Saturday, April 25
AI's Money, Muscle, and Limits
Google is committing up to $40 billion (opens in a new tab) to Anthropic in cash and compute, a figure that makes every other AI investment this cycle look modest and signals that the search giant views Anthropic as its primary hedge against OpenAI. That bet lands the same week Cohere acquires a Germany-based startup (opens in a new tab) to build what it calls a transatlantic AI powerhouse, positioning itself as the enterprise alternative for non-American customers wary of US provider lock-in. The capital concentration is striking when measured against performance reality: GPT-5.5, marketed as OpenAI's strongest agentic coding model ever (opens in a new tab), is posting embarrassing results on LiveBench, raising pointed questions about whether benchmark-driven marketing has fully decoupled from actual capability.
Palantir is having an unusually expansive week (opens in a new tab) inside the US federal government: the company has signed a $300 million deal with the USDA to manage food supply data and is simultaneously reported to be assisting the IRS in investigating financial crimes (opens in a new tab), giving it a footprint across agriculture, taxation, and — through Project Maven — military targeting. That breadth of government dependency on a single commercial AI vendor is precisely what critics warned about, and it now looks less like a warning and more like a description of the present. Meanwhile, Utah's medical board has called for the immediate suspension (opens in a new tab) of the state's AI doctor experiment, a sharp regulatory rebuke that suggests not every agency is moving at Palantir's pace.