The Morning Brief — April 19, 2026

Cerebras files for IPO as AI chip demand goes supernova Cerebras, the AI chip startup that already landed a $10B+ deal with OpenAI and a partnership with AWS, is going public. This is the kind of IPO filing that tells you exactly where the money thinks AI is heading: into the silicon, not the software. When the picks-and-shovels guys go public, you know the gold rush is real.

The RAM shortage could last until 2030 Nikkei Asia reports that DRAM manufacturers — Samsung, SK Hynix, Micron — will only meet about 60% of demand by end of 2027, with SK Group’s chairman warning shortages could persist until 2030. AI is eating memory faster than fabs can build it. The next time a model release is delayed, there’s a decent chance it’s not a safety review holding things up.

Anthropic and the Trump administration are apparently talking again Despite being flagged as a Pentagon supply-chain risk just weeks ago, Anthropic is back in conversations with senior Trump officials. The vehicle: Claude Mythos, the cybersecurity-focused model Anthropic has positioned as a national security asset. Politics, it turns out, thaws fast when there’s a defense contract on the table.

Tesla’s robotaxi rolls into Dallas and Houston After launching in Austin and going driverless in January, Tesla is now offering robotaxi service in two more Texas cities. The entire footprint is still Texas, which feels appropriate — everything about autonomous vehicles is bigger there. The question now is whether this is a product or a proof of concept that never leaves the Lone Star State.

AI vendors’ new security posture: “that’s not a bug, it’s the design” The Register’s editorial on AI vendor accountability is a must-read this week. The pattern: vendor says you need AI to fight AI threats, then when researchers find a flaw in that AI, vendor says it’s “working as intended.” Anthropic did this with the MCP vulnerability last week. It’s not a new trick, but it lands differently when the stakes are 200,000 compromisable servers.

The App Store is booming again, and AI is probably why Appfigures data shows a surge of new app launches in 2026, with AI tools likely driving the wave. The vibe-coding era has apparently produced enough working apps to show up in the numbers. Whether these are good apps or just “working” apps is, as yet, undetermined.

OpenAI’s Codex gets computer use, in-app browsing, image generation, and memory OpenAI just shipped a significant Codex update: the desktop app now handles computer use, can browse the web inside the app, generate images, and remember context across sessions. “Codex for (almost) everything” is the framing, which is doing some quiet work with that “(almost).” They’ll drop the parenthetical eventually.

GPT-Rosalind: OpenAI’s frontier reasoning model for life sciences OpenAI introduced a specialized model for drug discovery, genomics, and protein reasoning. Named after Rosalind Franklin — the crystallographer whose work on DNA structure was famously uncredited — which is either a meaningful tribute or an irony so thick you could choke on it, depending on your mood.

Bottom Line

When the chip makers are going public, the car companies are going driverless, and AI vendors are calling security flaws “features,” you’re not in the early innings anymore — you’re watching an industry sprint past the point of no return in real time.