As of December 2025, just three years after ChatGPT's debut, discussions about an "AI bubble" have reached a fever pitch. Deutsche Bank argues that the current AI boom is neither a pure bubble nor entirely risk-free—the key lies in distinguishing between different types of "bubbles."
On December 12, Deutsche Bank released a groundbreaking report categorizing AI bubbles into three dimensions: valuation bubbles, investment bubbles, and technological bubbles. The report highlights that valuations of large public tech companies are supported by earnings, investment growth aligns with trends and is cash-flow-driven, and technological progress continues. The real risks, however, lie in overvalued private companies, potentially unsustainable circular financing structures, and looming technological bottlenecks and supply constraints.
**Valuation Bubble: Divergence Reveals True Risks** Deutsche Bank's core thesis is that the AI boom comprises three distinct bubbles rather than a single one.
On valuation, the report notes that the Shiller CAPE ratio has exceeded 40, nearing the 44x peak of the dot-com bubble. This inflation-adjusted long-term earnings metric signals overheating. Historically, such high multiples precede market corrections.
However, Deutsche Bank attributes current valuations primarily to earnings growth rather than speculation. Since October 2022, the S&P 500 has maintained an annualized growth trend of 22.7%, now at the lower end of this channel. Crucially, large-cap tech’s 60% valuation premium is underpinned by a 20%+ earnings growth advantage.
The bank observes that tech valuations today are less extreme than during the dot-com era, with earnings growth broadening across sectors. In contrast, private companies appear overvalued: OpenAI trades at 38x 2025 revenue forecasts ($13B), while Anthropic commands 44x. Public tech giants seem reasonable—Nvidia at 22x, Microsoft 12x, Alphabet 9.9x, and Amazon at 3.5x—suggesting public markets remain rational.
**Investment Bubble: Cash Flow vs. Debt Risks** Deutsche Bank projects hyperscalers’ capex will hit $500B by 2026, potentially accumulating to $4T by 2030—10x the inflation-adjusted cost of the Apollo program. While unprecedented, this remains within the 12.3% annual tech capex growth trend since 2013.
Notably, ROI for big tech has risen consistently via cloud demand and AI-driven cost savings. Unlike the debt-fueled dot-com era, current AI investments are largely funded by free cash flow. Alphabet’s Q3 operating cash flow of $48B exemplifies this, with hyperscalers’ capex-to-cash-flow ratios below 1, reflecting financial health.
**Tech Bubble: Scalability Doubts vs. Breakthroughs** Generative AI still faces hallucinations and scalability challenges, with physical bottlenecks (e.g., chip-to-chip data transfer speeds) looming. Yet, Alphabet’s Gemini 3—launched in November 2025—showcases progress, outperforming GPT-5 Pro in multimodal reasoning by 3x on ARC-AGI-2 tests.
Demand-side metrics are equally striking: Alphabet now processes 1,300 trillion tokens monthly (up from 97 trillion in April 2024), while under 10% of U.S. firms use AI, signaling vast growth potential. Cost declines further fuel adoption, with top-tier LLM expenses dropping 1,000x, adhering to Jevons Paradox—efficiency gains beget demand.
**Potential Bubble Triggers** Despite robust fundamentals, Deutsche Bank flags five risks:
1. **Opaque Circular Financing**: Complex deals like OpenAI’s $1.4T compute commitments (involving Nvidia, AMD, Oracle, etc.) could trigger systemic risks if any link fails. 2. **Debt Spiral**: Even cash-rich hyperscalers are issuing debt, with 2025’s U.S. investment-grade bond sales exceeding $35B. Rising net-debt/EBITDA ratios (Microsoft, Alphabet, Meta, etc.) may force excessive borrowing. 3. **Diminishing Returns**: Scaling costs grow exponentially—training expenses surged from $10M (Llama 2) to $1B+ (Grok 4). Data center spend now implies just a 20% chance of AGI by 2030, down from near 100% in 2022. 4. **Sociopolitical Pushback**: Over 20% in the EU/UK fear job losses to AI, with resistance strongest in developed markets. Regulatory curbs or consumer backlash could slow adoption. 5. **Supply Crunches**: By 2030, power demand may quadruple 2020 levels. U.S. households already face record electricity prices (~17¢/kWh), threatening AI’s energy-intensive growth.
Comments