Google Harvests Your Photos While OpenAI Designs a 10GW Brain

The Consent Problem Is Now a Default Setting

Google is storing the images you drop into Search and using them to train AI models. The opt-out exists, buried where opt-outs live, which is to say nowhere you would think to look. This is the new shape of the data deal. You are not asked. You are informed, eventually, by Wired.

The pattern repeats across the slot. Anthropic is accusing Alibaba of spinning up fake accounts to extract Claude's capabilities, a kind of industrial espionage by chatbot. If true, it confirms what everyone already suspected, that frontier model weights and behaviors are the new strategic minerals and that the front lines run through signup forms. Anthropic's complaint also quietly admits that defending a model from determined scraping is roughly as feasible as defending a sandcastle from the tide.

Meanwhile in London, the Met Police got a twelve-month extension on its Palantir pilot while it shops for a permanent vendor. Pilots that get extended do not end. They become procurement. The Guardian's framing is polite. The reality is that predictive policing infrastructure is being installed under the legal cover of a trial that the public was never really invited to evaluate.

The Power Bill Comes Due

OpenAI is designing a custom chip that would draw ten gigawatts. That is not a data center number. That is a small-nation number. The New York Times reports it as engineering ambition. Read it as a forecast of the grid fights, water fights, and zoning fights coming to a county near you. Every frontier lab is now in the business of negotiating with utilities, and utilities are in the business of passing costs to ratepayers who did not ask for a chatbot.

The chip economy underneath all this is having its moment. Micron's profits are up fifteenfold as memory becomes the bottleneck nobody can build out of fast enough. Qualcomm just landed Meta as its first major data center customer and the stock jumped fifteen percent on the news. The semiconductor industry spent two years apologizing for the shortage. Now it is taking a victory lap and pricing accordingly. If you thought the AI boom was about software, the balance sheets disagree.

Hollywood, Crime, and the Cleanup Crew

Google DeepMind put seventy-five million dollars into A24, which until yesterday was the indie film studio that indie film fans were allowed to like without guilt. The backlash was immediate and predictable. The deal itself is smaller than a rounding error in OpenAI's chip plans, but it signals the direction. Every cultural institution with a recognizable brand is getting an AI partner whether the audience wants one or not. The pitch is always tools for filmmakers. The result is usually a licensing agreement nobody can read.

The one piece of good news comes from law enforcement, which managed to coordinate a global takedown of two widely used cybercrime tools under Operation Endgame. It is genuinely useful work and it will be undone within months, because the criminal supply chain is more resilient than the legitimate one. Still, credit where it is due. Someone is cleaning up while everyone else is building out.

What the Slot Adds Up To

The through line today is that the AI buildout is no longer a series of product launches. It is an infrastructure project running on default consent, contested data, and electricity nobody has budgeted for. Google is harvesting quietly. OpenAI is planning loudly. Palantir is settling in. Anthropic and Alibaba are litigating the boundaries of what counts as theft when the product is a probability distribution. The chipmakers are getting paid either way.

None of this is a single catastrophe. It is a series of normal Tuesday decisions that will be very hard to reverse by the weekend. The opt-outs are theoretical. The trials are permanent. The gigawatts are real. File this one under accelerating normal, which is the worst kind of normal, because by the time it feels abnormal the defaults have already shipped.