How This A.I. Company Collapsed Amid Silicon Valley’s Biggest Boom
Builder.ai went from a value of $1.5 billion to zero in a few months, amid questions over the sales of an A.I. product. Its downfall hints at a broader downturn.
By David Streitfeld, The New York Times, Aug. 31, 2025, 5:00 a.m. ET
…
Within a few months, Builder, which was based in London and had operations in India and California, went from a $1.5 billion unicorn to bankruptcy. It is now being liquidated in a Delaware court….
“Fake A.I. has long been pervasive in Silicon Valley, but with the bubble it’s really taken off,” said David Gerard, who runs the popular debunking site Pivot to A.I. “If you want funding, you just say a bunch of A.I. words — ‘machine learning’ and ‘large language models’ and ‘This is the future.’ You don’t have to actually have A.I.”… [end quote]
The article is about one company. It’s hard to say how much money has been invested in companies or AI projects overall which will eventually implode like so many of the dot.coms.
Excessive exuberance seems to be a universal phenomenon. It’s everywhere. Even in AI.
AI is a new technology with much potential. “Everyone” is investing heavily. No one wants to be left behind. I think we know 10 years from now a few companies will dominate the field. Just like Google and search. If you miss the surge and get left behind the best you can hope for a small specialty segment.
The mad scramble to find the winners is on. Yes, fake players are out there. Caveat emptor.
I’m not sure that’s true. There are some things which devolve to having only one or two survivors left standing, and yes Google is one. Facebook was, but isn’t any longer; while still dominant it now has at least a half dozen competitors (a couple of which the DoJ foolishly allowed Facebook to own.) EBay was the same, where the network effect made it the “go to” for auction sales.
But those sorts of limitations will not be as important in the use of AI, simply because there will be so many more uses of it. Those network effects really counted in “search” and “retail” and “auction”, but I think AI will be far more widely dispersed: think medical, production, automation, driving, [search], art, composition, architecture, cooking, music, and so on. I suspect there will be only a handful of winners for the hardware, mostly because it’s so capital intensive to be one of those producers - but on the other end, the software, it will be applied to everything from fast food prep to robotic surgery, and from game playing to sports odds prediction. There will be thousands of differing specialties on the software side, each trained appropriately (and that’s the key) and used discretely, alone and apart from all the others. We didn’t seem to find that much in, for instance, search. (Yes, there were a couple: FindLaw, etc. but those were minor.)
Once the software is perfected, one crawler can do it all. Why duplicate that? Yes, the user might see different brands on the output, but the input is a commodity. One collector is likely to dominate.
In 2017, a publicly traded beverage company called the Long Island Iced Tea Corp. announced that it was renaming itself Long Blockchain Corp. They gave no justification for the pivot.
But based on the name change alone to replace “iced tea” with “blockchain”, their stock price surged 380%. Temporarily, that is – their stock was later delisted.
Or, back then…
But really, the entire concept of valuing share prices on price-to-earnings ratios seemed passe during the internet bubble because so many of the hottest companies were unprofitable. So Wall Street even invented new metrics, like “mouse clicks” and “eyeballs”, to try to measure their growth without involving money. While that may seem crazy now, many investors at the time did not mind because they were betting on the limitless future baked into the internet’s growth assumptions.
Then again, the value creators will boom and the fakes weeded out - eventually.
Agreeing with Goofy, the game theory of the creation of monopolies suggests no such thing in the AI world. Palantir is the closest that I know of to being a monopoly but for how long?
The marketing of e-tailing has some similarities with what you describe for AI in that there are many specialized e-merchants. Nevertheless the general e-market is dominated by Amazon. That is because Bezos delivered on the concerns of the general public who wanted secure transactions, fast deliveries, and easy returns.
Analogously I think the big AI winner will be the company that can convince the public that its AI is unbiased and accurate, while also being user friendly and secure from hacking.
Perhaps a better comparison is the search engine competition. There were many competitors, from Yahoo to Altavista, to Ask Jeeves. Why Google eventually dominated may provide some insight as to what factors will determine who will win in the general use AI category.
And then there is the Apple analogy. A user-friendly AI designed for multiple functions that works only on proprietary hardware. The latter may be the best way to keep the software secure. Apple has come along way by optimizing ease-of-use and integration of its software on multiple devices.
Because it’s more efficient to design a purpose-driven vehicle than one that can do everything. You don’t need an AI with “all the information in the world” if the market you’re going to serve is just “medical”. Or “design”. Or “companionship”.
Yes, the one-big-beautiful-data center could do all of those, but why? It would be like designing a car that’s also a pick up truck that’s also a sports car that’s also an SUV that’s also a dump truck.
I’m not saying that’s the way it will play out at first, but eventually, given energy restraints, data center location and other impediments, it seems that they will disaggregate into smaller and smaller units - the same way “search” went from a few to one, and now Amazon has it’s own, Home Depot & Lowe’s have their own, Apple has its own, DuckDuckGo emerged, and so on.
I note that at this point Google is under assault not from competitors, per se, but from AI transformation which threatens their core business model (advertising referrals) as their AI is answering questions before people even get to any links. (I see scant evidence of that so far, except my own personal experience, but I’m willing to project based on that.)
At any rate, I expect there will (eventually) be purpose driven, smaller AI units rather than big, monstrous, energy gulping centers - although there will surely be some of those as well.
At the consumer level, or general purpose level, I’d agree. I think the analogies to Google are good. Google “won” the search business because it was good enough at general purpose searching to attract advertising dollars.
But I suspect the real financial winners will be more specialized AI. Generalized AI will probably always have problems with the Garbage In side of GIGO. If you want an AI that knows everything, you have to train it on everything available. And that will likely include some garbage. It could be as innocent as failing to recognize that the radio broadcast of War of the Worlds is a bit of fiction and not fact, to deliberately feeding the AI misinformation and disinformation for political or financial or personal gain.
More specialized AI would be trained on a limited (human curated??) set of information. Train a medical AI on only the top rated medical journals instead of every medical journal and news article and press release available. And there could be room for some exclusivity. “Welcome to Dr. Peter’s cancer AI. The only medical AI with exclusive access to the Tom and Jerry school of medicine’s pre-publication research on the latest cancer therapies.”
Or imagine Toyota creating an AI diagnostic system for their vehicles. Have the technician describe the symptoms, ask them to make some specific observations (are the bolts to the turboencabulator loose or missing?), and plug the computer into the car’s diagnostic port so the AI can make some direct inquiries into the vehicle’s systems. The AI can then ask the tech for further information as needed to complete a diagnosis. A single update to the AI with the latest technical bulletins - or letting the AI itself create technical bulletins based on its own diagnoses - keeps it up to date and highly useful.
Generalized AI looks good. But specialized AI is the one that will be the real transformative power in the future. At least that’s what I think.
Apparently some of the margin that the most popular AI chip seller is earning really comes from their software. They sell chips (really modules), but they also sell the underlying software tools that enable development of the AI tools that will eventually run on that hardware. I’m not sure where the line is drawn right now - how much of the critical software is from the chip seller and how much of it is from outside the chip seller?