 2.0 was designed to help agents and reasoning models run more efficiently, while supporting its Lucy and Oasis models.
Lucy: real-time generative video:
Lucy is Decart’s real-time video model. The company says its latest Lucy 2.5, announced in July 2026, can generate live video effects at 30 frames per second, with improvements in time consistency, visual quality and latency.
Oasis: interactive world models
Oasis is Decart’s world-model line, aimed at generating interactive environments in real time.
Its early Oasis demonstration became widely known because users could interact with a Minecraft-like environment generated by an AI model. The product was one of the company’s first major consumer demonstrations, while Decart has since expanded the concept toward physical AI and interactive simulation.
Decart introduced Oasis 3 in June 2026 as an interactive world model for physical AI.
[ALSO READ: OpenAI Acquires Tech Show TBPN, Cited by Sam Altman as His “Favorite” ]
Key detail
| Metric | Figure |
| Reported potential Decart acquisition | About $6 billion |
| Decart’s latest funding round | $300 million |
| Decart’s total funding after that round | More than $450 million |
| Decart’s latest reported valuation | Nearly $4 billion |
| Anthropic Series H funding | $65 billion |
| Anthropic post-money valuation | $965 billion |
| Anthropic run-rate revenue reported in May 2026 | More than $47 billion |
| Amazon compute agreement | Up to 5 GW |
| Google/Broadcom next-generation TPU agreement | 5 GW |
| SpaceX Colossus capacity | More than 220,000 Nvidia GPUs |
Why the Decart Deal Is Different From a Typical AI Acquisition
A large AI acquisition usually attracts attention because of the model, consumer product or customer base being acquired.
Decart is different.
The company has built an optimization layer for AI computing, with the Decart Optimization Stack, or DOS, alongside its own real-time AI models. Decart says DOS is created to optimize inference and training across several hardware platforms, while Lucy and Oasis sit on top of that infrastructure.
For Anthropic, that is a particularly relevant problem.
Anthropic says Claude is trained and run across AWS Trainium, Google TPUs and Nvidia GPUs. The company has continued to add capacity across all three.
So Decart AI’s ability to work across hardware architectures is closely related to a problem Anthropic already has.
[ALSO READ: Amazon, Nvidia, and SoftBank Power OpenAI’s Historic $110 billion Capital Raise ]
Who Funds Decart AI?
Decart’s investor list has grown substantially alongside the company.
Its $300 million financing round announced in May 2026 was led by Radical Ventures and included Nvidia, Sequoia Capital, Benchmark, Atreides and other investors. Radical Ventures said its co-founder and managing partner Jordan Jacobs would join Decart’s board.
According to The Wall Street Journal, additional prominent backers include Michael Eisner, OpenAI co-founder Andrej Karpathy and the Yamauchi family, alongside Nvidia and institutional investors.
In December 2024, TechCrunch reported that Decart had raised $32 million at a valuation above $500 million.
In August 2025, Fortune reported that the company raised $100 million at a $3.1 billion valuation.
By May 2026, the company had raised another $300 million, bringing total funding above $450 million, and the Wall Street Journal said the round valued Decart at nearly $4 billion.
If the reported $6 billion Anthropic transaction were completed at that price, it would represent roughly a 50% premium to Decart AI’s latest reported valuation.
There are several reasons the reported acquisition makes technical sense.
Anthropic is adding huge amounts of compute
Anthropic has been unusually explicit about its infrastructure requirements.
In April, the company announced an agreement with Amazon for up to 5 gigawatts of compute capacity, including new Trainium capacity. Anthropic said it currently uses more than one million Trainium2 chips to train and serve Claude.
In another April announcement, Anthropic said it had signed an agreement with Google and Broadcom for 5 gigawatts of next-generation TPU capacity, expected to come online from 2027.
Then Anthropic announced access to more than 220,000 Nvidia GPUs at SpaceX’s Colossus 1 data center, representing more than 300 megawatts of additional capacity.
At this level, adding hardware is only one part of the economics.
The organization must also manage how efficiently that hardware is used.
Inference is becoming a bigger cost question
Training gets most of the attention because the numbers attached to frontier-model training are enormous.
Inference can be less visible but is continuous.
Every time a customer asks Claude a question, runs an agent, generates code or uses an enterprise workflow, computing resources are consumed.
As usage rises, inefficiencies which seem small at the level of a single request can grow substantial across millions or billions of requests.
This is where optimization can become strategically important.
A faster inference stack does not replace GPUs or TPUs. But it can improve throughput, decrease latency, improve hardware utilization or reduce the amount of capacity required for a given workload.
That is a very different value proposition from simply acquiring another AI model.
[ALSO READ: US Lifts Export Controls, Anthropic Restores Access to Fable 5 and Mythos 5 AI Models ]
What It Means for Claude
A Decart acquisition would not necessarily produce a new Claude feature immediately.
The more likely effect would be underneath the product: inference performance, hardware utilization, deployment efficiency and potentially the cost of serving models.
Anthropic appears to be interested in Decart AI’s engineering capability and infrastructure technology, not simply in acquiring Lucy or Oasis as consumer products.
Over time, better infrastructure could translate into faster responses or additional capacity, but that is still a potential outcome rather than a publicly announced benefit.
The Real Value May Be the Engineering Stack
Anthropic does not need another general-purpose AI startup simply for the sake of adding products. It already operates one of the world’s largest frontier-model businesses.
What it does need is greater control over the systems that determine how Anthropic Claude is trained and served, Decart has spent its relatively short history working on precisely that layer.
The company began with training and inference optimization, built an enterprise business around it, and then used that technical foundation to develop real-time AI products.
That history makes the reported acquisition more interesting than a conventional AI startup purchase.
It suggests that AI compute efficiency is becoming valuable enough to be acquired at frontier-model scale.
