How the AI Bubble Bursts (insights.som.yale.edu)

Tech-industry deals and massive compute bets have spun into a financial web that some leaders now call a bubble. Recent moves — OpenAI taking a 10% stake in AMD, Nvidia committing roughly $100 billion to OpenAI, Microsoft as both major OpenAI shareholder and big customer of CoreWeave (in which Nvidia also holds equity), plus OpenAI’s reported $300 billion, five‑year compute pact with Oracle — highlight extreme concentration of capital and interlocking ownerships. Venture and public-market flows have followed: PitchBook found nearly two‑thirds of U.S. deal value went to AI/ML startups in H1 2025, AI-related stocks drove ~75% of S&P 500 returns since ChatGPT, and AI capex accounted for ~1.1% of GDP growth in early 2025. Those dynamics matter technically and economically. Huge, front‑loaded investments in data centers and chips lock firms into long depreciation cycles and raise contagion risk if AI revenues underperform. Empirical warning signs include an MIT study showing 95% of 52 organizations saw zero ROI despite $30–40B spent on GenAI initiatives, and concerns about benchmark “data contamination” that can exaggerate model capabilities. Governance and safety gaps, plus the possibility of disruptive chip or quantum breakthroughs, could render large infrastructure investments stranded. The result: meaningful downside if hype outpaces capability — winners and losers will be determined by who actually delivers reliable, generalizable AI and who overinvests into ephemeral advantages.