Bloomberg reported on August 13, 2026 that Anthropic is in talks to acquire Decart, an Israeli AI infrastructure and video-generation startup, for approximately $6 billion. The report has since been corroborated by Haaretz, Calcalist, i24News, and a Yahoo Finance syndication of the Bloomberg story — but every one of those outlets describes the same thing: early-stage talks, not a signed deal. Nothing is finalized, and talks of this size fall through often enough that "reportedly in talks" deserves to be taken literally.
If it does close, it would be Anthropic's largest known acquisition to date — arriving as the company is reportedly gearing up for a major public offering and has been aggressively expanding the compute behind Claude. Here's what's actually been reported, and why it matters if you build on Claude rather than just watch the ticker.
TL;DR: Quick Reference
| Question | Answer |
|---|---|
| Is this confirmed? | No. Bloomberg calls it "early-stage talks." Anthropic and Decart have not confirmed a signed deal. |
| How much? | Reportedly around $6 billion — Anthropic's largest known acquisition if it closes. |
| Who is Decart? | An Israeli startup building AI infrastructure/optimization tech plus its own models, including real-time video generation and editing. Founded by Unit 8200 veterans Dr. Dean Leitersdorf (CEO) and Moshe Shalev (chief product officer). |
| How much has Decart raised? | ~$450M total, including a $300M round in May 2026 at a ~$4B valuation, with Nvidia among investors. |
| Why now? | Reported alongside Anthropic's IPO preparations and a push to add compute capacity — including hiring engineers for custom chip work. |
| What does this mean for Claude API users? | Nothing changes today. If the deal closes, the plausible upside is cheaper/faster inference and, longer-term, video generation on Claude's roadmap. |
| Where was this first reported? | Bloomberg, corroborated by Haaretz, Calcalist, i24News, and Yahoo Finance. |
What did Bloomberg actually report?
According to Bloomberg, Anthropic is in talks to buy Decart for roughly $6 billion, with the discussions described as being at an early stage. Haaretz, Calcalist/Ctech, and i24News all picked up and corroborated the Bloomberg reporting through their own sourcing, and Yahoo Finance carries the Bloomberg wire piece directly.
None of that reporting describes a signed agreement, a term sheet, or a closing date. "In talks" in deal reporting typically means a bilateral negotiation is underway — not that both sides have agreed on price, structure, or even that either party is committed to closing. Big-number tech acquisitions that reach this stage do sometimes collapse over valuation, diligence, or regulatory concerns, so treat the $6 billion figure as the number currently on the table, not the number that will necessarily appear on a closing document.
If this deal does eventually get confirmed, it would be worth tracking against the pattern of other AI-adjacent acquisitions this year, like AMD's acquisition of Taalas for inference-optimized silicon — a similar thesis (buy the chip-efficiency layer rather than build it) playing out at a different scale and for a different buyer.
Who is Decart, and what does it actually build?
Decart is an Israeli AI startup founded by two veterans of Unit 8200, the Israeli intelligence corps' elite technology unit that has produced a disproportionate share of the country's cybersecurity and AI founders: Dr. Dean Leitersdorf (CEO) and Moshe Shalev (chief product officer).
Decart's business has two parts, per the reporting:
- AI infrastructure and optimization tech — tooling that helps developers extract more performance out of different chip architectures, i.e. inference-efficiency and hardware-abstraction work rather than a single-chip product.
- Its own AI models, including real-time video generation and editing technology — Decart is one of the more closely watched names in the fast-moving real-time and agentic video-generation space that's drawn increasing attention over the past year.
Decart has raised approximately $450 million total, including a $300 million round in May 2026 at a roughly $4 billion valuation, with Nvidia among its investors. A $6 billion acquisition price implies roughly a 1.5x step-up from that valuation in just three months — a steep premium, though not unusual for a strategic acquisition where the buyer values technology and team over the round's original financial math.
Why is Anthropic reportedly doing this now?
Two threads in the reporting point at the strategic logic, and both matter more to Claude users than the price tag does.
1. Anthropic is preparing for an IPO and needs more compute
The reporting ties this deal to Anthropic's broader push toward a public offering and its aggressive buildout of available compute for Claude's services. Anthropic filed a confidential S-1 back in June 2026 and had bankers scheduling investor meetings by mid-July, pointing to a possible listing later this year. That buildout has been visible for months on the compute side too — AMD's up-to-$5 billion investment tied to 2 gigawatts of MI450 Instinct capacity being the clearest recent example. An IPO raises the bar on demonstrating a durable, scalable compute strategy to public-market investors, not just a research roadmap — and inference-efficiency technology like Decart's is exactly the kind of asset that improves the unit economics investors scrutinize before a listing.
