AI’s weird, frothy, bubble-icious summer

“This time it’s different.”

Those four words, the official slogan of every economic bubble, have been weaponized in the age of AI. They’re the rallying cry of utopians and evangelists insisting the technology will reshape every corner of life, and that the financial gains surrounding it are therefore excusable. At the same time, those same boosters, along with their critics, wield the phrase as a cudgel in their endless sparring.

The AI conversation now feels as fractured—and as split-screen—as debates over politics or the economy. Every new headline becomes ammunition, seized on to reinforce whatever larger argument someone wants to make. OpenAI CEO Sam Altman captured it rather well in August: “Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes,” he said at the time. “Is AI the most important thing to happen in a very long time? My opinion is also yes.”

Of course, only the first part of the quote got aggregated into countless even Sam Altman says AI is a bubble stories.

Predicting whether we are or aren’t in a bubble is a fool’s game, and I won’t play it. But what Altman effectively elides is that the financial frenzy part risks the most important thing to happen part.

The greed is endangering the good.

You don’t need to talk to a chatbot until it induces psychosis; you just need to follow the money in AI.

In this Premium piece, you’ll learn:

• How the insatiable appetite for elite AI talent is breaking the social contract of startup creation
• What the race to build ever larger data centers to power AI use is really about
• Why you should closely watch the use of special purpose vehicles (SPVs) to buy and sell shares in AI startups

This time is different: Whatever happens in AI will not be exactly what happened to inflate and then pop the dotcom bubble in 2000 (though, naturally, SoftBank CEO Masayoshi Son has costarring roles in each drama). So let’s go through the events of just the last three months or so and you decide what the hell is going on.

Sama-Jony-ding-dong

Just before Memorial Day, OpenAI acquired Jony Ive’s io AI hardware company, which has never released a product or even announced one. But the potential riches if you build what my colleague Mark Wilson dubbed the fourth great computing interface are such that the startup commanded $6.5 billion in OpenAI stock.

The nine-minute video that accompanied the news reportedly cost $3 million.

Weeks later, a startup named iyO, which is also making AI-powered hardware, sued OpenAI alleging trademark infringement and unfair competition, saying that iyO met with OpenAI and Ive’s LoveFrom as far back as 2022 and as recently as spring 2025 sharing its vision and tech for “screenless natural language human-computer interaction.”

A judge ruled iyO made enough of a case to merit a hearing and ordered OpenAI not to discuss the io brand. OpenAI then took down the video.

ARRrrrgh

Cursor, the buzzy AI code editor (or “vibe coding app”), announced in June that it had raised $900 million at a $9.9 billion valuation. The company also claimed its annualized recurring revenue (ARR) hit $500 million—up from the $100 million milestone it bragged about in February, which at the time made it the fastest company ever to get there, in just 12 months.

But ARR is a slippery metric. It’s calculated by taking a single month of revenue (almost always the company’s best) and multiplying it by 12. The result is a gaudy number that implies annual revenue without actually being annual revenue.

Cursor’s “record” quickly became the AI industry’s version of the four-minute mile. In July, rival AI coding startup Lovable crowed that it hit $100 million ARR in only eight months.

Not to be overshadowed, Anthropic (OpenAI’s nerdier LLM competitor) reportedly hit $3 billion ARR in May, $4 billion in June, and $5 billion in July. By year’s end, it says it will reach $9 billion ARR.

Yet even if you take those claims at face value, the company’s real revenue pencils out closer to $4.3 billion on a back-of-the-envelope calculation. That didn’t stop Anthropic from raising at a $183 billion valuation, or roughly 42.5 times this guesstimate of its 2025 revenue.

While ARR is soaring, so too are concerns that AI companies’ negative gross margins, meaning it costs more to do what it does than it makes from selling it, will not easily be reversed.

But hey, number go up, right?

