The warnings are everywhere. Nvidia’s $32 billion quarterly profit. OpenAI valued at $500 billion despite losing money until 2030. Companies borrowing $1 trillion to build data centres. Open AI’s Sam Altman committing $1.4 trillion his company doesn’t have.
Even the optimists are nervous. Sundar Pichai, CEO of Microsoft, one of the world’s most profitable companies, admits the spending is driven partly by “irrationality.” Sam Altman concedes investors are “overexcited.” Mohamed El-Erian, a professor at the Wharton School, calls it what it is: a bubble.
But, writing in the New York Times, El-Erian makes a critical distinction. This isn’t tulip mania or crypto speculation. This is a “rational bubble.” Although still going strong at the time of writing, it’s one that will burst. But it will leave the world better off. The bubble will pop.. The question is: who survives?
And that answer reveals why the AI crash will be nothing like the dot-com collapse of 2000.
The Dot-Com Playbook Won’t Work This Time
The dot-com bubble taught us a simple lesson: hype doesn’t equal value. Companies added “.com” to their names and watched valuations soar. Pets.com spent millions on Super Bowl ads while burning cash with no revenue model. When reality hit, the survivors weren’t the loudest or most hyped. They were the ones with revenue.
Amazon survived because it sold books. eBay survived because it facilitated transactions. Google survived because advertisers paid for clicks. The crash was brutal but the filter was simple: Do you generate revenue or don’t you?
The AI bubble shares disturbing similarities. OpenAI is worth $500 billion but won’t be profitable for five years. The stock market values Anthropic at $183 billion and yet it still operates in the red. Thinking Machines Labs, founded in February, is worth tens of billions. It just launched its first product.
It gets worse. Goldman Sachs estimates that 15% of Nvidia’s 2026 sales will come from circular deals. Nvidia invests in OpenAI, which then buys Nvidia chips. OpenAI receives billions from tech companies, then sends billions back for computing power. Is it financial engineering that makes the market look stronger than reality?
Gil Luria, head of technology research at financial services company D.A. Davidson, doesn’t mince words. “What OpenAI is engaged in is the most dramatic case of ‘Fake It Until You Make It’ that we have ever seen,” she said.
So yes, it’s a bubble. But here’s where the comparison breaks down.
The Three Forces That Will Pop the Bubble
El-Erian identifies three specific forces that will burst the AI bubble.
First, unsustainable spending. The arms race between tech giants requires elite engineers in high demand. It needs sprawling data centres, and massive energy consumption. Costs continue to soar. Google and Microsoft can lean on their diverse revenue streams. Other players must rely on debt funding or circular financing. That will prove unsustainable.
Second, AI-washing. Companies are slapping AI labels onto mundane services to attract less informed investors. This is reminiscent of the dot-com bubble when adding “.com” to a startup’s name was a shortcut to inflated valuations. Investors will expose these pretenders when they demand proof of value.
Third, external shocks. Sudden regulatory changes, bad actors, geopolitical fractures of the AI supply chain, or lack of widespread adoption could derail firms. Any of these would hinder the ability of companies building AI models to generate enough revenue to survive.
These three forces are already visible. But they won’t kill everyone.
Why This Bubble Is Different
The dot-com boom was built on potential. AI is built on deployment and data.
When the dot-com bubble burst, most companies hadn’t delivered transformative technology. They had websites. They had ideas. They had PowerPoint presentations about the future of e-commerce. But they weren’t changing how businesses operated.
AI already is.
AI is a general-purpose technology like electricity. It is already pervasive. It will alter a range of economic activities. The evidence is already here and, realise it or not, we are all already using AI every day.
Developers using AI coding assistants report 35% productivity gains. ChatGPT has 800 million weekly users. These aren’t pilot programmes. This is adoption at scale.
AI is already embedded in enterprise operations in ways dot-com companies never achieved.
Take SAP, the business enterprise software giant. While AI start-ups pitch potential, SAP has deployed AI across 80% of its most used tasks. Their AI co-pilot is integrated in S/4HANA. These aren’t experimental features tacked onto products. They are production systems running operations for 25,000 companies globally.
