May 25, 2026
Models got cheaper. The pressure got heavier.
Google shipped a faster, cheaper frontier model at I/O while tech kept cutting jobs for AI (and some employers began rehiring), a nine-member jury sank Musk's OpenAI suit on timing, and the EU's general-purpose AI fines go live August 2, 2026.
Two things are happening at once. A frontier AI model got faster and cheaper, and companies cut thousands more jobs to pay for AI. That is the whole market in one picture. Here are the five stories that move your work.
TECH: AI in production
Google put a frontier model in front of everyone on May 19, and it undercut the field on price.
At Google I/O, Sundar Pichai announced Gemini 3.5 Flash, live that day across Google products and the API. Google says it runs four times faster than other frontier models and costs less than half of comparable ones. Google says it is already processing more than 3 trillion tokens a day. A bigger Gemini 3.5 Pro was promised for the following month.
Speed and price are the story. When output gets four times faster and cheaper, work that was too slow or too costly last quarter becomes worth building this quarter.
The money play: Re-run your unit economics. Take one AI feature you shelved because tokens cost too much, price it again at half, and if it clears now, build the smallest version this week.
Sources: Google I/O keynote | Google Developers
BIZ: build, fund, scale
Meta began cutting about 8,000 jobs, roughly 10% of staff, to free up cash for AI.
The cuts are part of a year where tech has shed more than 160,000 roles, with AI the most cited reason. Some of these bets are already going sideways. By July, several employers who blamed AI for layoffs had started rehiring for the same work.
The money play: Do not fire before you verify. Run the AI system next to the human process for one full cycle. Cut only what it actually replaces, measured, not promised.
Sources: Tech layoffs tracker | CNBC
LAW: rights and fights
A jury threw out Elon Musk's case against OpenAI in under two hours.
On May 18, a nine-member jury in California found Musk waited too long to sue over OpenAI's shift from nonprofit to for-profit. Judge Yvonne Gonzalez Rogers agreed and tossed the case on the statute of limitations. Musk says he will appeal. The court never ruled on whether the conduct was wrong, only that the clock had run out.
The money play: Diary your deadlines. If a partner or vendor wrongs you, the limitation clock usually starts when you first knew, not when you finally act. Write down the date you learn of a problem the day you learn it.
POL: policy, tax, borders
The EU's rules for general-purpose AI models get real teeth on August 2, 2026.
These rules are already law. From August 2 the European Commission can fine providers of general-purpose AI models up to 15 million euros or 3% of global yearly revenue, whichever is greater. Providers must also publish a summary of the data used to train their models. If you ship AI into the EU, the paperwork is now part of the product.
Founders who move across borders have a second clock to watch: your own tax residency.
The money play: Map your exposure now. List which of your models or features count as general-purpose, and start the training-data summary before the August date. If you operate from more than one country, pin down your tax residency on the same timeline.
Sources: European Commission | EU AI Act Article 101
SCI: breakthroughs to build on
AI for real science stopped being a demo and started shipping tools.
Alongside I/O, Google unveiled Gemini for Science and put 10 million dollars into five universities to pair AI with quantum work in the life sciences. Its AlphaEvolve system is now aimed at supply chains, chip design, molecular systems, and power grids. The pattern to notice: general models are being wrapped into narrow, checkable tools for one field at a time.
The money play: Copy the pattern, not the budget. Pick one repeatable task in your niche, wrap a model in a tight checklist that verifies its output, and you have something a general chatbot cannot match.
Sources: Google AI updates | Google I/O
Before you go
The through-line: models got cheaper, and the pressure to use them got heavier. Cheaper tools only help if you point them at real work. That is the whole job over at nomadtechnologist.com. Kobe, my dog, verified none of this and slept through all of it. He still called the AI layoff reversals before the CEOs did.
Not legal, financial, or tax advice.
Get this every weekday.
Free. 5 minutes. Unsubscribe whenever.