The week's AI headlines reveal a fracturing consensus: the companies that built the technology are now fighting over who controls it, while the infrastructure beneath the entire enterprise, power, data, regulatory authority, becomes the actual prize.
Start with the political rupture. Anthropic submitted sworn declarations to a federal court late Friday pushing back against Pentagon allegations that it poses an "unacceptable risk to national security," claiming the government's case rests on technical misunderstandings. This came a week after Trump declared the relationship kaput. Simultaneously, the Trump administration released its National Policy Framework for Artificial Intelligence, which reads as a federal preemption playbook designed to undercut state regulations and shift child safety burden to parents rather than platforms. The framework emphasizes lighter-touch rules for tech companies. Yet Big Tech is fracturing over the attacks on Anthropic, with former Trump allies now offering unprecedented criticism. The administration is trying to block state-level laws while facing fierce pressure to offer a federal rule book. What this reveals is not philosophical disagreement about AI safety but a power struggle over who gets to write the rules: the executive branch, states, or the companies themselves. Trump's framework isn't about safety. It's about centralized control.
Beneath the regulatory theater, the real bottleneck is infrastructure. Nvidia's CEO Jensen Huang declared that every company needs an "OpenClaw strategy" and projected one trillion dollars in AI chip sales through 2027. But power has become the biggest constraint on rolling out new AI data centers, creating openings for energy tech investors. Meanwhile, the industry is quietly optimizing away at the margins: engineers are cutting training costs through "toggle-away efficiencies" inside the training loop, and recommendation system engineers are discovering that energy consumption is now a north star metric alongside accuracy and engagement. Microsoft rolled back Copilot bloat on Windows, Photos, Widgets, Notepad, after user and developer resistance to forced AI integration. OpenAI is reportedly folding ChatGPT, Codex, and its browser Atlas into a single desktop superapp, signaling a shift toward enterprise and developer audiences away from the consumer market that made it a household name. These aren't product refinements. They're admissions that the consumer AI wave has peaked and the real money is in enterprise infrastructure and developer tools.
The final pattern is distribution lock-in disguised as convenience. WordPress.com now lets AI agents write and publish posts directly. Google introduced AI into its Stitch UI design tool, enabling developers to describe interfaces in natural language. Amazon is building a smartphone called Transformer to serve as a central hub integrating shopping, streaming, and voice-controlled services through Alexa. LinkedIn banned an AI agent that had conquered the platform. Each move lowers friction for adoption while raising switching costs and centralizing control. PwC told staff they must embrace AI or face replacement. Google told researchers to stop submitting AI-generated bug reports to its open-source vulnerability program due to hallucinations and low quality. The pattern is clear: AI adoption is no longer optional, quality control is breaking down at scale, and the winners will be whoever controls the platform through which work flows.
Sloane Duvall