Analysis
The Talks
Anthropic is negotiating to acquire Decart, an Israeli AI startup, for roughly $6 billion, according to Bloomberg, which first reported the discussions. Fortune's follow-up confirms the same figure and notes the talks remain early enough that they could still collapse -- neither Anthropic nor Decart has commented on the record. If it closes, this would be Anthropic's largest disclosed acquisition to date, well ahead of the smaller tuck-in deals it has made for research talent and tooling over the past two years.
What Decart Actually Builds
Decart's public product is real-time, interactive video generation -- AI models that can produce and modify video on the fly rather than rendering it offline, a capability closer to a live game engine than a traditional generative-video tool like OpenAI's Sora, Runway's Gen-4, or Google's Veo. The less-public but strategically more important piece is Decart's GPU-efficiency software, which is designed to squeeze more usable inference out of the same chips. That's the part Bloomberg's sourcing points to as Anthropic's real motivation: the company has spent 2026 signing one compute deal after another -- including a $10 billion agreement with Volta Infra Holdings and roughly $45 billion in committed compute from Elon Musk's xAI in May -- because Claude demand has consistently outpaced the GPU capacity Anthropic can secure. Software that stretches existing GPU fleets further is a cheaper lever than buying more chips.
Company Background
Decart was founded by Dr. Dean Leitersdorf, who serves as CEO, and Moshe Shalev, its chief product officer -- both veterans of Unit 8200, the Israeli military's elite technology unit that has produced a disproportionate share of the country's cybersecurity and AI founders.
The company's funding history moves fast even by 2026 standards:
- May 2026 Series -- $300 million, Pulse previously covered, valuing Decart at $4 billion
- Total raised to date -- roughly $450 million
- Reported acquisition price -- about $6 billion, a premium of roughly 50% over the May mark set just three months earlier
The Competitive Field
Real-time video generation is a crowded and well-funded category. Runway, Pika, Luma AI, and the incumbent labs' own tools (OpenAI's Sora, Google's Veo) are all racing on latency and interactivity, not just output quality. On the inference-efficiency side -- arguably the acquisition's real prize -- Decart competes indirectly with Groq, Cerebras, SambaNova, and Together AI, all of which sell speed and cost advantages on AI workloads rather than raw model capability. Anthropic buying rather than partnering with an efficiency player is notable: Nvidia, Microsoft, and Google have all preferred long-term supply contracts over acquisitions when solving the same compute-crunch problem this year.
Numbers in Context
A $6 billion price for a company that has raised $450 million total puts Decart's multiple on invested capital in territory usually reserved for frontier labs, not applications built on top of them.
That pricing lands in a month full of similar signals:
- Cognition -- reportedly fielding funding talks north of $40 billion
- SpaceX / Anysphere (Cursor) -- closed a $60 billion all-stock purchase in June
Three separate deals pricing AI-tooling companies off infrastructure scarcity as much as product traction.
The Counterweight
None of this is finalized. Bloomberg's own reporting frames the talks as early-stage, and acquisition rumors at this size fall apart more often than headlines suggest -- price, integration terms, and regulatory review (Decart is an Israeli company being bought by a US AI lab now bulking up ahead of a possible IPO) are all still open questions. Even if the deal closes, folding an outside GPU-efficiency stack into Anthropic's existing infrastructure is a real integration risk, not a guaranteed unlock -- efficiency software tuned for Decart's own use case doesn't automatically generalize to Claude's training and serving stack.
Ahead
Watch whether Anthropic files anything with regulators before its own IPO process advances -- a $6 billion acquisition this close to a potential public listing adds a data point bankers and investors will want explained in the S-1, not buried in a footnote.
