What it means that Anthropic is negotiating to buy Decart for 6 billion — August 13, 2026
· CompaniesAutomation
Flash Radar Edition: Anthropic is negotiating the largest purchase in its history, about 6 billion for the Israeli firm Decart, and it's not about smarter models but about cheaper inference. What it means for what you pay for AI.
Flash edition. Anthropic is negotiating the purchase of the Israeli firm Decart for approximately $6 billion: it would be the largest acquisition in its history and isn't about smarter models, but about making every response cost less electricity. A supplement to today's Radar.
What happened
Bloomberg reported that Anthropic is in talks to acquire Decart for about $6 billion, in what would be its largest purchase to date (Source). Decart, founded in 2023 by Dean Leitersdorf and Moshe Shalev, has raised about $450 million in total, and its last round of $300 million in May valued it at $4 billion: the offer implies a premium of approximately 50% in three months. The company is known for its world models—Oasis, which generates simulated environments for training robotics and autonomous driving, and Lucy, which edits video in real-time—but what Anthropic seems to want is the "boring" layer: software that squeezes performance out of different chips and lowers the cost of training and inference. Reuters points out that the Decart team would integrate into Anthropic's inference and performance organization (Source). The talks are in an early stage and could fall through; neither company has commented.
Why it matters
Paying 6 billion for efficiency, rather than for a larger model, says a lot about where the pressure in the sector lies today: the bottleneck is no longer the model's intelligence, but the cost per token and the computing capacity to absorb demand. Anthropic is making this move while preparing for its IPO, and facing Chinese competition that has set the price floor very low. For those who buy AI via API, the translation is direct: the structural trend of price per token remains downward, and the margin is being fought by providers at the infrastructure level, not on your bill this month. That said, this is an early-stage negotiation, not a price announcement: nothing changes in your contract tomorrow.
For your company
Three concrete things. One: if you are signing a yearly AI spending commitment at today's prices, negotiate a downward revision clause or quarterly commitments; locking yourself into the current rate for twelve months is betting against the only clear trend in the sector. Two: do not design your automation around a single provider. Isolate model calls behind your own layer to be able to switch engines when price or latency moves; consolidations like this rearrange the map every few months. Three: if you discarded a use case because it was too expensive—massive transcription, video analysis, agents with long reasoning paths—set a review date for three months instead of considering it dead. If you want that portable architecture in writing, an AI consultancy can have it set up before the next price drop.
Frequently Asked Questions
Is what I pay to use Claude going to go down?
Not immediately or guaranteed. The purchase, if closed, improves Anthropic's inference efficiency, and that efficiency usually ends up in the price or in more available capacity. But it is an early-stage negotiation: plan with today's prices and take advantage of the drops when they arrive.
Should I wait for this to clear up before automating something with AI?
No. What changes is the price and speed of the engine, not the underlying work: mapping the process, cleaning the data, defining the human control point. That takes weeks and works just as well with any provider.
What if the deal doesn't close in the end?
The underlying reading doesn't change. The fact that Nvidia, Amazon, and others have looked at the same company confirms that computing efficiency is the strategic asset of the moment, whether Anthropic gets it or not.