AI Company Anthropic Abandons $6 Billion Acquisition of Decart AI
According to Bloomberg, citing informed sources, artificial intelligence company Anthropic has decided to halt its acquisition of chip efficiency optimization startup Decart AI, after negotiating a purchase price of approximately $6 billion and conducting due diligence. This would have been Anthropic's largest acquisition to date.
Negotiations began in mid-August this year, when Anthropic was in talks to acquire Decart for around $6 billion. Israeli media reported that besides Anthropic, Elon Musk's SpaceX had also shown interest in acquiring Decart. Currently, representatives from both Anthropic and Decart have declined to comment, and it remains unclear whether the two companies will explore other forms of collaboration in the future.
Decart AI was founded by an Israeli team, with core products including the chip optimization platform DOS (Decart Optimization Stack) and "world models" Lucy and Oasis. DOS optimizes across-chip neural networks, reducing tuning cycles from months to weeks and increasing the processing speed of AI agents to over 1,600 tokens per second, about eight times the industry average. Lucy can modify objects in video frames in real-time and generate training data for robots, while Oasis can create three-dimensional simulated environments for logistics warehouses.
In May of this year, Decart announced the completion of a new round of financing of $300 million, with a post-money valuation close to $4 billion, led by Radical Ventures, with participation from Nvidia, Adobe Ventures, Toyota Ventures, and OpenAI co-founder Andrej Karpathy as an angel investor. The company has generated significant revenue by licensing DOS technology to cloud service providers and AI labs.
For Anthropic, the core consideration for acquiring Decart was believed to be related to improving its financial metrics ahead of its IPO: on June 1, Anthropic secretly submitted an IPO prospectus draft to the U.S. Securities and Exchange Commission (SEC), with Morgan Stanley, Goldman Sachs, and JPMorgan serving as lead underwriters. The Series H financing completed in May raised its valuation to approximately $96.5 billion. Market news indicates that Decart's chip utilization optimization technology can increase hardware utilization from the industry average of 40%-50% to over 80%, theoretically providing a computational power gain equivalent to the additional purchase of $10 billion to $15 billion in hardware without actual new purchases.
From a financial and industrial perspective, Anthropic's decision to abandon the acquisition occurs against the backdrop of its ongoing significant investments in computational infrastructure: over the past year, the company has signed cloud computing agreements worth $35 billion with Nvidia-supported cloud service provider Lambda, reached a $45 billion investment agreement with Nscale, signed a $50 billion agreement with Fluidstack, and established a $45 billion partnership with SpaceX, accumulating a massive scale of commitments related to computational power. Under such substantial infrastructure expenditure pressure, a $6 billion acquisition, even if it could enhance chip utilization, would need to be weighed against other options such as directly expanding computational power procurement. The company ultimately chose to abandon the acquisition, interpreted by the market as a preference for relatively controllable and easily explainable methods (building or leasing computational power) to improve operational efficiency rather than integrated mergers and acquisitions as it approaches the IPO window.
Additional information: It should be noted that specific financial metrics such as "the cost per comparison revenue decreased from $0.71 to $0.56" and "target gross margin of 77% by 2028" are primarily reported by some tech media and have not been officially confirmed by Anthropic or independently verified by authoritative financial media. Readers should consider them as market rumors rather than confirmed disclosure data.
Source: Public Information
ABAB AI Insight
Anthropic's historical behavior shows that the company has consistently relied on computational power partnerships rather than acquisitions for expansion: since its Series A financing in 2021 (valued at approximately $623 million), its growth has primarily depended on strategic investments from Amazon (holding about 21%, with a total investment of $33 billion) and Google (holding about 15%), as well as large-scale computational supply agreements with Lambda, Nscale, Fluidstack, and SpaceX, rather than acquiring companies themselves. The Decart deal, which was set to become its largest acquisition, was a clear deviation from this usual strategy, and now abandoning the acquisition can be seen as a return to its familiar expansion path.
From a financial perspective, rather than spending $6 billion to fully acquire Decart's chip efficiency technology, Anthropic could simply license Decart's existing DOS platform like other cloud service providers and AI labs, achieving similar results with much less capital commitment. For Decart, the $300 million financing led by Radical Ventures in May, with participation from Nvidia, Adobe Ventures, Toyota Ventures, and angel investor Andrej Karpathy, indicates that it is not short on funds. The investor lineup spans chip manufacturers and industrial capital, giving Decart the confidence to turn to other potential buyers like SpaceX or continue to develop independently, even if the acquisition talks with Anthropic fell through.
In the context of the industry, this reflects a common pattern in the current AI infrastructure boom: leading model companies like OpenAI, Anthropic, and Google are increasingly inclined to sign long-term computational supply agreements with large-scale cloud service providers like Lambda, CoreWeave, Nscale, and Fluidstack, rather than directly acquiring startups that develop infrastructure optimization software. This is similar to how cloud computing giants historically preferred to sign long-term supply agreements with Nvidia and TSMC rather than directly acquiring them. Decart itself spans two popular but still early-stage fields: chip efficiency tools and "world model" simulations, attracting several well-funded potential buyers (including SpaceX), which allows the company to hold out for a better offer rather than being locked into a single acquirer's bid.
Structurally, this is essentially a capital allocation decision close to the IPO window: the closer a company gets to a highly scrutinized IPO, the more its acquisition decisions will lean towards options that can clearly and cleanly improve investor-focused metrics (gross margin, revenue per unit of computational power) without introducing additional integration risks. A $6 billion acquisition, no matter how attractive the technology, would introduce a series of complex issues that would require additional explanations to public market analysts before the roadshow, such as merger integration, core team retention, and revenue recognition methods. In contrast, continuing to sign large-scale, standardized computational supply agreements (which investors are already familiar with through cases like OpenAI and Meta) is a "safer" and more easily understood signal—this is why, while this acquisition was shelved, the company's computational-related expenditure continued to expand.