Welcome back! The AI you use today lives in someone else's data center and charges you by the token. Nvidia's new chip wants to drag that whole brain onto your laptop and end the monthly bill, which is a bigger deal than it sounds. The same hunger for compute is driving the rest of the day. SoftBank is spending $81 billion to turn French nuclear power into AI data centers, and OpenAI is putting its locked biology model in government hands. Ford, meanwhile, became this month's hottest AI stock without earning a cent from the business behind the hype.
In today's Generative AI Newsletter:
Nvidia: What changes when AI stops living in the cloud and moves inside your laptop?
OpenAI: Why is OpenAI giving governments free access to a biology model most people can't touch?
SoftBank: What is France really selling in Europe's biggest AI buildout?
Ford: How did a 122-year-old carmaker become the month's hottest AI trade?
Every time you use AI today, you are renting it. Your words travel to a data center, get processed there, travel back, and you pay for the privilege one token at a time.
Nvidia wants to end that by putting the whole AI brain inside your computer.
At Computex in Taiwan, Jensen Huang held up the RTX Spark, a chip built to run AI on your own machine, and said it will change the PC as much as the smartphone changed the phone.
It pairs a 20-core Arm CPU with a Blackwell GPU and 128GB of unified memory, with context up to a million tokens.
It runs Windows on Arm and was built with MediaTek, putting Nvidia up against Intel, AMD, Qualcomm and Apple in PCs for the first time.
Dell, HP, Lenovo, Microsoft, Asus and MSI ship RTX Spark machines this fall, with Acer and Gigabyte to follow.
In plain terms, a powerful AI model that used to run only in the cloud can now live on your laptop, answer instantly, keep your files private and never run up a cloud bill.
Every major PC maker lined up behind one chip, which tells you they believe the next computer is one that runs its own AI. The open question is whether people buy a pricey new laptop when the cloud version already runs on the machine they own.
Special highlight from our network
That's the gap between how we think and how search systems work. (They match words, that’s it.)
AI is changing that, and honestly, it changes a lot about how we find information.
MongoDB built a free course that covers embeddings, how to store and search them, and reranking and optimization techniques like chunking. It's the practical foundation for understanding how meaning-based search works.
With it, your search will never fail you again.
G42 kicked off its Agentathon last week, and the software running it is our Hackathon app, the same platform Oracle and Red Bull Racing have used for theirs.
The app is already pulling in revenue and users, and our community can now own a piece of it along with the rest of what we are building.
→ We grew our revenue 2.4X in less than a year
→ Got tens of thousands of users first month
And now, this can be your success story too. Own an early piece of an already winning product. Join the round.
OpenAI is giving government biodefense teams free access to its most capable biology model through a new program called Rosalind Biodefense.
The model itself is nothing new. It launched in April and has been locked to a short list of approved pharma and research labs ever since, kept from the general public on purpose.
The reason it stays locked is worth slowing down on.
The same skills that help a scientist design a defense against a virus can help someone design the virus, and OpenAI says that risk is exactly why access is gated.
GPT-Rosalind is the sharpest biology brain yet, sitting behind a vetting process and what the company calls safety and accountability controls.
What is new today is who gets in.
Handing the model to government biodefense teams points the cybersecurity playbook at biology. The labs already sell government-tuned hacking-defense models through approved-only channels, and now a pathogen-capable model gets the same treatment.
Free access for governments also looks like a foot in the door for the federal contracts that follow. Naming a model this locked down after Rosalind Franklin, who got cut out of the DNA discovery she helped crack, is its own kind of irony. The thing to watch is who makes the approved list, and what happens the first time someone on it should not be.
SoftBank pledged up to 75 billion euros, around $81 billion, to cover France in AI data centers, the biggest such project Europe has seen.
Why? Companies can buy all the chips they want, but there is barely enough power to run them, and France sits on a surplus of cheap nuclear.
That surplus is what SoftBank is really buying.
The first 45 billion euros builds 3.1 gigawatts in northern France by 2031, with 2 more planned.
A hub in Dunkirk, built with Schneider Electric, sits close enough to serve London, Brussels and Amsterdam.
The full 5-gigawatt site would draw about as much power as five nuclear stations, roughly New York City at peak.
The deal came together over dinner.
Macron and Son met in Tokyo in April, where the president sold France on two things it actually has, cheap nuclear power and permits that move fast.
A single gigawatt costs around $50 billion all in, so a 5-gigawatt build runs closer to $250 billion than $81 billion, and SoftBank needs partners it has not named to cover the rest.
Ford climbed 44% in May, its best month since 2009, back when it was dodging the bankruptcy that took down GM and Chrysler.
This time the story is batteries, and it shows how stretched the AI trade has gotten.
A Morgan Stanley analyst wrote 12 that Ford's battery business could be worth $10 billion and might land deals with the data center giants desperate for power. That was enough.
The pitch is Ford Energy selling storage to utilities, data centers and heavy industry, the same market where storage made up 13.5% of Tesla's 2025 revenue.
One problem is that Ford will not earn a dollar from it until 2028. A fund manager called the run "more about hopes and dreams than facts."
"The market is rewarding AI adjacency almost as much as actual AI right now," another investor put it, and Ford is the clean example.
A 122-year-old carmaker that earns nothing from storage for two more years became a market darling on one analyst note. It was the 461st cheapest stock in the S&P 500 before that note landed.
This is what the AI trade looks like when it runs out of real AI companies to buy. Money is paying a premium for anything that plugs into a data center, down to a Detroit automaker.
It stays a momentum trade in an AI costume until Ford signs a real deal with one of those giants.
Returns are where e-commerce margins go to die, and Returnalyze exists to find out Replay records your browser test runs so that when one fails, an AI agent opens the recording and tells you why. It captures every function call, DOM change and network request as the test runs, then travels back through that timeline to find the root cause.
When it has an answer it drops a comment on your GitHub PR with the failure sequence, a suggested fix naming the exact file and code change and an evidence trail pulled from the real run.
Try this yourself:
Swap Replay into your Playwright or Cypress config. Your tests stay exactly as they are.
Run the suite. Replay records each failing test deterministically, down to the line of code.
Let the agent investigate. It posts the root cause, the step-by-step failure and a suggested fix on your PR, with a confidence level.
Step through any recording yourself in the time-travel DevTools, adding print statements that run backward and forward in time.
Japan is testing AIBeS, a system that uses thermal sensors and AI to spot a bear and then fire commercially available repellent on its own, running around the clock.
Runway is making London its European headquarters and putting more than $200 million into UK AI by 2028, following Anthropic and OpenAI into the city after a $315 million round valued the world-models company at $5.3 billion.
Belgian AI company Kantify is feeding stem-cell models of rare neuromuscular diseases into an EU project called DREAMS, screening 2,700 approved drugs to find ones that could be repurposed for conditions like Duchenne.
Researchers in Guangzhou built an EEG headband that reads brain activity and pairs AI with mindfulness to cut car sickness, and it helped 83% of more than 100 riders in real road tests.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
