Inside The AI Bubble: Debt, Depreciation, and Losses — With Gil Luria
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
Gil Luria, head of technology research at D.A. Davidson, joins Big Technology Podcast to dissect the AI bubble concerns gripping Wall Street. (05:31) The discussion reveals a nuanced view: AI represents revolutionary technology with insatiable demand, but unhealthy behaviors around debt financing and speculative investments are creating bubble-like conditions. While companies like Microsoft, Amazon, and Google are making sound cash-flow backed investments, others like CoreWeave are borrowing money to build data centers for loss-making startups like OpenAI. (18:54) The episode explores depreciation issues, with Michael Burry warning that companies are understating costs by extending useful lives of AI chips artificially, potentially overstating earnings by hundreds of billions. The conversation concludes with analysis of the prisoner's dilemma in AI pricing and the systemic risks of hundreds of billions in speculative debt.
• Core Theme: Distinguishing healthy AI investments from bubble behavior through debt analysis, depreciation concerns, and competitive dynamicsSpeakers
Gil Luria
Gil Luria serves as the head of technology research at D.A. Davidson, where he analyzes technology trends and provides investment guidance to Wall Street. He has established credibility for calling both positive and negative aspects of tech investments, particularly around AI infrastructure spending. Luria brings experience from previous financial bubbles to his analysis of current AI market dynamics.
Alex Kantrowitz
Alex Kantrowitz hosts Big Technology Podcast and has been covering technology trends and their economic implications. He conducts in-depth interviews with industry analysts and executives to break down complex technology and business developments for his audience.
Key Takeaways
Distinguish Healthy from Unhealthy AI Investment Behavior
The AI revolution is real and transformative, but not all investment approaches are sound. (05:31) Luria emphasizes that companies like Microsoft, Amazon, and Google are making healthy investments using cash flow from existing profitable businesses, with clear customer relationships and balanced risk management. In contrast, companies like CoreWeave represent unhealthy behavior - borrowing money to build data centers for loss-making startups without predictable cash flows or assets to back the loans. The key difference is whether investments are backed by existing customer relationships and cash flow, or by speculative promises from unprofitable companies.
Debt Financing Should Match Asset Types and Cash Flow Predictability
Financial fundamentals matter even in revolutionary technology cycles. (09:34) Luria explains that debt should be used for assets with predictable cash flows or tangible backing (like mortgages), while equity should fund speculative growth investments. When Oracle borrows money based on OpenAI's promise of $300 billion in revenue - from a company currently losing over $20 billion annually - this violates basic financial principles. This confusion between debt and equity financing creates systemic risk when applied at scale across hundreds of billions of dollars.
Depreciation Timelines Don't Match AI Hardware Reality
Companies may be artificially inflating profits by extending AI chip depreciation beyond realistic useful lives. (22:42) While accountants initially set 5-6 year depreciation schedules for AI chips, the rapid pace of improvement means newer chips can be 10 times better annually. Michael Burry warns this could lead to $176 billion in understated depreciation from 2026-2028. Even if older chips still function, they may only generate 1% of their original revenue capacity, making them economically worthless despite technical functionality. This discrepancy could force major earnings restatements across the industry.
Focus on Core Competencies Rather Than Vertical Integration
OpenAI's best path forward involves focusing on their ChatGPT product strength rather than expanding into infrastructure. (37:26) Luria advises that OpenAI should concentrate on maintaining their frontier model advantage and growing their 800 million weekly active users responsibly, rather than committing to building their own data centers and hardware. Their current overcommitment strategy of promising $1.4 trillion to various partners while losing money creates unsustainable expectations. Companies succeed by leveraging their core advantages rather than trying to control entire value chains.
Winner-Takes-All Mentality Drives Persistent Losses
The prisoner's dilemma in AI pricing creates a race to the bottom that may persist for years. (54:05) Major players like Meta, Google, Microsoft, and Amazon view AI as a winner-takes-all market, leading them to prioritize market share over profitability through unlimited usage models and flat-rate pricing. Mark Zuckerberg's signaling that he'll spend whatever it takes to win demonstrates this dynamic. Only companies with the deepest pockets can survive this war of attrition, which explains why smaller players like OpenAI face existential challenges despite having innovative products.
Statistics & Facts
- OpenAI has committed approximately $1.4 trillion in total revenue promises to various partners including $300 billion to Oracle, $200 billion to Microsoft, $38 billion to Amazon, and $25 billion to CoreWeave, while the company will only have about $15 billion in revenue this year and will lose more than $20 billion. (11:11)
- By Michael Burry's estimates, hyperscale cloud companies will understate depreciation by $176 billion from 2026 to 2028, with Oracle potentially overstating earnings by 26.9% and Meta by 20.8% by 2028 due to artificially extended useful lives of AI compute equipment. (21:01)
- ChatGPT has reached 800 million weekly active users, representing the fastest growth of any startup ever, yet OpenAI operates as a negative gross margin business where it costs more to answer queries than they generate in revenue. (36:26)
