“This time it’s totally different.”
These 4 phrases, the official slogan of each financial bubble, have been weaponized within the age of AI. They’re the rallying cry of utopians and evangelists insisting the expertise will reshape each nook of life, and that the monetary video games surrounding it are due to this fact excusable. On the identical time, those self same boosters, together with their critics, wield the phrase as a cudgel of their infinite sparring.
The AI dialog now feels as fractured—and as split-screen—as debates over politics or the economic system. Each new headline turns into ammunition, seized on to bolster no matter bigger argument somebody needs to make. OpenAI CEO Sam Altman captured it rather well in August: “Are we in a part the place buyers as an entire are overexcited about AI? My opinion is sure,” he stated on the time. “Is AI an important factor to occur in a really very long time? My opinion can also be sure.”
After all, solely the primary a part of the quote bought aggregated into numerous even Sam Altman says AI is a bubble tales.
Predicting whether or not we’re or aren’t in a bubble is a idiot’s recreation, and I gained’t play it. However what Altman successfully elides is that the monetary frenzy half dangers an important factor to occur half.
The greed is endangering the nice.
You don’t want to speak to a chatbot till it induces psychosis; you simply have to observe the cash in AI.
On this Premium piece, you’ll be taught:
- How the insatiable urge for food for elite AI expertise is breaking the social contract of startup creation
- What the race to construct ever bigger information facilities to energy AI use is basically about
- Why you need to intently watch the usage of particular goal automobiles (SPVs) to purchase and promote shares in AI startups
This time is totally different: No matter occurs in AI won’t be precisely what occurred to inflate after which pop the dotcom bubble in 2000 (although, naturally, SoftBank CEO Masayoshi Son has costarring roles in every drama). So let’s undergo the occasions of simply the final three months or so and also you resolve what the hell is happening.
Sama-Jony-ding-dong
Simply earlier than Memorial Day, OpenAI acquired Jony Ive’s io AI {hardware} firm, which has by no means launched a product and even introduced one. However the potential riches when you construct what my colleague Mark Wilson dubbed the fourth great computing interface are such that the startup commanded $6.5 billion in OpenAI inventory.
The nine-minute video that accompanied the information reportedly price $3 million.
Weeks later, a startup named iyO, which can also be making AI-powered {hardware}, sued OpenAI alleging trademark infringement and unfair competitors, saying that iyO met with OpenAI and Ive’s LoveFrom way back to 2022 and as not too long ago as spring 2025 sharing its imaginative and prescient and tech for “screenless pure language human-computer interplay.”
A decide dominated iyO made sufficient of a case to advantage a listening to and ordered OpenAI to not focus on the io model. OpenAI then took down the video.
ARRrrrgh
Cursor, the buzzy AI code editor (or “vibe coding app”), introduced in June that it had raised $900 million at a $9.9 billion valuation. The corporate additionally claimed its annualized recurring income (ARR) hit $500 million—up from the $100 million milestone it bragged about in February, which on the time made it the quickest firm ever to get there, in simply 12 months.
However ARR is a slippery metric. It’s calculated by taking a single month of income (virtually at all times the corporate’s greatest) and multiplying it by 12. The result’s a gaudy quantity that means annual income with out really being annual income.
Cursor’s “document” rapidly grew to become the AI business’s model of the four-minute mile. In July, rival AI coding startup Lovable crowed that it hit $100 million ARR in solely eight months.
To not be overshadowed, Anthropic (OpenAI’s nerdier LLM competitor) reportedly hit $3 billion ARR in Might, $4 billion in June, and $5 billion in July. By yr’s finish, it says it’ll reach $9 billion ARR.
But even when you take these claims at face worth, the corporate’s actual income pencils out nearer to $4.3 billion on a back-of-the-envelope calculation. That didn’t cease Anthropic from elevating at a $183 billion valuation, or roughly 42.5 instances this guesstimate of its 2025 income.
Whereas ARR is hovering, so too are issues that AI corporations’ adverse gross margins, which means it prices extra to do what it does than it makes from promoting it, won’t simply be reversed.
However hey, quantity go up, proper?
