Hello, dear readers!

This is our weekly brief on remarkable AI topics, so you can stay informed without spending your entire Monday catching up on AI news.

Today's focus — the sizable AI bill. For the past year, companies have been told to use as much AI as possible and worry about the costs later. Now the invoices are arriving, and some executives are discovering that "AI-first" can be an expensive lifestyle choice. How long before every prompt needs manager approval?

Also in this week's edition:

The End of Tokenmaxxing

For the past year, companies have been trying to use as much AI as possible.

Developers got access to Claude, ChatGPT, Cursor and other tools. CEOs encouraged teams to experiment. Investors rewarded companies that could show growing AI adoption.

Now many businesses are discovering that all those AI prompts cost real money.

Uber reportedly burned through its entire annual AI coding budget by April. Microsoft has started pulling back some Claude Code licenses. One company allegedly ended up with a $500 million Claude bill after forgetting to set usage limits. Another CTO told Axios that employees were using expensive AI models to check the weather.

The funny thing is that AI itself is getting cheaper. Token prices keep falling. But usage is growing even faster.

More employees are using AI every day. Newer models can handle longer tasks. Agents can spend thousands of tokens completing work in the background. As a result, companies are consuming far more AI than they expected when they signed those enterprise contracts.

The result is what some industry observers call the end of "tokenmaxxing" — the belief that maximizing AI usage would automatically maximize business value. The idea was simple. Give everyone access to the best models and worry about costs later.

That made sense when companies were still figuring out what AI could do. But now finance departments are asking harder questions. Is AI actually saving time? Is it generating more revenue? Or is it just generating bigger invoices?

The answers are not always obvious. Some studies show that developers using the most AI tools produce more code. They also produce more bugs and spend more time fixing things. More tokens do not automatically mean more value.

A whole new market is already emerging around this problem. Startups are building tools that track AI spending, limit usage, route requests to cheaper models, and measure ROI. The Linux Foundation has even launched a new initiative focused on standards for AI costs and token accounting.

Just a few months ago, the main concern was that companies were not adopting AI fast enough.

Now the concern is that they adopted it a little too enthusiastically.

How Much AI Should the Public Own?

For years, the assumption was that if AI created trillions of dollars in value, the winners would be obvious: AI companies, their investors, and their employees.

Now politicians are starting to ask whether everyone else should get a cut too.

Last week, Donald Trump said he plans to meet with major AI companies to discuss ways ordinary Americans could benefit from their success. The idea remains vague. Trump suggested the public could effectively become a "partner" in AI companies, potentially receiving some share of the industry's gains.

At the other end of the spectrum sits a much more concrete proposal from Senator Bernie Sanders. His plan would give the federal government a 50% ownership stake in AI companies, allowing the public to directly share in future profits.

The gap between those proposals is enormous. One sounds more like a dividend. The other would fundamentally reshape ownership of the industry's most valuable companies.

But both are responding to the same question.

If AI really does automate large amounts of work and generate unprecedented profits, should those gains flow mainly to shareholders? Or should society as a whole receive some share of the upside?

Until recently, AI policy debates focused on safety, regulation, and competition. Increasingly, they're becoming debates about economics.

AI Designs Its First Vaccine

Most AI products today generate text, code, or images.

Researchers at the University of Cambridge are trying something more ambitious: using AI to design vaccines.

Last week, the team announced early human trial results for what they describe as the first vaccine whose key component was designed entirely by AI. Instead of targeting a single coronavirus variant, the vaccine is meant to provide protection against a whole family of related viruses — including ones that do not yet exist.

Before anyone gets too excited, this is still very early research. The first trial involved just 39 people and was designed primarily to test safety rather than effectiveness. The vaccine is years away from any real-world use, assuming larger trials succeed.

Still, it's a reminder that some of AI's most interesting applications may happen far away from chat windows. While consumers debate which chatbot is best, researchers are increasingly using AI to tackle problems in medicine, biology, and other sciences.

Thanks for reading AIport. Until next Monday — by then, AI will almost certainly have found a new way to spend your money.