Anthropic targets Decart AI acquisition for $6 billion to cut costs

The acquisition of Decart AI signals a shift from pure model development to infrastructure optimization as AI labs face capacity constraints and rising data center costs. By integrating world models and chip-level efficiency software, Anthropic aims to serve more users on existing hardware, a critical necessity as OpenAI initiates an 80% price cut on new models. This consolidation reflects a broader industry trend where American labs must aggressively lower operating expenses to compete with low-cost Chinese alternatives. Investors should monitor Anthropic's official S-1 filing to verify if realized revenue and cash burn rates align with the aggressive $120 billion annualized revenue projections touted by insiders.

Anthropic has been aggressively building its infrastructure stack ahead of a planned public offering. In early 2025, the company raised $3.5 billion in a Series E round led by Lightspeed Venture Partners, valuing the Claude maker at $61.5 billion and giving it one of the largest war chests among private AI labs. That capital has funded rapid expansion of compute partnerships, including a multi-year deal with Amazon Web Services that gives Anthropic access to custom Trainium chips. Decart AI, the Israeli startup at the center of the acquisition talks, built its reputation on real-time world models, most notably Oasis, a fully AI-generated interactive game engine that renders video frames on the fly without traditional game engines. Decart raised $32 million in a Series A led by Benchmark in November 2024, with investors citing the company's ability to generate interactive environments at low latency as a key differentiator.

The competitive landscape around AI lab consolidation has intensified as companies race to reduce inference costs before public market scrutiny arrives. OpenAI completed its restructuring into a for-profit entity in late 2025, clearing a path toward its own eventual IPO, while Google DeepMind merged its Gemini and DeepMind teams under a single leadership structure in early 2025 to accelerate model development and reduce redundant compute spend. Meanwhile, Chinese labs continue to pressure Western competitors on cost. DeepSeek's R1 model, released in January 2025, demonstrated competitive reasoning performance at a fraction of the training cost of frontier Western models, prompting a sharp selloff in AI-linked equities and renewed urgency around efficiency. Anthropic's pursuit of Decart AI fits squarely into this cost-reduction imperative, as world-model inference optimization could lower per-token serving costs for Claude across enterprise and consumer workloads.

On the technical side, Decart AI's Oasis engine demonstrated that transformer-based world models can generate interactive video at 20 frames per second on a single NVIDIA H100 GPU, a result that the company published alongside its open-source release in late 2024. That efficiency benchmark is notable because traditional video generation models typically require clusters of GPUs for real-time output. For Anthropic, integrating this capability could extend Claude's multimodal offerings into interactive simulation and synthetic environment generation, areas where OpenAI has also invested through its Sora video model. OpenAI launched Sora as a standalone product in December 2024, signaling that video generation is becoming a core product category for frontier labs rather than a research demo. The combination of Anthropic's language model scale with Decart's real-time world-model inference could position the merged entity to compete across both conversational AI and generative video workloads heading into a public listing.