Constructing the Avengers of AI
In June, Meta CEO Mark Zuckerberg paid $14.3 billion for 49% of Scale AI, an organization which gives mannequin makers with coaching information. Scale cofounder and CEO Alexandr Wang and another Scale staff moved over to Meta to arrange the social media large’s new Superintelligence Labs (MSL).
The transfer raised questions on whether or not Massive Tech was but once more deploying a intelligent workaround to federal regulatory scrutiny for a big acquisition by spending a big sum for a stake within the firm whereas its most precious staff be part of the investor. Good questions!
Much more so after Meta rivals, which had been Scale prospects, backed away, and in July Scale laid off 14% of its employees, 200 individuals, and ended relationships with 500 contractors.
Quickly the tumult came for Meta itself, with Meta in August instituting an AI hiring freeze, doing a reorganization of the AI groups it had simply organized (the fourth reorg in six months, however who’s counting), and a few of these high-profile hires altering their thoughts as soon as they needed to present as much as work at Meta.
At the least all this came about over the leisurely cadence of about two and a half months.
Uneven Surf
The drama over the acquisition of the AI coding app Windsurf performed out over simply 72 hours, throughout a mid-July weekend.
This too regarded like an organization being raided for its management and prime expertise (this time by Google). Within the warmth of a summer season Friday information dump, it appeared these 40-plus Windsurf poachees have been leaving their former colleagues excessive and dry whereas they reaped a share of $2.4 billion.
AI, whether or not due to the chance, the assets required, or simply the assured cash, was shattering the heretofore understood premise that founders, like sea captains, went down with the ship.
However with Massive Tech providing yachts, not lifeboats, startups grew to become simply one other manifestation of “I bought mine,” with founders now on the entrance of the road to get theirs.
On the Monday after the Friday expertise raid, Cognition, yet one more AI software program engineering device (!), took the hero’s mantle, buying the remainder of Windsurf and promising to make the staff complete.
Cognition CEO Scott Wu wrote that the deal “consists of Windsurf’s world-class individuals, among the greatest expertise in our business, whom we’re privileged to welcome to our workforce.”
These good vibes lasted all of three weeks, when Wu confirmed that he had supplied these “world-class” Windsurf individuals buyouts, citing Cognition’s “extreme performance culture” that interprets to “many people actually reside the place we work.” The be aware implied that Windsurf maybe didn’t share that oh-so-healthy work tradition and permitting that its staff hadn’t signed as much as grind themselves into mud.
The one higher privilege than welcoming Windsurf to the workforce was displaying them the door.
Mine’s greater
Stargate: It’s not only a sci-fi franchise however a three way partnership OpenAI’s information middle enlargement, an “AI infrastructure” firm it stood up in January with assist from SoftBank, Oracle, and others. In response to OpenAI, Stargate “intends to speculate $500 billion over the following 4 years” and this ”infrastructure will safe American management in AI, create lots of of hundreds of American jobs, and generate huge financial profit for the whole world.” (Emphasis mine.)
The primary Stargate undertaking in Texas, which began working whereas nonetheless underneath building, is promising “357 full-time jobs as soon as building has completed,” according to Bloomberg, which added that that’s “slightly greater than the common Walmart Supercenter.” If we interpret “lots of of hundreds” as, say, 200,000, we solely have to construct one other 560 information facilities to meet that objective.
Gotta begin someplace, I assume, however in fact, OpenAI got to that jobs number by way of the standard financial influence sleight of hand, including in “short-term building roles” and “oblique jobs like manufacturing and native service roles.”
Good American jobs, AI supremacy over the world, no matter. Let’s not lose sight of what’s necessary right here: Sam Altman’s information middle was bigger than Elon Musk’s xAI Colossus information middle, consuming 300MW of energy to Colossus’s 250MW. The tech website Tom’s {Hardware} proclaimed, “It’s the world’s largest single building.”
Concern not, although, Colossus 2 is underway in Memphis. Certain, each of those tasks threaten the soundness of the electrical grid, and Colossus might or is probably not poisoning the residents of Memphis whilst you learn this, however irrespective of: Musk merely should show his is larger.
