Anthropic is about to spend $6B to fix its inference bottleneck. Your AI cost assumptions just died.

Anthropic is in talks to buy Nvidia-backed Decart AI for $6 billion.

Not to build a better model. To squeeze more compute out of every chip it already owns.

Decart raised $450M total, including a $300M round just three months ago at a $4B valuation. Anthropic is offering a 50% premium. That is not a negotiation. That is desperation.

The real story is why.

Anthropic filed a confidential S-1 in June. It is racing toward an IPO while demand for Claude is crushing its infrastructure. Every query requires inference capacity. Every efficiency gain at the chip level translates directly into lower costs, faster latency, and more customers served.

Decart builds inference optimization software and AI infrastructure tooling. Their team joins Anthropic's inference and performance organization if this closes.

This is the structural shift nobody is talking about.

The AI race just changed geometry. It is no longer about who trains the smartest model. It is about who serves it cheapest and fastest. That is an infrastructure problem, not a research problem.

If you are an enterprise running Claude, your pricing just got pegged to Anthropic's inference efficiency. If they buy Decart and it works, costs drop. If it does not close, costs stay high and Anthropic burns cash.

Audit your AI vendor contracts today. The companies winning the inference war will set the price floor for everyone else.

SOURCE: https://techstartups.com/2026/08/13/anthropic-in-talks-to-buy-nvidia-backed-ai-startup-decart-for-6-billion-ahead-of-mega-ipo/
VERIFIED: Bloomberg, Reuters, Calcalist
SIGNAL: Anthropic's largest acquisition attempt signals inference efficiency is becoming as valuable as model training. Enterprise AI costs are about to be reshaped by whoever wins this infrastructure war.