Anthropic Closes In on a $7 Billion Decart Deal After Beating Nvidia to the Table

In a summer that has already delivered a string of eye-popping AI deals, Anthropic may be about to sign its biggest check yet. The Claude maker is closing in on an acquisition of Decart AI, the Israeli startup behind some of the industry’s most aggressive inference-efficiency technology, at a valuation of roughly $7 billion — up from the $6 billion figure reported when talks first became public on August 13. According to CTech, most of the consideration is expected to be paid in Anthropic shares, and the deal would be the company’s largest acquisition to date.

The story of how the deal got here is as interesting as the price tag. According to reporting from CTech and LinkedIn summaries of the negotiations, Anthropic had to beat back rival interest from Nvidia itself — Decart’s own investor — to get to the table. Google and SpaceX were reportedly still in the mix as potential alternative buyers as late as this weekend.

Why Decart, and why now

Founded in 2023 by Dean Leitersdorf and Moshe Shalev, Decart has made a name for itself in two intertwined areas. The first is AI infrastructure optimization: software that squeezes dramatically better utilization out of expensive GPU clusters, making training faster, cheaper, and more reliable. The second is real-time generative video — Decart’s “world models” can generate interactive video at scale, a capability that has drawn comparisons to more famous world-model efforts and made the company a fixture in demos of real-time AI simulation.

The company grew to roughly 80 people in its first two years and counts Nvidia among its backers — which is precisely what makes Anthropic’s win notable. When the world’s dominant AI chip company wants to buy an inference-efficiency startup and loses it to one of its largest customers, that tells you where the leverage in this market now sits.

For Anthropic, the strategic logic is straightforward: compute is the constraint on everything. The company’s revenue has exploded — its annualized run rate has climbed from $9 billion at the end of 2025 past $30 billion by spring 2026, with recent reporting indicating quarterly revenue roughly 14 times higher year-over-year — but serving that demand requires enormous amounts of inference capacity. Decart’s technology attacks the cost side of that equation directly. Reuters reported that if the deal completes, Decart’s team would join Anthropic’s inference and performance organization, which tells you exactly how Anthropic intends to use the acquisition: not as a product line, but as a weapon for gross margin.

The IPO shadow over the deal

The timing is impossible to ignore. Anthropic raised $65 billion at a $965 billion post-money valuation earlier this year, making it the most valuable private AI company in Silicon Valley. Multiple outlets report the company is moving toward an IPO, and an acquisition of this size — paid largely in stock — makes the most sense as a pre-IPO move: absorb a strategic asset, consolidate its team and technology, and present public-market investors with a cleaner efficiency story.

Paying in shares rather than cash also preserves Anthropic’s balance sheet at a moment when the AI industry’s capital intensity is under intense scrutiny. It mirrors the broader pattern of 2026’s mega-deals, where the industry’s richest players use their own inflated paper to buy the capabilities they need — OpenAI’s recently finalized Nvidia-SoftBank compute arrangements and Stripe’s $7 billion-plus acquisition of OpenRouter follow the same template of using scale and currency to lock in scarce capabilities.

What it says about the inference wars

The most telling detail in the reporting is who else wanted Decart. Nvidia’s interest — and Google’s and SpaceX’s — confirms that inference efficiency has become one of the most contested frontiers in AI. When model providers, chipmakers, and hyperscale-curious outsiders are all bidding on the same small pool of companies that can make GPUs do more with less, the “efficiency layer” of the AI stack is no longer a niche. It is the margin between profit and loss at frontier scale.

For Anthropic specifically, the economics are stark. The company just posted its first-ever quarterly profit on the back of surging Claude enterprise adoption, and its biggest cost line is compute. Every point of inference efficiency Decart’s technology can deliver flows almost directly to Anthropic’s bottom line — and, if the IPO chatter is real, straight into its public-market valuation multiple.

Israel’s deepening role in frontier AI

The deal also underscores how central Israel’s AI ecosystem has become to the frontier layer of the industry. Decart is the second high-profile Israeli AI exit story of the year, and the country’s combination of elite systems engineering talent and military-grade optimization culture has made it an unlikely but durable supplier of exactly the kind of infrastructure talent that companies like Anthropic now compete over. A $7 billion price for a company founded in 2023 — roughly three years from founding to a multi-billion-dollar exit — will only intensify that competition.

What to watch

The deal has not been formally announced, and talks of this size can still fall apart. But the direction of travel is clear. If it closes, expect Anthropic to fold Decart’s team into its inference organization quickly, with the efficiency gains showing up in Claude’s serving costs rather than as a standalone product. Expect Nvidia to keep shopping in the same aisle — it has both the motive and the wallet. And expect the pre-IPO acquisition window to stay wide open: when your stock is your most valuable acquisition currency, the rational move is to spend it before the market gets a vote.

For an industry that has spent 2026 consolidating at breakneck speed, the Decart deal is less a surprise than a confirmation: the scarce resource in AI is no longer ideas or even talent — it is the ability to run the models cheaply at planetary scale. Anthropic is betting $7 billion that Decart is one of the keys to that capability. The fact that it had to outbid Nvidia to get it may be the strongest signal of all.