Key takeaways
- Decart is more than an AI video startup: its core asset is a vertically integrated optimization stack designed to make AI training and inference faster and less expensive.
- Decart complements Anthropic at a critical bottleneck. Its infrastructure expertise could help Claude serve rapidly growing demand while improving latency and compute economics.
- The acquisition could also establish Anthropic’s first major Israeli R&D operation, giving it immediate access to approximately 100 employees and one of the world’s densest AI ecosystems.
For several days, the Israeli tech ecosystem was buzzing with reports that Decart was nearing a $6–7 billion sale, representing a roughly 50% premium to Decart’s $4 billion valuation just three months earlier.. SpaceX emerged as the most intriguing potential buyer, although Nvidia, Amazon and Nebius were also reportedly in the frame. Now, Anthropic appears to be the company at the table.
According to Reuters, Anthropic is in early-stage talks to acquire the Israeli-founded AI startup for approximately $6 billion. If completed, Decart’s team would reportedly join Anthropic’s inference and performance organisation. Neither company has confirmed the negotiations, and early acquisition talks have a habit of changing direction, as the apparent pivot from SpaceX already demonstrates.
But the potential transaction, rumoured to be priced around $6 billion could make strategic sense.
So, what exactly does Decart do?
Decart is sometimes described as an AI video company, but that only captures its most visible output.
Founded in 2023 and led by Dean Leitersdorf and Moshe Shalev, Decart initially focused on reducing the cost of training and running large AI models. Its core technology, the Decart Optimization Stack (DOS), is a vertically integrated training and inference platform spanning model architecture, compilers, kernels, memory management and the underlying hardware.
In plain English, Decart aims to make demanding AI models run faster and more cheaply on existing chips.
The company used that infrastructure to develop several increasingly ambitious public products:
- Oasis, launched in 2024 with chip startup Etched, was an early real-time “world model.” Instead of rendering a conventional video game with a graphics engine, Oasis generated a Minecraft-like environment frame by frame in response to the player’s actions. It was imperfect and dreamlike, but it offered a compelling glimpse of AI-generated interactive worlds.
- Mirage, since folded into the Lucy product family, applied similar technology to live video transformation. It could change a person, background or visual style while a video was streaming—technology that Wired described as poised to reshape streaming.
- Lucy is Decart’s current family of real-time video models. It can alter clothing, insert products, transform environments and add effects to live footage. The latest versions run continuously rather than producing short, disconnected clips, opening potential applications in advertising, commerce, entertainment and virtual try-on.
- Oasis 3 takes the world-model thesis beyond gaming and into “physical AI”: generating interactive environments for robotics, autonomous driving, simulation and synthetic training data.
Decart therefore operates across three layers: infrastructure through DOS, immersive media through Lucy, and simulation for physical AI through Oasis. The common thread is not video itself, but low-latency inference, getting sophisticated models to respond at the speed of human interaction.
That technical ambition has attracted serious capital. In May 2026, Decart raised $300 million at a reported $4 billion valuation, bringing its total funding to roughly $450 million. Radical Ventures led the round, with Nvidia joining existing investors including Sequoia, Benchmark and Zeev Ventures.
Why Decart is complementary to Anthropic
Anthropic is best known for Claude, but the frontier-model race is increasingly being determined below the application layer. Training matters, but inference economics, how quickly and cheaply a model can serve millions of users and agents, may ultimately matter just as much.
This is where Decart could be particularly valuable.
First, DOS could help Anthropic extract more performance from scarce and expensive compute. Even modest improvements in utilisation or latency can translate into enormous savings at Anthropic’s scale.
Second, Decart brings deep expertise across the full stack, from hardware-aware optimisation to model architecture. Anthropic has already been recruiting engineers to help co-design chips, systems and models. Acquiring a team that has built this capability from scratch could accelerate that effort considerably.
Finally, Decart would give Anthropic an entry point into world models and real-time multimodal AI. Claude is increasingly able to reason, code and operate software. Decart’s technology points towards agents that can also understand, simulate and manipulate dynamic visual environments.
From a VC perspective, this is the more interesting interpretation of the deal. Anthropic would not be paying $6 billion for a collection of clever video demos. It would be buying an unusually ambitious systems company whose public models demonstrate what becomes possible when inference is fast enough, cheap enough and responsive enough.
There is also a second layer to the transaction: Israel. Anthropic has been quietly increasing its engagement with the Israeli ecosystem, including the acquisition of infrastructure startup Runhouse and the appointment of an Israel-focused go-to-market executive. Israel reportedly ranks first globally in Claude usage relative to its working population and remains one of the world’s densest concentrations of AI talent. Acquiring Decart would give Anthropic something partnerships and remote hiring cannot: an immediate, roughly 100-person team and a credible foundation for a significant Israeli R&D operation.
For Israel’s startup ecosystem, a successful transaction would be another landmark outcome, and a reminder of its enduring strength in deep infrastructure, optimisation and technically complex AI. But Decart may also be reaching an inflection point where its technology can create more value inside a frontier-model platform than as an independent supplier.
The SpaceX rumours were not entirely surprising: Decart’s combination of Nvidia optimisation, world models and robotics simulation would fit naturally with Elon Musk’s expanding AI ambitions. But Anthropic may be an even more complementary home. Claude provides the intelligence layer; Decart could help make that intelligence faster, more visual and ultimately more grounded in the real world.
The talks may still come to nothing. Strategically, however, the match is compelling.
He is a former General Partner at Google Ventures (GV) in Europe, former head of Google for Entrepreneurs in Europe, and founding head of Campus London, Google's first startup hub. Eze writes on Israeli tech, venture capital, artificial intelligence, and founder strategy.
He is also the founder of Techbikers, a nonprofit that brings together the startup ecosystem on cycling challenges in support of Room to Read.
- Why Anthropic May Want to Acquire Israel’s Decart - August 13, 2026
- The $9 Billion Bet on Forward Deployed Engineers - August 12, 2026
- Weekly Firgun Newsletter – August 7 2026 - August 7, 2026