Building the Avengers of AI

In June, Meta CEO Mark Zuckerberg paid $14.3 billion for 49% of Scale AI, a company which provides model makers with training data. Scale cofounder and CEO Alexandr Wang and some other Scale employees moved over to Meta to set up the social media giant’s new Superintelligence Labs (MSL).

The move raised questions about whether Big Tech was yet again deploying a clever workaround to federal regulatory scrutiny for a large acquisition by spending a large sum for a stake in the company while its most valuable employees join the investor. Good questions!

Even more so after Meta rivals, which had been Scale customers, backed away, and in July Scale laid off 14% of its staff, 200 people, and ended relationships with 500 contractors.

Soon the tumult came for Meta itself, with Meta in August instituting an AI hiring freeze, doing a reorganization of the AI teams it had just organized (the fourth reorg in six months, but who’s counting), and some of those high-profile hires changing their mind once they had to show up to work at Meta.

At least all this took place over the leisurely cadence of about two and a half months.

Choppy Surf

The drama over the acquisition of the AI coding app Windsurf played out over just 72 hours, across a mid-July weekend.

This too looked like a company being raided for its leadership and top talent (this time by Google). In the heat of a summer Friday news dump, it appeared those 40-plus Windsurf poachees were leaving their former colleagues high and dry while they reaped a share of $2.4 billion.

AI, whether because of the opportunity, the resources required, or just the guaranteed money, was shattering the heretofore understood premise that founders, like sea captains, went down with the ship.

But with Big Tech offering yachts, not lifeboats, startups became just another manifestation of “I got mine,” with founders now at the front of the line to get theirs.

On the Monday after the Friday talent raid, Cognition, yet another AI software engineering tool (!), took the hero’s mantle, acquiring the rest of Windsurf and promising to make the employees whole.

Cognition CEO Scott Wu wrote that the deal “includes Windsurf’s world-class people, some of the best talent in our industry, whom we’re privileged to welcome to our team.”

Those good vibes lasted all of three weeks, when Wu confirmed that he had offered those “world-class” Windsurf people buyouts, citing Cognition’s “extreme performance culture” that translates to “many of us literally live where we work.” The note implied that Windsurf perhaps didn’t share that oh-so-healthy work culture and allowing that its employees hadn’t signed up to grind themselves into dust.

The only greater privilege than welcoming Windsurf to the team was showing them the door.

Mine’s bigger

Stargate: It’s not just a sci-fi franchise but also a joint venture to fuel OpenAI’s data center expansion, an “AI infrastructure” company. Stargate launched in January with help from SoftBank, Oracle, and others. According to OpenAI, Stargate “intends to invest $500 billion over the next four years” and this ”infrastructure will secure American leadership in AI, create hundreds of thousands of American jobs, and generate massive economic benefit for the entire world.” (Emphasis mine.)

The first Stargate project in Texas, which started operating while still under construction, is promising “357 full-time jobs once construction has finished,” according to Bloomberg, which added that that’s “a little more than the average Walmart Supercenter.” If we interpret “hundreds of thousands” as, say, 200,000, we only need to build another 560 data centers to fulfill that goal.

Gotta start somewhere, I guess, but in truth, OpenAI got to that jobs number through the usual economic impact sleight of hand, adding in “short-term construction roles” and “indirect jobs like manufacturing and local service roles.”

Good American jobs, AI supremacy over the world, whatever. Let’s not lose sight of what’s important here: Sam Altman’s data center was larger than Elon Musk’s xAI Colossus data center, consuming 300MW of power to Colossus’s 250MW. The tech site Tom’s Hardware proclaimed, “It’s the world’s largest single building.”

Fear not, though, Colossus 2 is underway in Memphis. Sure, both of these projects threaten the stability of the electric grid, and Colossus may or may not be poisoning the residents of Memphis while you read this, but no matter: Musk simply must prove his is bigger.

