The infrastructure layer is consolidating faster than the applications built on top of it. Memory chip makers are printing money, Micron's profit jumped to $28.2 billion from $1.88 billion year-over-year, while inference chip startups like Cerebras are discovering that margin forecasts matter more than technical achievement. OpenAI and Broadcom's Jalapeño chip announcement, Qualcomm's $4 billion acquisition of Modular, and Meta's first Big Tech datacenter chip deal with Qualcomm all point to the same pattern: whoever controls the hardware layer controls the economics of AI deployment. The silicon race is not about innovation anymore. It is about who gets to extract rent from the compute everyone else needs.
Meanwhile, the talent market is inverting. Researchers are leaving Google for Anthropic and Anthropic is building enterprise products, not research papers. Engineering hiring is up while layoffs dominate headlines because companies are not replacing commodified roles, they are building teams to operate AI systems at scale. Entry-level workers now need senior-level judgment to land jobs because AI has collapsed the career ladder. Token rationing has replaced tokenmaxxing within months because consumption-based licensing turned infrastructure costs into line items that finance departments actually see. The memory chip shortage paid off for Micron and created margin pressure everywhere else, but it also forced companies to think about AI spending as a budget constraint, not an unlimited resource.
Regulation and geopolitics are reshaping who can build what. Europe is pushing back on Washington's chip war by noting that the MATCH Act would restrict machines a decade old, suggesting the real leverage lies in current-generation tools, not yesterday's hardware. Anthropic accused Alibaba of illicit access to Claude. The Trump White House replaced Anthropic's CEO Dario Amodei with Tom Brown at high-stakes meetings. Anthropic's Mythos model found vulnerabilities in classified US government systems, raising questions about who controls AI security research. These are not separate stories. They are the same story: AI companies are becoming infrastructure providers whose products touch national security, and governments are learning that access and control over models matter more than access to compute.
Sloane Duvall