The week's signal is unmistakable: AI infrastructure and the companies that control it are consolidating power faster than regulation can track, while the actual use cases that matter are shifting from models to agents. The friction points reveal where money and leverage are concentrating.
Start with what's not in the headlines. Notion is killing its email product because users have already moved on to agents managing their inboxes. Patronus AI landed $50 million because companies are desperate to test agents before deploying them. General Intuition raised $320 million betting that video game data can train agents that behave like humans. These aren't feature announcements. They're evidence that the market has moved past the model wars. Claude is gaining paid users against ChatGPT's market lead not because it's incrementally better but because the product experience around agents matters more than model size. Meanwhile, the Trump administration is asking OpenAI to slow-roll GPT 5.6, which means government now sees model releases as infrastructure decisions requiring vetting. That's a signal that the regulatory frame is shifting from "should AI exist" to "who controls distribution."
The real leverage is hardware. Apple is raising prices across all products except the iPhone because AI server demand has created a memory shortage. IBM claims sub-1 nanometer transistors. Amazon is investing $13 billion more in India's AI infrastructure. Alibaba was caught using 25,000 fake accounts to extract Claude's capabilities across 28.8 million exchanges, which Anthropic escalated directly to the Senate Banking Committee. That's not corporate tattling. That's a company signaling that model theft is now a national security concern worthy of government attention. The Linux Foundation is launching an Agent Name Service framework to establish identity and trust for AI agents because enterprises deploying agents across organizations need to know who owns what. This is the infrastructure layer materializing in real time. The companies selling memory, compute, and trust frameworks are the ones printing money. The companies selling models are fighting over scraps while managing state pressure, IP theft, and the need to prove their agents work better than open-source alternatives.
Sloane Duvall