Amazon Reportedly Considered Acquiring Israeli AI Startup Decart Valued at Approximately $6 to $7 Billion
Reports in August indicated that Israeli AI startup Decart was in potential sale negotiations, with a transaction valuation of approximately $6 billion to $7 billion, with Amazon, SpaceX, and Nebius listed as interested parties.
Decart initially approached an agreement with NVIDIA, but another bidder intervened, leading the founder to turn to a new buyer. Founded in 2023, the company has about 100 employees and has raised approximately $450 million, with a valuation nearing $4 billion after a $300 million funding round in May. Investors include NVIDIA, Sequoia, and Radical Ventures.
Its core technology is a chip efficiency optimization software (Decart Optimization Stack), which claims to enhance the speed of AI models on the same hardware (including NVIDIA GPUs, Google TPUs, and Amazon Trainium) by up to 8 times, while also involving real-time video and world models. Amazon has previously become a strategic customer.
Subsequent reports indicated that Anthropic entered deeper negotiations in mid-August, with a valuation of about $6 billion (partly paid in stock), with the founder and major investors leaning towards Anthropic. In early September, Bloomberg reported that Anthropic withdrew from the deal after completing due diligence, and both parties may explore other collaboration methods; NVIDIA had previously offered a higher bid but did not finalize a deal.
As of October, there is no public information confirming that Amazon has entered exclusive or advanced negotiation stages, nor has any formal acquisition agreement been announced. Decart has not commented on related reports.
Market mechanisms indicate that the event is driven by the scarcity of AI computing efficiency assets, with several giants competing for the same target. Beneficiaries are buyers (cloud and model companies) that can reduce training and inference costs, while pressured parties include unsuccessful bidders and AI infrastructure players reliant on external optimization tools.
Source: Public Information
ABAB AI Insight
Decart was founded in 2023 by Dean Leitersdorf and Moshe Shalev, who have backgrounds in Israel's Unit 8200, quickly shifting from world models and real-time video to chip optimization layers, with its valuation rising from $3.1 billion to nearly $4 billion after the May funding round. NVIDIA is both an investor and a potential buyer, indicating a dual strategy by hardware manufacturers regarding software efficiency tools.
In terms of capital pathways, giants aim to reduce the unit computing power costs of their own chips or cloud clusters through the acquisition of efficiency software. Amazon's focus relates to its Trainium ecosystem and AWS customer relationships; Anthropic hopes to use the technology to absorb more Claude demand with existing computing power, with transaction structures often involving stock payments to preserve cash for infrastructure. After the deal did not finalize, both parties retain the possibility of collaboration.
A comparable case is NVIDIA's acquisition of Run:ai and other scheduling tools to enhance GPU utilization, as well as Anthropic's previous limited large-scale acquisitions, expanding more through cloud partners (Amazon, Google). Currently, AI is in an "expansion phase where computing power costs become a core constraint," making efficiency software a scarce asset that can be independently priced.
Structurally, this represents capital concentration and technological substitution. The scaling of models makes inference and training costs bottlenecks, with optimization software shifting hardware utilization from chip manufacturers to the software layer, making acquisition a path to quickly gain control over that layer. The mechanism is that under fixed hardware capacity, whoever first controls the efficiency multiplier can provide higher effective computing power at the same capital expenditure, thus altering the cost curve for cloud and model companies.
ABAB News · Cognitive Laws
Efficiency software equals the leverage multiplier of computing power
Bidding enthusiasm does not equal final transactions
Due diligence is a filter between valuation and reality