June 29, 2026
California put Claude on every state desk. Colorado switched its AI law off.
In one week California put Claude on every state desk at a 50% discount, Colorado's first-in-the-nation AI discrimination law had its core duties cut and its start pushed to 2027, OpenAI and Broadcom unveiled the Jalapeño inference chip, Menlo Ventures raised a record $3B fund on its Anthropic stake, and a Nature Medicine study found general models beat FDA-cleared clinical tools.
In one week, AI stopped being a demo and started being infrastructure. A venture firm raised $3 billion to back it. OpenAI taped out its own chip to run it cheaper. California put it on every state worker's desk. And a medical study found the general models now beat the tools regulators already cleared. Meanwhile the one state law built to police any of this quietly switched off. Here is where the money is.
TECH: AI in production
OpenAI now makes its own silicon, and it took nine months.
On June 24, OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom chip, built for one job: running large language models cheaply. They went from design to manufacturing tape-out in about nine months, which they call the fastest cycle ever for a chip this advanced. Early samples run in the lab, with performance per watt they say beats today's best. First deployment is aimed at late 2026.
The point for builders is simple. When the biggest lab controls its own inference silicon, the cost of every AI feature you ship keeps falling.
The money play: Do not lock long vendor contracts at today's token prices. Architect your product so a 50% drop in inference cost lands in your margin, not your provider's.
Sources: OpenAI | TechCrunch
BIZ: build, fund, scale
One bet on Anthropic just printed a $3 billion fund.
On June 23, Menlo Ventures closed $3 billion, the largest raise in its 50-year history. The reason is a single position: Menlo's stake in Anthropic, now worth nearly $14 billion. Menlo and Anthropic also run Anthology, a fund that has backed more than 60 startups building on Claude.
Capital is pooling around one ecosystem. That is good if you are inside it and a headwind if you are not.
The money play: If you are raising, build something real on Claude and apply to Anthology before you pitch generalist funds. Ecosystem money moves faster and asks fewer questions than cold outbound.
Sources: TechCrunch | Menlo Ventures
LAW: rights and fights
The first broad US state AI law was set to switch on June 30. It won't.
Colorado's AI Act was the first state law in the country to make companies answer for algorithmic discrimination. It was due to take effect June 30, 2026. In May, Colorado enacted SB 189, which cut the core duty of care, dropped mandatory impact assessments, and pushed the start to January 1, 2027. What survives is mostly disclosure. The rules meant to give people a claim against biased AI decisions got hollowed out before they ever bit.
This is the pattern everywhere. States and courts keep moving the line on what AI can use and who is liable.
The money play: Do not hard-code compliance to one state. Log model inputs, outputs, and decisions by default so you can prove fairness under whichever regime wins.
Sources: Colorado General Assembly | Hunton
POL: policy, tax, borders
California just put Claude on every state desk at half price.
On June 29, Governor Newsom announced a first-of-its-kind deal: Anthropic's Claude is available to all California state agencies, plus any city or county that opts in, at a 50% discount, with free worker training and technical help. Newsom framed it as help, not replacement. AI, he said, "should help our workers move faster, solve problems more effectively, and deliver better results for Californians." It is the largest state government AI deal in the country.
The money play: Governments are now buying AI in bulk. If you sell software, build a public-sector tier with a real discount and procurement-ready docs. That is where new budget is opening in 2026.
Sources: Office of the Governor | TechCrunch
SCI: breakthroughs to build on
General AI models now beat the cleared medical tools.
A study published in Nature Medicine on June 23 tested three general models, OpenAI's GPT-5.2, Google's Gemini 3.1 Pro, and Anthropic's Claude Opus 4.6, against two FDA-cleared clinical tools, OpenEvidence and UpToDate Expert AI, on real questions from practicing doctors. The general models won on every benchmark tested. The catch is that no one has a settled way to validate them for the messy questions that show up at the bedside.
Read the lesson past medicine. A frontier model plus good context often beats a narrow, purpose-built tool. Your edge is the workflow and the data, not the model.
The money play: If you are building a vertical AI product, stop fine-tuning a bespoke model first. Wrap a frontier model, spend your effort on validation and workflow, and be clear about who owns and answers for the output.
Sources: Nature Medicine | Clinical Trial Vanguard
Before you go
Five stories, one theme: the tools are sprinting and the rules are jogging. That gap is where builders make money and where sloppy builders get sued, so we track both here every week. Kobe, my dog, remains unimpressed by all of it and is still holding out for the model that opens the treat drawer. More at nomadtechnologist.com.
Not legal, financial, or tax advice.
Get this every weekday.
Free. 5 minutes. Unsubscribe whenever.