You didn’t think Meta CEO Mark “Cage Fighter” Zuckerberg was going to sit this out, right? In July, he announced Prometheus and Hyperion (nicknamed “The Beast”), his data center projects that promise multi-gigawatt power consumption. As Zuck bragged, “Just one of these covers a significant part of the footprint of Manhattan.”

SoftBank’s Masayoshi Son, although already involved with Stargate (with money he doesn’t actually have), was not deterred by those realities from reportedly pitching a $1 trillion industrial campus in Arizona devoted to building AI and robotics, dubbed Project Crystal Land.

Can’t help but wonder if these guys are compensating for something here.

It’s the money, stupid

Of course, the hot AI summer of 2025 hasn’t just been about bragging rights. It’s about money, which can be converted into bragging rights.

To wit:

• In July, former OpenAI CTO Mira Murati raised a record $2 billion seed round for her Thinking Machines Lab AI startup.
• As of mid-August, arguably a lifetime ago in AI time, the analyst firm CB Insights calculated that there were 498 “AI unicorns,” with an aggregate value of $2.7 trillion, leading CNBC to conclude, “AI is creating new billionaires at a record pace.” On paper, at least.
• Thankfully there’s a way for the founders (and maybe even the unwashed AI rank and file actually doing the work) to cash in that doesn’t involve abandoning their startups: secondary share sales to investors clamoring to get a piece of these companies. In August, OpenAI reportedly initiated a $6 billion offering (or maybe it’s $8 billion) to investors (including, naturally, SoftBank), which would value the company at $500 billion. OpenAI currently has a $13 billion ARR (see above!). So let’s call this maybe 50 times revenue, with, again, accelerating losses.
• Elite AI researchers and “10x” engineers spent the summer changing teams like college football stars in the transfer portal, with Zuckerberg reportedly offering $300 million, four-year deals to woo them to Meta. The coup de grace was hiring away Apple’s head of AI models with a compensation package that dwarfed that of Apple CEO Tim Cook.

Meanwhile, investor demand has gotten so frenzied that

1) companies like Meta are using financial instruments known as special purpose vehicles (SPVs) to finance data center investments off its books, obscuring the risk from its stock investors

2) OpenAI warned investors that unauthorized

• sales of OpenAI equity;
• investments in SPVs that own OpenAI equity;
• tokenized interests in OpenAI equity (as Robinhood launched earlier in the summer, starting in Europe) and promised or an SPV holding OpenAI equity; and
• “forward” contracts and other forms of purported economic interests as violating its terms and could result in those shares being invalidated

3) OpenAI’s investing arm launched its sixth SPV, mere weeks before warning investors about SPVs offering its own shares, to invest outside of its main fund, seeking another $69.5 million

4) VCs are buying into each other’s SPVs with high fee structures to get any kind of exposure to these those private AI high-fliers. As Javier Avalos, cofounder and CEO of Caplight, a secondary deal tracking platform, explained to TechCrunch, “Buying units of the SPV means [VCs] won’t own shares in the actual company; they’ll technically be an investor in another investor’s fund.” Of course they can then mark up those investments and sell them to a greater fool high-net-worth investor, so everything’s fine.

These are the kinds of things that happen when the “smart money” convinces itself that “this time is different.”

The next crash will be different, and it may well not be caused by AI. (It could be private credit! Or crypto! So many possibilities.)

Yes, every single one of these actions this summer has some kind of perfectly logical explanation. Why wouldn’t you bet 1 or 2% of your company to have a shot at owning the next iPhone, being the dominant player in the fourth industrial revolution, or controlling data centers, the “railroads” of the 21st century?

Then again, everything looks shiny if you’re living inside a gossamer sphere floating into the atmosphere.

Stay careful out there, folks. If anyone offers you a chance to buy into an SPV, maybe take all your money and put it in something tangible and safe, like Labubus.

https://www.fastcompany.com/91397736/ai-weird-frothy-bubblicious-summer?partner=rss&utm_source=rss&utm_medium=feed&utm_campaign=rss+fastcompany&utm_content=rss