Anthropic is currently in negotiations to acquire Decart AI for approximately $6 billion, a move that would represent the largest takeover in the history of the Claude developer. The Arabian Post reported that this potential transaction reflects a strategic pivot toward controlling the underlying infrastructure layer to mitigate the extreme capital expenditure required to operate frontier artificial intelligence models.
Decart AI, founded by brothers Dean and Orian Leitersdorf, specializes in software designed to extract maximum performance from existing hardware accelerators. By optimizing how models interact with silicon, the company aims to increase the volume of user requests handled per unit of compute power.
The firm has developed the Decart Optimization Stack, or DOS, which functions across diverse hardware architectures including Nvidia GPUs, Google TPUs, and Amazon Trainium chips. This hardware-agnostic approach provides a significant degree of flexibility for large-scale deployments, allowing developers to shift workloads between different processor types without rewriting core code.
Performance benchmarks released by Decart suggest that its DOS 2.0 software can process over 1,600 tokens per second for agent-based workloads. While these figures remain unverified by independent audits, they highlight the company’s focus on the critical bottleneck of inference speed.
The firm also claims its systems can manage high-definition world-model workloads at rates reaching 100 frames per second. These capabilities demonstrate potential applications in robotics and autonomous simulation, areas where real-time data processing is essential for operational success.
Decart’s infrastructure software addresses the fundamental challenge of memory management and kernel execution, which often limits the efficiency of large language models. By optimizing these low-level processes, the company provides a pathway to higher utilization rates for expensive data center assets.
The financial structure of the deal remains fluid, and discussions could still conclude without an agreement. Decart recently secured $300 million in a funding round led by Radical Ventures in May, which valued the company at roughly $4 billion.
A $6 billion acquisition would provide a significant premium for investors, including Nvidia, who participated in the earlier financing efforts. This valuation reflects the high market demand for specialized software that can squeeze more utility out of expensive, scarce hardware.
Anthropic currently faces intense pressure to improve its gross margins, which remain significantly lower than those of traditional software enterprises. Investors are monitoring the company’s path toward a projected 77 per cent margin by 2028, a target that depends heavily on reducing the unit cost of inference.
Bringing Decart’s optimization expertise in-house would allow Anthropic to refine its own inference and performance operations directly. This move addresses the broader industry challenge of hardware dependency by moving optimization closer to the hardware layer—specifically targeting kernels, memory management, and model execution.
The acquisition would enable Anthropic to reduce its reliance on constant hardware procurement. This strategy aligns with the company’s broader efforts to demonstrate long-term profitability ahead of a potential stock-market listing, proving that management views computing efficiency as central to its financial profile.
The competitive landscape for frontier AI laboratories is shifting from a focus on raw model capability to the economics of deployment. As inference costs become a primary differentiator, companies are increasingly prioritizing the efficiency of their compute clusters over simple model scaling.
Anthropic’s interest in Decart underscores the necessity of mastering the software-hardware interface to maintain a sustainable business model. Future developments will likely hinge on whether Anthropic can successfully integrate Decart’s engineering team into its existing infrastructure pipeline.
If the deal proceeds, the focus will shift to whether the promised efficiency gains can be realized at the scale of Claude’s massive user base. Observers will be watching for any signs of improved utilization rates in Anthropic’s future financial disclosures and technical updates.