Is the AI Bubble Real? How It Differs From Dot-Com

TL;DR

AI companies are spending far beyond their means, with OpenAI committing to roughly $1.5 trillion in deals despite targeting only $13 billion in 2025 revenue against $8.5 billion in losses. Bain estimates the industry needs $2 trillion in annual revenue by 2030 to be profitable, a figure with no clear path, fueling bubble fears alongside circular financing deals centered on Nvidia.

Transcript

Hey everyone, it's Richard. You're watching the plane bagel. It's been about three years since chatbt first launched. And since that time, we've seen generative AI develop at a remarkable pace. AI images and videos have moved from horrific to nearperfect. And large language models have gotten so good that companies have started laying off employees... Read More

Key Insights

• OpenAI was recently valued at $500 billion despite generating just over $10 billion in annual revenue and losing even more than that to expenses, making it one of the biggest money furnaces in the AI space.
• Circular financing is a core concern: Nvidia pledged to invest $100 billion in OpenAI in return for OpenAI buying its chips, while also investing in CoreWeave, a data center that sells compute back to OpenAI.
• OpenAI has committed to around $1.5 trillion in AI deals, including Project Stargate ($500 billion), a $300 billion Oracle compute deal, and chip purchases from Nvidia and AMD forecast to cost $500 billion and $300 billion.
• Bain & Company estimates AI companies will need $2 trillion in annual revenue by 2030 to be profitable, more than the combined 2024 revenue of Microsoft, Meta, Alphabet, Amazon, Apple, and Nvidia, with no clear path there.
• The AI supply chain has three tiers: chip companies like Nvidia and AMD, infrastructure providers like Amazon, Microsoft, and Oracle, and the AI model companies themselves that buy the compute.
• Vendor financing risks overstating Nvidia's profits because investing in customers so they can buy your chips effectively sells products at a discount without recording a hit to margins, a tactic popular with Nortel late in the dot-com bubble.
• The scale is enormous: OpenAI plans to consume 26 gigawatts of data center capacity, where a single gigawatt roughly equals one nuclear plant's output and powers nearly 900,000 households.
• Sentiment signals abound: 54% of global fund managers believed we were in a bubble per a Bank of America survey, Michael Burry announced short positions against AI companies, and Sam Altman himself admitted we are probably in a bubble.

Questions & Answers

Q: Are we in an AI bubble right now?

Many indicators suggest bubble conditions. A Bank of America survey found 54% of global fund managers believed we were in a bubble, the IMF and Bank of England warned about soaring valuations, investor Michael Burry announced short positions against key AI companies, and OpenAI CEO Sam Altman himself admitted we are probably in a bubble. The CAPE ratio has reached levels not seen since the dot-com bubble of 2000, despite the space not really making money at the moment.

Q: Why is OpenAI's business model considered questionable?

OpenAI was valued at $500 billion despite just over $10 billion in annual revenue while losing even more than that to expenses. It targets $13 billion in revenue and $8.5 billion in losses for 2025, and its own estimates see it burning through $115 billion by 2029. The company even loses money on its highest $200-a-month subscription because people use the service more than expected, and most users are free subscribers melting its GPUs.

Q: What is the circular financing concern in the AI industry?

The concern is that key AI financing deals loop back on each other. Nvidia pledged to invest $100 billion in OpenAI in return for OpenAI buying its chips, invested in CoreWeave, a data center that sells compute to OpenAI, and agreed to backstop CoreWeave. AMD supplied OpenAI warrants for shares in return for chip purchases. OpenAI pays Oracle $300 billion for data centers, and Oracle buys Nvidia chips, which invests in OpenAI, forming a circle.

Q: How much money is OpenAI committing to AI deals?

OpenAI has committed to around $1.5 trillion in AI deals. This includes Project Stargate, a $500 billion initiative to build US data centers, a $300 billion deal with Oracle to purchase compute over roughly five years, and chip deals with Nvidia and AMD that the Financial Times forecasts could cost $500 billion and $300 billion respectively. It also announced a $250 billion deal with Microsoft's Azure and a $38 billion deal with Amazon's AWS.

Q: How does the current AI situation differ from the dot-com bubble of 2000?

While both involve investors pouring money into a promising technological revolution and pushing the CAPE ratio to similar heights despite little profit, key differences exist. Many AI infrastructure players such as Amazon, Microsoft, and Oracle have very strong balance sheets, and companies like Meta have very good financials. The video aims to add context rather than sensationalize, noting the situation is more nuanced than a simple dot-com 2.0 repeat, though vendor financing echoes late dot-com tactics.

Q: How much revenue does the AI industry need to become profitable?

Bain & Company estimates that AI companies will need $2 trillion in annual revenue by 2030 to be profitable. As the Wall Street Journal noted, that is more than the combined 2024 revenue of Microsoft, Meta, Alphabet, Amazon, Apple, and Nvidia, and five times more than the entire software-as-a-service market. There does not seem to be a clear path toward that level of profitability, even though 88% of companies use AI in their workflows to some capacity.

Q: Why does vendor financing raise concerns about Nvidia's profits?

If Nvidia invests in its customers to give them the cash they need to buy its chips, it is effectively selling products at a discount without recognizing the hit to its profit margin, since investments are not factored into margin calculations. This could mean Nvidia is overstating its profits. This type of vendor financing became very popular with companies like Nortel toward the end of the dot-com bubble, adding to concerns that demand is being propped up beyond what is sustainable.

Q: How large is the AI infrastructure buildout in terms of power and capital?

The scale is enormous. OpenAI plans to build up to an additional 26 gigawatts of data center capacity, where a single gigawatt roughly equals one nuclear power plant's output and powers just shy of 900,000 households. McKinsey estimates data centers and infrastructure will need nearly $7 trillion in capital expenditures over five years. Venture capitalists poured a record nearly $200 billion into AI startups this year alone, over half of total year-to-date VC investment.

Summary & Key Takeaways

• Three years after ChatGPT launched, generative AI has advanced rapidly, but a growing chorus warns of a bubble. A Bank of America survey found 54% of fund managers believed we were in one, the IMF and Bank of England flagged soaring valuations, Michael Burry shorted AI companies, and Sam Altman conceded it is probably a bubble.

• The video maps three AI tiers: chip sellers like Nvidia, whose market cap hit a record $5 trillion, and AMD; infrastructure providers like Amazon, Microsoft, and Oracle; and AI companies from well-financed Meta to over 1,300 startups valued above $100 million and nearly 500 unicorns, led by OpenAI.

• OpenAI targets $13 billion revenue against $8.5 billion losses for 2025 and may burn $115 billion through 2029. Circular deals center on Nvidia investing in customers who buy its chips. Bain estimates $2 trillion in annual revenue is needed by 2030 for profitability, with no obvious path to that demand.