The story today is electricity, not silicon

The most telling news of the day isn't a chip or a model — it's a utility bill. According to the-decoder.com, AI's growing energy appetite is pushing Nvidia and Amazon to put billions into the power that runs artificial intelligence rather than into the hardware or software layers everyone talks about. When the two companies most identified with the AI boom start buying into generation, it's a signal that compute scarcity is quietly becoming energy scarcity.

That abstraction has a very concrete cost in Emporia, Kansas. Tom's Hardware reports the town has moved its city council meetings online after death threats against city leaders intensified over a proposed gigawatt-scale AI data center. The move had a side effect that may outlast the threats: it ended public comment at those meetings. A community fight over where AI's power goes has now produced a community with less say in the outcome — a pattern worth watching as more towns face the same proposal.

The upside case for what infrastructure can do to a place is visible in London. TechCrunch traces the remaking of King's Cross, a notorious red-light district two decades ago and today a home for frontier technology. It's a reminder that these transformations take twenty years, not two — and that Emporia's residents are arguing about a version of that timeline.

Robots are a compute story before they're a robot story

Barclays analysts, cited by Yahoo Finance, argue that the next wave of AI chip and data centre demand could come from humanoid robots — machines that haven't shipped yet. It's a bullish read that folds neatly into the energy story above: every embodied robot is a claim on compute somewhere else.

What those robots will actually do first is less glamorous. Business Insider reports that several heavily funded robotics startups have converged on the same opening task: folding laundry. Towels aren't the kind of work that usually attracts serious venture money, but they're an honest benchmark — deformable, unpredictable, and impossible to fake with a demo reel.

At the other end of the ambition scale, Interesting Engineering reports China plans to integrate four-legged robots into the International Lunar Research Station, its future Moon base, by 2035. Robot dogs on the lunar surface and robot arms on the laundry pile are the same bet placed at opposite extremes.

Anthropic hands Claude Code more of the wheel

Two changes landed for Anthropic's command-line coding assistant, and together they point the same direction. Claude Code version 2.1.224 lets separate sessions send messages directly to one another, per a report surfaced via Bing News — agents coordinating with agents rather than routing everything through a human. And starting August 14, a Techmeme summary reports, auto mode becomes the default for Pro, Max, and Team users. Defaults are the strongest product opinion a company can express; this one says the tool should mostly run itself.

Deals, payouts, and one very odd checkout page

OpenAI has acquired NextSlide, a startup focused on presentation software, TechCrunch reported on August 8. The most concrete detail is where the team went: they're already working on ChatGPT.

X is retiring the payout system that has defined creator monetization on the platform. TechCrunch reports the company is winding down Revenue Sharing in favor of a new program called Original Content Rewards. Whatever the mechanics, a change in how a platform pays is a change in what it wants made.

And in Taiwan, GPU scarcity has taken a farcical turn. Tom's Hardware reports that the ecommerce platform PChome24h has been listing Nvidia's flagship RTX 5090 bundled with other hardware — in one case, eight motherboards. Buy the card, take the pile.

Countermeasures

Finally, a small story with long implications: TechCrunch reports a security researcher has built an algorithm that generates adversarial patterns designed to make people, faces, and vehicles invisible to surveillance cameras. It's the mirror image of everything above — as the industry pours capital into building bigger machine perception, someone has automated the production of blind spots.

The through-line

Read together, the day says AI's constraints have moved outward. Not to the model, but to the grid that feeds it, the town council that has to approve it, the robot that has to earn its keep on a laundry pile, and the retail buyer who can't get a card without eight motherboards. The hard parts are increasingly physical and political rather than algorithmic.