- Anthropic is reportedly in early-stage talks, which could still fall through, to acquire Israeli AI startup Decart for approximately $6 billion, targeting its real-time video and simulated environment capabilities in products called Lucy and Oasis.
- Strategic M&A deal values rose 36% year-to-date in 2026 per Bain & Company, with acqui-hires allocating substantial deal value to retention packages, Google reportedly paid $2.4 billion primarily to secure engineering staff from Windsurf in July 2025.
- Strategic buyers in 2026 are requiring proven agentic deployment, clear data moats and net revenue retention above 110% before engaging, lab-stage demonstrations no longer qualify for premium valuations.
A reported $6 billion offer for Israeli startup Decart would be Anthropic’s largest acquisition to date, and it puts a concrete price on what specialised inference infrastructure is worth to a frontier lab right now.
Anthropic’s Infrastructure Play
The reported interest in Decart centres on three products: Lucy, a model for real-time video editing; Oasis, which generates simulated environments for robotics and autonomous driving systems; and DOS (Decart Optimization Stack), infrastructure and compiler tooling that squeezes more inference and training performance out of existing chips, which multiple reports describe as the actual driver of Anthropic’s interest. According to reports, discussions were ongoing as of August 13, 2026, though multiple outlets describe them as early-stage and note the deal could still fall apart. For Anthropic whose core business is large language model development, these are capabilities that would take significant time and resource to build in-house. Acquiring them directly addresses a gap in computing capacity and model efficiency at a moment when demand is pressing hard against both.
Why Buy Is Beating Build
Strategic M&A deal values rose 36% year-to-date as of mid-2026, according to Bain & Company’s M&A midyear report, part of a broader shift toward buying AI capability rather than building it in-house.
The logic is straightforward. As foundational models commoditise, the competitive advantage shifts to proprietary datasets, the pipelines that generate and maintain that data, and the workflows that make those systems hard to replicate. Building those structural advantages in-house takes years. Acquisition compresses that timeline, and in a market where acquisition targets are being identified earlier in their lifecycle buyers who move fast capture assets before valuations reset.
Proprietary Data and the Talent Scramble
What acquirers are actually buying is rarely the model itself. The premium sits in proprietary datasets, the pipelines that maintain them, and the vertical workflows that competitors starting from public data cannot close quickly. A company with years of transaction data in a defined sector carries a compounding structural advantage that is explicitly priced into deal terms.
Talent follows the same logic. In July 2025, Google reportedly paid $2.4 billion primarily to secure key engineering staff from AI code-generation startup Windsurf, according to reports, a figure that illustrates how much deal value in this cycle flows to retention packages rather than product assets. Acqui-hires are structured to keep critical personnel through integration, because the expertise walks out if the people do.
AI-Native Targets Command the Premium
Strategic buyers in 2026 are drawing a hard line between companies that bolt AI onto existing products and those built AI-native from the start. In an AI-native company, product architecture, data model and value proposition are all designed around AI from the ground up. That structural depth is what makes these targets difficult to replicate and why they attract acquisition interest over companies with more superficial integrations.
The market has also shifted its test from model capability to production evidence. Acquirers want agentic systems already generating measurable revenue with real customers, not lab-stage demonstrations. This mirrors the broader enterprise demand for proof of ROI rather than proof of concept, a tension well documented in recent board-level scrutiny of AI spend. Infrastructure, data provenance and operational reliability now matter more to acquirers than benchmark scores.
Valuations Tighten Around Production Proof
Overall tech M&A has moderated in 2026, but AI-specific deals remain active. Buyers are entering conversations with sharper criteria than they applied during the 2022 to 2025 capital surge: proven agentic deployment, clear data moats, demonstrated net revenue retention. Vertical AI companies solving defined problems in construction, legal or financial services are seeing consistent buyer interest. Those that can show strong net revenue retention and defensible data advantages are commanding the highest multiples in the lower middle market.