2. Anthropic has been hiring for custom chip work
Per the reporting, Anthropic said just last week that it was hiring engineers with hardware and software stack experience to work on custom chips — the same direction OpenAI took with its Jalapeño inference chip built with Broadcom and Google has pursued with its own custom TPU silicon. Decart's stated specialty — squeezing more performance out of different chips — lines up directly with that hiring push. Buying an infrastructure-optimization team is a faster path to chip-level inference gains than building the expertise from scratch, and it would put Anthropic in the same "buy the efficiency layer" pattern AMD just followed with Taalas.
What would this mean for people building with Claude?
Nothing changes today — talks are not a signed deal, and even a signed deal typically takes months to close and integrate. But it's worth reasoning through the "if it closes" scenario, because it maps onto two things developers already care about.
Compute capacity and inference cost. Anthropic has faced API pricing and capacity pressure as demand for Claude has scaled, and rivals continue to undercut it on raw per-token price even as Anthropic emphasizes cost-per-completed-task. Inference-optimization technology that squeezes more throughput out of existing chip fleets is one of the few levers that improves margin and capacity at the same time, without needing new fabs or new GPU allocations. If Decart's tech genuinely delivers on that promise inside Anthropic's stack, the plausible downstream effect is fewer capacity crunches and more room to hold or improve pricing rather than raise it.
A possible video-generation feature on Claude's roadmap. Anthropic has so far stayed out of the real-time video generation race that OpenAI, Google, and video-native startups have been running hard at. Decart's real-time video generation and editing tech would be the fastest way for Anthropic to acquire that capability rather than build it in-house — though there's no confirmation yet that any Decart video product would ship inside Claude specifically, versus staying a separate product line or being absorbed purely for its underlying optimization research.
Both of those are downstream of a deal that hasn't happened yet. Worth watching, not worth planning around.
What people are asking
Why would Anthropic pay a premium over Decart's May valuation? Strategic acquirers routinely pay above the last private round when they value team, technology, and speed-to-integration over the financial math the round was priced on. A $2 billion premium over a three-month-old $4 billion valuation is large but not unprecedented for a buyer racing to close a compute-efficiency gap ahead of a public listing.
Could this fall through? Yes. Bloomberg explicitly frames this as early-stage talks. Price, structure, regulatory review (cross-border deals involving Israeli-founded companies and US buyers can draw extra scrutiny), and simple negotiation breakdown are all plausible off-ramps between "in talks" and "signed."
Does this affect Claude pricing or API limits right now? No. There's no reported change to Claude's API pricing, rate limits, or product roadmap tied to this news as of publication. Any effect would only materialize after a deal closes and Decart's technology is actually integrated.
Is Decart's video-generation product going to become a Claude feature? Unconfirmed. The reporting establishes Decart's video capability as part of what makes it attractive to Anthropic, not that a specific Claude video feature is planned.
The honest read
This is a story about compute strategy, not a done deal. Bloomberg's reporting is solid and has been corroborated by four other outlets, which is enough to take the talks themselves seriously — but "in talks" and "acquired" are different facts, and this article will need an update the moment either happens (or the moment talks are reported dead). What's genuinely notable, regardless of outcome, is the size of the number being discussed and what it says about how seriously Anthropic is treating inference efficiency and compute capacity heading into a public offering: enough to consider its largest acquisition yet for a team that specializes almost entirely in that problem.
Related reading
- Anthropic IPO Path 2026: S-1, Banker Meetings, and What Changes for Builders — the public-offering timeline this deal is reportedly tied to.
- AMD Invests Up to $5B in Anthropic for 2GW of MI450 Capacity — the compute buildout this reported deal would extend.
- AMD Acquires Taalas: The Chip That Etches Model Weights Into Silicon — a similar "buy the inference-efficiency layer" acquisition, for comparison.
- OpenAI's Jalapeño: First Custom AI Chip, Co-Developed With Broadcom — the custom-silicon path a rival lab has already taken.
- Anthropic Makes Claude Sonnet 5 Pricing Permanent at $2/$10 — the pricing and capacity pressure this deal's efficiency thesis would bear on.
- VIMAX: Agentic Video Generation — Complete Guide — context on the real-time video-generation space Decart also competes in.
- Dario Amodei Worries New Anthropic Hires Are Chasing Pay, Not Mission — more on Anthropic's talent and hiring strategy this year.
- Nvidia's $500 Billion Compute-as-an-Asset-Class Pitch to Wall Street — the broader compute-financing backdrop this acquisition talk sits inside.
- Claude Sonnet 5 vs GPT-5.6 Luna Max: Which Is the Cheaper Workhorse? — where Claude's cost-per-task currently stands against rivals.
Sources: Bloomberg — Anthropic Said in Talks to Buy AI Startup Decart for $6 Billion (Aug 13, 2026) · Haaretz · Calcalist/Ctech · i24News · Yahoo Finance (Bloomberg syndication)
This article reflects reporting as of August 13, 2026. The Anthropic-Decart talks are described by all sources as early-stage and unconfirmed; deal terms, valuation, and whether the acquisition closes at all may change. This post will be updated if the talks are confirmed, revised, or called off.