The infrastructure wars have shifted from hardware to control over what gets built on it. Google is flooding the market with chips, agents, and integrations across every surface of work, Workspace, Chrome, Maps, Gmail, betting that owning the plumbing means owning the workflow. Anthropic is rationing Claude Code on its Pro plan because demand has become unmanageable, a constraint that pushes users toward paid tiers or alternatives. SpaceX just paid a $10 billion collaboration fee to Cursor and secured a $60 billion acquisition option, a move that telegraphs something clearer than any press release: the real money is not in building models but in controlling the interface between models and actual work. Tesla is burning $25 billion in capex this year alone, betting its entire future on in-house AI for vehicles and manufacturing. The pattern is unmistakable. The companies winning are not the ones with the best research papers. They are the ones with distribution, capital, and the ability to embed AI so deep into existing systems that switching costs become prohibitive.
Underneath this is a brutal sorting of power. High-earning workers are adopting AI tools at triple the rate of lower earners, according to an FT survey, which means AI is not flattening hierarchies, it is reinforcing them. Meta is tracking employee keystrokes and screen activity to train workplace agents, a move that reveals the actual transaction: workers surrender behavioral data for automation that may ultimately replace them. Anthropic's Claude Mythos found 271 vulnerabilities in Firefox, a demonstration of AI's security capabilities that is also a demonstration of whose AI gets trusted with critical infrastructure. Google Cloud is cutting deals with Mira Murati's Thinking Machines Lab for multibillion-dollar infrastructure commitments, and Infosys is packaging OpenAI's tools for enterprise clients, the consulting layer is thickening, which means the actual leverage is consolidating upward. North Korean hackers are using AI to scale their social engineering and malware, which tells you that capability diffusion is not a feature of AI adoption, it is a guarantee.
The regulatory response has been to propose more frameworks, more governance, more committees. Senator Josh Hawley is pushing Republicans to reject $300 million in AI lobby funding, which is real pressure but not enough to stop what is already in motion. Private equity is courting OpenAI and Anthropic for joint ventures to deploy AI in businesses, which means the model is becoming standardized, financialized, and folded into the existing machinery of capital allocation. By the end of 2025, half of all companies had deployed AI in at least three business functions. The question is no longer whether AI will be adopted. The question is who controls the terms of adoption, who captures the margin, and who bears the cost when the system breaks.
Sloane Duvall