Daily Briefing — April 20, 2026

The world is about to run short on RAM. According to Nikkei Asia, memory makers are only expected to meet 60 percent of demand by the end of 2027. The SK Group chairman has floated the possibility that shortages persist all the way to 2030.

To keep pace with demand, production would need to grow by 12 percent a year across 2026 and 2027. What is actually planned, according to Counterpoint Research, is 7.5 percent. Samsung, SK Hynix, and Micron are all building new fabs, but almost none of that capacity comes online before 2027 at the earliest. SK opened one facility in Cheongju back in February, and that is it for new production among the big three this year.

The reason demand is so brutal right now is AI. Every large language model, every inference cluster, every data centre buildout is consuming memory at a pace nobody fully anticipated two years ago. So while consumer tech feels the squeeze, the real pressure point is enterprise and cloud infrastructure.

Anthropic just launched Claude Design, a new product under the Anthropic Labs umbrella that lets you generate and iterate on visual work like prototypes, slides, one-pagers, and polished designs through conversation. It runs on Claude Opus 4.7 and is currently in research preview for Pro, Max, Team, and Enterprise subscribers. So if you're already paying for Claude, you may already have access or will soon.

Claude reads your codebase and existing design files to build a working design system for your team automatically. After that, every project inherits your brand's colors, typography, and components without you lifting a finger. That's a genuine workflow shift for teams where design resources are stretched thin.

For non-designers specifically, this is the gap that's been quietly costing people. Founders pitching investors, PMs trying to communicate product ideas, marketers needing something fast but not embarrassing have all had to either hire out, beg a designer, or ship something rough. That middle option is now a lot more accessible.

Vibe coding is exactly what it sounds like. You describe what you want in plain English, and an AI spits out working software. No computer science degree required, no Stack Overflow rabbit holes, no waiting on a developer who has seventeen other tickets ahead of yours. Andrej Karpathy named it back in February 2025 and the concept spread fast, because of course it did. It feels like a superpower.

Here is the part nobody puts in the demo reel. When your colleague in marketing spins up a little internal tool using Claude or ChatGPT, that code does not arrive with a certificate of origin. The AI assembled it from patterns across the entire internet, which includes brilliant open source work, yes, but also code written by people whose intentions you would not want anywhere near your company's systems. Your colleague has no way to know the difference, and honestly neither does the AI. It just pattern matched its way to something that runs.

Anthropic just handed British banks access to Mythos, its newest Claude model, and the timing is notable because the company itself flagged this thing as too dangerous to release to the general public. Anthropic saying out loud that the model can find vulnerabilities in IT systems at a level they have not seen before. UK financial institutions are getting it within days, according to Anthropic's UK head Pip White.

The rollout so far has been deliberately tight. Amazon, Apple, and Microsoft were in the first wave, all US based, all with the kind of security infrastructure and legal teams that can absorb the liability questions that come with this territory. UK banks are next, which tells you something about where financial services sits in Anthropic's expansion strategy.