Daily Briefing — May 26, 2026

UK companies are doing what companies always do when a new buzzword smells like money: slapping it on everything and hoping nobody looks too closely. PR firms are openly complaining that clients running basic automation tools (think scheduling software or simple rule-based systems) are demanding to be pitched to journalists as AI companies. One publicist described it as "yoga-level" stretching, which is a generous way of putting it.

This trick works until it doesn't. Reporters are already exhausted by it — one south London publicist noted you can practically hear the eye rolls when AI comes up in a pitch. That skepticism is going to spread fast, and when it does, the companies that built their entire brand identity around a label they never earned are going to look very silly very quickly.

For anyone working in tech, communications, or anything adjacent to AI product development, the signal-to-noise ratio is getting worse by the week. Real AI capabilities are getting buried under a pile of rebranded spreadsheets, which makes it harder for actual practitioners to be taken seriously, and harder for hiring managers, clients, and partners to tell the difference between someone who knows this stuff and someone who updated their LinkedIn after a YouTube binge.

The short version: Oracle signed a $300 billion deal with OpenAI last September, its stock shot up 36% in a single day, Larry Ellison briefly became the richest person on the planet, and then over the next ten months the whole thing unravelled to the tune of a 43% share price drop that erased all those gains. The credit default swap market, where sophisticated investors go to bet that a company might struggle to pay its debts, has also started moving against Oracle.

The broader point is about the "picks and shovels" thesis that everyone keeps trotting out as the sensible, boring, safe way to play the AI boom. The idea is that you don't bet on who wins the gold rush, you sell the shovels. Nvidia and Oracle are the two names most often dropped as examples. But Oracle's situation is a useful reminder that "infrastructure" is not automatically a hedge against hype — you still have to actually build the things, fund the things, and find enough paying customers to cover what are absolutely enormous capital commitments.

The AI infrastructure buildout is real, and so are the questions about whether demand will keep pace. When a company takes on significant debt to build data centers at a scale that depends on the AI boom continuing exactly as advertised, the downside scenario stops being theoretical pretty fast.

A post on Hacker News lays out a practical EU infrastructure stack for solo builders trying to validate an idea without lighting money on fire. The core argument is simple: no serious EU cloud provider offers permanent free compute, so stop chasing that myth. The sweet spot is Hetzner's CX33 at around €7 a month, which gets you 4 vCPUs, 8 GB of RAM, and 80 GB of disk. That is enough to run a full Django or Rails app with Postgres, Redis, and background workers — overkill for most side projects.

The point is more about discipline than infrastructure. A lot of would-be builders stall out because they spend weeks architecting a cloud setup that could handle a million users before they have even found ten. The post tells you to stop doing that. Get something running for less than the cost of a mediocre lunch and find out if anyone actually cares about your idea.