AI Might Actually Be Too Big to Fail
With so much divisiveness in today’s society, it’s not often that we reach a consensus about anything, so it’s somewhat striking that so many are saying the same thing about artificial intelligence. Pick up a newspaper, switch on the TV or browse the web, and it seems that every news organization, podcast and blogger parrots the same prediction – AI is in a bubble, and it’s going to cause chaos when it bursts.
In fact, even the companies leading the AI industry are saying it. Amazon’s founder Jeff Bezos recently referred to AI as an “industrial bubble,” while Google CEO Sundari Pichai described the market as having "elements of irrationality." Sam Altman of OpenAI thinks the industry is in a bubble too, and his company’s chairman of the board Bret Taylor agrees.
But perhaps even more unusual than the fact everyone seems to agree on the AI bubble talk, they’re not really acting like it’s true. On the contrary – investors continue to shovel hundreds of millions of dollars at AI companies every month, while Google, OpenAI, Microsoft, Meta and Amazon are still racing to build data centers like there’s no tomorrow, spending billions in the process.
Is it possible that maybe, just maybe, this bubble isn’t what it seems?
Cultivating Actual Businesses
Instead of the bubble bursting, it’s far more likely to deflate slowly, argues Zeev Farbman, Chief Executive of the AI video software company Lightricks. In a recent column, Farbman opined that AI has become synonymous with massive valuations and unlimited scalability, leading to a feeling that no one can compete with the industry’s biggest players. But he says it has now become apparent that’s not really true.
Farbman believes the AI industry is shifting towards a different kind of future, where AI companies won’t be desperately searching for subscriptions to try and claw back the huge amounts of money they’ve spent developing models and building data centers. Instead, he sees an industry focused on how AI models can be used to make money for businesses. And that means the resources being invested in newer, more powerful models, will slowly dwindle.
“Scale alone is no longer delivering step-function gains,” Farbman said. “Execution, distribution and ecosystem now matter more than raw model size. Adjusting expectations to this new reality will allow the growing AI bubble to slowly deflate, rather than burst and wreak havoc on the economy and financial markets like the dotcom bust did a quarter century ago.”
The spending on AI won’t dry up overnight, but that’s fine too, because most of the industry’s biggest spenders can easily afford to keep throwing money at it.
Commentator Derek Thompson makes a compelling case for it. Writing in Substack, he pointed out that most of the companies leading the dotcom boom were spending millions, despite not making any money at all. A case in point is Pets.com, which spent almost $12 million on advertising in 1999 while earning only $700,000 in revenue that year.
The hyperscalers leading the AI boom are very different beasts, Thompson said. They’re some of the most profitable companies that the world has ever seen, generating billions of dollars in annual revenue and profit. Unlike the dotcom companies, they’re not totally reliant on future profits and they’re not playing with borrowed money.
They’re also making quite a lot of money from AI already, despite claims to the contrary. Thompson explains that the likes of Microsoft, Meta Platforms and Google are benefiting in more of a roundabout way by using AI to enhance their existing businesses. For instance, Meta claims to have increased its ad sales by over $1 billion thanks to AI, while Microsoft AI services recently achieved a $13 billion run rate, seeing 157% growth year over year.
“The bottom line here is that the AI build-out is unlike anything we’ve seen before,” Thompson wrote. “The richest companies in the world, with the most profitable core businesses in modern memory, are using their obscene cash flow to collectively fund a new national infrastructure project. That’s not like the railroads, or the dot-com bubble, or the housing bubble. It’s not like anything that’s ever happened.”
Investing in Something Much Bigger
The idea that AI is somehow “infrastructure” is not new. Technologist and futurist Jason Snyder believes that people assume AI is in a bubble because they compare the industry to traditional software. But he argues that AI doesn’t behave like software, because it’s building something far more transformational – an entirely new technology that will forever benefit human society.
“AI’s economics resembles the economics of infrastructure,” Snyder wrote. “Valuations may appear disconnected from productivity. Capital may look like it is circulating in a self-reinforcing pattern. Spending may appear excessive. Yet these dynamics appear irrational only through the lens of consumer technology.”
According to Synder, OpenAI’s enormous losses may look problematic, but he says that’s only true if its finances are viewed in terms of the cost structure of apps and social platforms. But if we imagine it’s building the infrastructure that’s going to underpin modern life, its spending makes a lot more sense.
It’s analogous to the early days of the electricity grid and railroads, which were built at huge expense with very little initial returns. But in the end, those investments totally transformed the economy, just as AI will eventually do.
“OpenAI’s spending is no more indicative of a bubble than Edison’s power stations or Bell’s early switchboards,” Snyder wrote. “The economics only appear flawed if one assumes the system they are building already exists.”
Groundwork for Value and Revenue
The infrastructure AI needs to scale large enough to change the world doesn’t yet exist, but that will change soon enough. And once the infrastructure is in place, companies will quickly work out how to create value and generate the revenue that ultimately pays for it, Farbman believes. “There is no question that AI is the most amazing technology we have today,” he said.
Farbman is confident that the value will arrive soon, because the financial barrier is no longer that big. He explained that many of AI’s fundamental components, such as the transformer and diffusion architectures that underpin most large language models, are already open and accessible. So the most pressing challenge now is simply building reliable systems out of these components.
“These products and services no longer require investors to front trillions of dollars,” Farbman said. “The defensibility of AI spending now lies in infrastructure optimization, proprietary data, and integration depth. Entrepreneurs with good ideas for solutions that carefully craft or use models with that end-performance in mind will win out over those that seek massive models that can later be scaled for different potential uses.”
So instead of thinking of the AI market as a bubble, it might be wiser to look at the bigger picture. The huge amount of money being thrown at the industry is building the foundation of something so consequential that we cannot even conceptualize how big it’s going to be. After all, who in 1999 could have predicted the rise of social media, smartphones, mobile applications and everything else the internet later evolved into?
The internet became the springboard for our modern, connected lives, and its impact is lasting. AI will be much the same, and so the real test is not about economics, but about how well we can craft this technology into something truly useful and economically viable. So long as we can do that well enough, chances are good that the road ahead for AI in 2026 won’t be exceedingly rocky.
Read More