A global consumer goods company uses Joule to automate 60% of HR service requests. A European auto parts supplier cut dispute resolution time by 40%. SAP itself runs over 100 Joule use cases internally with documented cost savings.
By year’s end, SAP will have 400+ AI features in production. The company released 130+ generative AI capabilities in 2024 alone.
This matters because embedded AI has switching costs that dot-com services never had. You could cancel your Webvan grocery delivery with one click. But if you weave AI into your ERP, and supply chain systems – systems running your operation – removal means operational chaos.
SAP doesn’t have customers experimenting with AI. It has customers whose businesses run on it.
Data foundations create defensibility. It’s a survival factor the dot-com era didn’t have. AI models are only as intelligent as their training data. Enterprise AI needs to understand business reality. How companies operate, what regulations require, which processes work.
This is where established enterprise players have structural advantages start-ups can’t replicate.
SAP sits on 50+ years of enterprise transactional data from the world’s largest companies. Not aggregated web content. Not synthetic training sets. Actual financial transactions, supply chain movements, HR records, manufacturing schedules, and compliance filings across 190+ countries.
When SAP’s AI recommends a solution, it draws on decades of implemented business processes. When a start-up’s AI recommends a solution, it extrapolates from documentation and blog posts.
Consider regulatory compliance, a reality Pivot encounters every day. When regulations change, generic AI can explain what changed. But SAP’s AI knows which configuration tables need updating. It knows which modules are affected. And it knows how similar regulations were implemented in other countries. Because it learned from regulatory implementations, not theoretical compliance guides.
Start-ups can’t replicate this data advantage. And even if you could access it, you’d need decades to accumulate the depth SAP possesses.
This creates a compounding moat. Every transaction processed through SAP systems makes its AI smarter. Every regulatory change implemented adds training data. Every business process optimization enriches the models. The data advantage widens with time.
SAP vs. OpenAI: Who survives?
The dot-com crash killed indiscriminately because almost everyone was burning VC cash. The AI crash will be more surgical.
Microsoft, Google, Amazon, and Apple reported $110 billion in combined profits in their most recent quarters. They spent $360 billion on data centres last year, but they can afford it. Their AI investments are funded by revenue, not debt.
SAP generates $34 billion in annual revenue from subscriptions and services. It can outlast any funding drought.
But OpenAI, Anthropic, and the wave of AI startups? They’re burning cash at historic rates. All while borrowing up to $1 trillion collectively for infrastructure. OpenAI won’t be profitable until 2030. That’s five years of negative cash flow, betting everything on reaching artificial general intelligence. It’s a goal Sam Altman admits they don’t know how to achieve.
Anton Korinek, an economist at the University of Virginia, puts it bluntly: “It’s a bet on A.G.I. or bust.”
When the bubble pops, profitable giants weather the storm. Debt-fuelled start-ups with no path to profitability don’t. Companies such as OpenAI, relying on debt and circular financing will collapse. Those with profits and diverse revenue streams will survive.
What the Rational Bubble Leaves Behind
El-Erian is right to call this a rational bubble. The potential payoffs justify eye-watering levels of investment. AI will transform productivity, raise economic growth rates, and solve problems in health, education, and beyond. Some bets will deliver thousandfold returns.
But rationality doesn’t mean everyone survives. The losing horses go to the glue factory. The race itself, however, produces lasting innovation that improves outcomes for everyone.
The AI crash will leave us with embedded enterprise AI, productivity gains, and transformed business processes.
When regulations change overnight and you’re implementing 98 pages of tariff rules with impossible deadlines, you need AI trained on decades of enterprise reality that’s embedded in systems that can’t be easily replaced, backed by companies that will still exist when the bubble bursts.
That’s not everyone. But it’s not no one, either.
The bubble will pop. The pretenders will disappear. The survivors will have built something that transforms how the world works.
That’s the difference between a bubble and a rational bubble. Both burst. Only one leaves us better off.
