OpenAI and Broadcom Unveil ‘Jalapeño’ — Their First Custom AI Chip
OpenAI has revealed its first in-house AI inference chip, built with Broadcom and named — I’m not making this up — Jalapeño. The ASIC is purpose-built for LLM inference, which is the part of the AI pipeline that actually costs money at scale, so this is a serious strategic move to cut dependency on Nvidia and control their own silicon destiny. Whether it actually delivers competitive performance remains to be seen, but the name alone tells you these folks have a sense of humor about the arms race they’re running. Hot take: naming your first chip after a pepper is either very cool or the kind of thing that sounds great in a keynote and gets quietly rebranded in 18 months.
The Trump White House Is Over Anthropic CEO Dario Amodei
Wired is reporting that Dario Amodei has been quietly sidelined from high-stakes White House meetings, replaced by co-founder Tom Brown — with one official apparently calling Amodei a “weirdo.” In a town where the currency is access, getting swapped out for a different face is a significant demotion, and it comes as Anthropic is simultaneously pursuing its IPO and navigating its feud with the administration. This is what happens when you try to be the “responsible AI” company while also needing federal goodwill — at some point, someone at the table decides your vibe isn’t working for them. The irony of the safety-focused AI CEO being too weird for Washington is genuinely rich.
[I Met With China’s Top AI Experts. They’re Freaking Out, Too](https://www.wired.com/story/ai-arms-race-china-us-cooperation/
The AI arms race between the US and China has researchers on both sides quietly terrified of a “Chernobyl moment” — a catastrophic failure driven by competitive pressure overriding caution. This is a genuinely important piece because it punctures the comfortable narrative that AI risk anxiety is just a Western luxury or a regulatory talking point. When the people actually building the systems in both rival superpowers are privately scared, that’s a data point worth taking seriously. The question is whether that shared fear translates into any actual cooperation, or just two teams racing faster while nervously looking over their shoulders.
The $27 Million AI Proxy War Over Alex Bores Ends in a Draw
The Anthropic vs. OpenAI super PAC spending war in New York’s 12th Congressional District ended with Bores narrowly losing the Democratic primary — meaning $27 million in AI industry money moved the race just enough to blow it. So the net result of the most expensive local election intervention in recent memory is: the AI-industry-backed candidate lost, the backlash candidate barely won, and everyone looks bad. If the goal was to buy goodwill in Congress, both companies might want to ask for a refund. As proxy wars go, this one managed to unite voters against the people spending the money — not exactly a template for future influence operations.
Congresswoman Denies Staff Used AI to Write Defense Funding Amendment
Rep. Anna Paulina Luna is doing the congressional equivalent of “I was just holding it for a friend” — her staff used Claude for “spellcheck” on a defense bill amendment summary, but absolutely, definitely not for the actual legislation. Screenshots circulating on X suggested otherwise, which prompted the denial in all-caps: “NO Legislation is ever drafted with AI.” Whether or not you believe her, this story is notable because it will not be the last time we have this exact conversation, and the line between “AI assisted” and “AI drafted” is going to get harder to defend the more these tools get used in offices that run on words.
AI Was Supposed to Kill Engineering Jobs, But New Data Suggests They’re the Most Resilient
New data from SignalFire shows that engineers are actually making up a larger share of new hires even as AI dominates the layoff narrative — which cuts directly against the “AI will replace all the coders” argument that’s been running for three years. The most plausible read is that AI tools are making engineers more productive, not redundant, and that demand for people who can actually build and maintain AI systems is expanding faster than automation displaces them. This won’t be the final word, but for now the “robots taking your job” story is more complicated than the headlines — at least if your job involves building the robots.
Qualcomm Buys Chip Startup Modular for Nearly $4 Billion
Qualcomm is dropping nearly $4 billion on Modular, one of the more quietly important AI chip software startups working on making heterogeneous hardware actually usable for AI workloads. This is less about the hardware itself and more about the software layer that makes different chips play nicely together — which is a chokepoint that matters enormously as the industry diversifies away from Nvidia monoculture. Qualcomm buying this suggests they’re serious about competing in the AI infrastructure stack, not just the edge device market where they already live. At $4 billion for a chip software company, you can feel the desperation to own something defensible in the AI stack.
AI Researchers Continue to Leave Google for Its Rivals
Two more top researchers — Jonas Adler and Alexander Pritzel — are leaving Google for Anthropic, continuing a talent bleed that has already cost the company Noam Shazeer and John Jumper. Google invented much of the foundational architecture that powers modern AI, and it keeps watching the people who know how to advance it walk out the door to better-funded, more focused competitors. At some point this becomes less of a personnel problem and more of a cultural one — and culture is a lot harder to fix than compensation.
Bottom Line
The AI industry is simultaneously naming chips after peppers, losing Senate proxy wars, getting sidelined by the White House, and watching its researchers flee — all while both American and Chinese scientists privately wonder if the whole thing is going to Chernobyl on them.