# Anthropic in Talks to Pay $6B for Decart, Its Biggest Deal Ever By Ray Termsheet · 2026-08-13 · Deals & Funding · https://datacommenter.com/anthropic-in-talks-to-pay-6b-for-decart-its-biggest-deal-ever/ About the author: Follows the money in the data economy: venture rounds, M&A, and strategic stakes across data vendors and infrastructure. > Anthropic is negotiating to acquire Israeli AI infrastructure startup Decart for roughly $6 billion, Reuters and Bloomberg reported August 12, 2026 — a 50% markup on the $4 billion valuation… Original reporting: [Reuters](https://www.reuters.com/technology/anthropic-talks-buy-decart-ai-source-says-2026-08-13/) _AI-assisted commentary, editorially reviewed. Quoted excerpts belong to the original outlet._ If it closes, this is Anthropic’s largest acquisition to date and a bet that owning inference-optimization talent matters as much as owning the model itself. Decart, founded in 2023 by two Unit 8200 alumni, builds software that squeezes more performance out of Nvidia, AWS Trainium, and Google TPU chips — exactly the kind of unglamorous plumbing that becomes strategically vital once a company like Anthropic is running Claude at a reported $47 billion annualized revenue run rate and burning cash on compute deals with Microsoft, AWS, and Fluidstack. > A 50% valuation jump in under three months isn’t due diligence pricing Decart’s technology — it’s Anthropic pricing scarcity and a bidding war it may be losing control of. The number worth sitting with: Decart raised $300 million in May 2026 at a $4 billion valuation, with Nvidia and Toyota participating, and is now reportedly being valued at $6 billion — a jump that outpaces almost anything else in Israeli tech this cycle, according to Proactive. That escalation reportedly happened because Nvidia itself was in advanced talks to buy Decart before higher bids arrived from Amazon, Nebius, and (briefly) SpaceX, per Calcalist’s reporting cited by Proactive. When your own strategic investor gets outbid for your target, the seller has all the leverage, and Anthropic’s willingness to pay up suggests it sees Decart’s Lucy and Oasis video-generation models as a shortcut past its acknowledged multimodal weakness versus rivals. ### Pattern, not one-off This fits a now-familiar Anthropic playbook: buy small, technical teams rather than products. Bun (December 2025), Vercept (February 2026), and Stainless (May 2026, reportedly $300 million-plus) were all developer-tooling and infrastructure plays, not consumer acquisitions. Decart would be the biggest by an order of magnitude, and the first aimed squarely at inference cost — the line item every AI lab is quietly panicking about as usage scales faster than margins. The competitive subtext is OpenAI, which has run a louder, pricier M&A campaign of its own, including the $6.5 billion Jony Ive io deal in May 2025. Anthropic buying Decart at a comparable price tag, timed just ahead of a possible IPO as soon as next month, reads as much like signaling to public-market investors as it does engineering strategy. Watch whether the deal actually closes at $6 billion, whether Nvidia’s earlier interest resurfaces as a competing bid, and whether Decart’s ~100-person team survives the integration or scatters the way acquired talent often does once the earn-out clock starts. > Anthropic in talks to buy Decart AI, source says > — [Reuters](https://www.reuters.com/technology/anthropic-talks-buy-decart-ai-source-says-2026-08-13/) [Read the full story at Reuters →](https://www.reuters.com/technology/anthropic-talks-buy-decart-ai-source-says-2026-08-13/) --- Cite this analysis: https://datacommenter.com/anthropic-in-talks-to-pay-6b-for-decart-its-biggest-deal-ever/ Cite primary facts: https://www.reuters.com/technology/anthropic-talks-buy-decart-ai-source-says-2026-08-13/ Need the underlying datasets (alt data, market data, AI training data)? Source licensed vendors via Brickroad: https://brickroad.network More machine-readable access: https://datacommenter.com/llms.txt ## Participate - Comment on a passage: MCP `add_note` (include `source_url` when available). - Suggest an editorially reviewed correction: MCP `suggest_edit`. - Open factual questions: none.