Here's why concerns about an AI bubble are bigger than ever

Investor euphoria around generative AI has produced dramatic market moves and massive capital deployment — Nvidia’s stock has surged ~300% in two years as companies pour billions into compute — with OpenAI said to be doing $20bn of revenue and planning $1.4tn of data-center spending over eight years and Big Tech slated to spend roughly $400bn this year. Analysts and economists warn the buildout is being financed with risky structures and heavy borrowing (Goldman: hyperscalers took on $121bn of debt last year, a >300% rise; Morgan Stanley: ~$3tn of AI infrastructure through 2028 with firms covering only half from cash), including off‑balance‑sheet SPVs and circular deals (e.g., a reported $100bn Nvidia–OpenAI arrangement, Meta/Blue Owl $27bn SPV), which could leave lenders and sponsors exposed if demand disappoints and capacity is overbuilt. High‑profile investor selling and shorts (Peter Thiel, SoftBank, Michael Burry) alongside cautionary comments from industry leaders and Nobel laureates underscore the risk that current financing and valuation dynamics could presage a painful correction or broader credit losses if AI revenue growth fails to match the spending binge.

Analysis

Investor euphoria around generative AI has driven extreme capital deployment and valuation moves: Nvidia’s market value has surged roughly 300% in two years and OpenAI is reported to be generating $20 billion of revenue while planning $1.4 trillion of data-center spending over the next eight years. Major cloud and tech players (Amazon, Google, Meta, Microsoft) are slated to spend about $400 billion this year on AI infrastructure, with some devoting ~50% of current cash flow to data-center construction.

Financing patterns amplify risk: Goldman analysts say hyperscalers took on $121 billion of debt last year (a >300% rise from typical levels), Morgan Stanley projects ~$3 trillion of AI infrastructure spend through 2028 with firms covering only half from cash, and sponsors are using SPVs and third‑party debt (e.g., Meta/Blue Owl $27 billion) that can keep liabilities off corporate balance sheets. Circular arrangements (reported $100 billion Nvidia–OpenAI tie-up, CoreWeave/OpenAI stock-for-capacity deals) and guarantees that shift downside back to sponsors raise amplification and moral‑hazard concerns.

These dynamics increase the probability of overbuilt capacity and credit losses if AI monetization fails to match capex expectations; prominent sellers and shorts (SoftBank, Peter Thiel, Michael Burry) and cautionary statements from industry leaders and economists underscore tangible downside risk to equity and lender positions.

Key Decisions for Investors

• Reduce concentrated long exposure to the most levered AI beneficiaries (notably NVDA and META) or hedge positions with protective options given valuation and financing risks
• Audit and limit direct credit/counterparty exposure to SPV financiers and lenders backing data‑center builds (for example parties like Blue Owl) until contractual guarantees and balance‑sheet implications are transparent
• Favor companies with diversified revenue streams and strong free cash flow that are less dependent on continued exponential AI capex, and avoid fresh allocations to speculative infrastructure names lacking clear paying customer demand
• Monitor SEC filings and lease/guarantee disclosures, utilization metrics for new data centers, and ARPU or customer‑paid adoption indicators as predefined triggers to alter positioning