You didn’t assume Meta CEO Mark “Cage Fighter” Zuckerberg was going to take a seat this out, proper? In July, he introduced Prometheus and Hyperion (nicknamed “The Beast”), his information middle tasks that promise multi-gigawatt energy consumption. As Zuck bragged, “Simply one in every of these covers a big a part of the footprint of Manhattan.”
SoftBank’s Masayoshi Son, though already concerned with Stargate (with cash he doesn’t even have), was not deterred by these realities from reportedly pitching a $1 trillion industrial campus in Arizona dedicated to constructing AI and robotics, dubbed Mission Crystal Land.
Can’t assist however marvel if these guys are compensating for one thing right here.
It’s the cash, silly
After all, the new AI summer season of 2025 hasn’t simply been about bragging rights. It’s about cash, which might be transformed into bragging rights.
To wit:
- In July, former OpenAI CTO Mira Murati raised a document $2 billion seed spherical for her Considering Machines Lab AI startup.
- As of mid-August, arguably a lifetime in the past in AI time, the analyst agency CB Insights calculated that there have been 498 “AI unicorns,” with an combination worth of $2.7 trillion, main CNBC to conclude, “AI is creating new billionaires at a record pace.” On paper, at the very least.
- Fortunately there’s a method for the founders (and possibly even the unwashed AI rank and file really doing the work) to money in that doesn’t contain abandoning their startups: secondary share gross sales to buyers clamoring to get a bit of those corporations. In August, OpenAI reportedly initiated a $6 billion providing (or possibly it’s $8 billion) to buyers (together with, naturally, SoftBank), which might worth the corporate at $500 billion. OpenAI at the moment has a $13 billion ARR (see above!). So let’s name this possibly 50 instances income, with, once more, accelerating losses.
- Elite AI researchers and “10x” engineers spent the summer season altering groups like school soccer stars within the switch portal, with Zuckerberg reportedly providing $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 deal that dwarfed that of Apple CEO Tim Cook dinner.
In the meantime, investor demand has gotten so frenzied that
1) corporations like Meta are utilizing monetary devices often called particular goal automobiles (SPVs) to finance data center investments off its books, obscuring the danger from its inventory buyers
2) OpenAI warned investors that unauthorized
- gross sales of OpenAI fairness;
- investments in SPVs that personal OpenAI fairness;
- tokenized pursuits in OpenAI fairness (as Robinhood launched earlier in the summertime, beginning in Europe) and promised or an SPV holding OpenAI fairness; and
- “ahead” contracts and different types of purported financial pursuits as violating its phrases and will lead to these shares being invalidated
3) OpenAI’s investing arm launched its sixth SPV, mere weeks earlier than warning buyers about SPVs providing its personal shares, to invest outside of its main fund, searching for one other $69.5 million
4) VCs are buying into each other’s SPVs with excessive price buildings to get any sort of publicity to those these non-public AI high-fliers. As Javier Avalos, cofounder and CEO of Caplight, a secondary deal monitoring platform, defined to TechCrunch, “Shopping for items of the SPV means [VCs] gained’t personal shares within the precise firm; they’ll technically be an investor in one other investor’s fund.” After all they will then mark up these investments and promote them to a higher idiot high-net-worth investor, so all the pieces’s advantageous.
These are the sorts of issues that occur when the “sensible cash” convinces itself that “this time is totally different.”
The following crash will probably be totally different, and it might nicely not be brought on by AI. (It might be non-public credit score! Or crypto! So many potentialities.)
Sure, each single one in every of these actions this summer season has some sort of completely logical clarification. Why wouldn’t you guess 1 or 2% of your organization to have a shot at proudly owning the following iPhone, being the dominant participant within the fourth industrial revolution, or controlling information facilities, the “railroads” of the twenty first century?
Then once more, all the pieces seems shiny when you’re residing inside a gossamer sphere floating into the environment.
Keep cautious on the market, of us. If anybody affords you an opportunity to purchase into an SPV, possibly take all of your cash and put it in one thing tangible and protected, like Labubus.
