The AI Bubble, the Productivity Paradox, and India’s AI Summit

Rebecca Patterson and Sebastian Mallaby work through three linked questions on AI’s spillovers: whether markets are in a bubble, why the real-economy productivity payoff remains unproven, and what the fizzling of global AI governance means after 86 countries gathered at India’s AI summit. Their shared conclusion — no broad bubble, but OpenAI is a company-specific one worth watching closely.

Key ideas

1. Not an AI bubble, but possibly an OpenAI bubble. Mallaby rejects the 2008 comparison outright — that crisis was financial engineering layered on a static asset, whereas AI is ‘an A+ technology’ improving fast on longer context, multimodality and agentic capability. But he rates individual companies, above all OpenAI, fragile: ‘an A+ technology, but an F- business model for now.’ OpenAI’s roughly 900 million users are not sticky — half a dozen good competing models make switching easy — so subscription revenue lags user growth even as the company reportedly projects $660bn in losses between now and 2030.
2. OpenAI’s financing gap is the number to watch. OpenAI raised $41bn privately in 2025, the largest private fundraising round in history, and is reportedly seeking around $100bn more at a valuation near $800bn — up from $500bn entering 2026, even as data-centre partner Oracle has fallen more than 50%. Mallaby puts a 50% chance on OpenAI hitting a financing cliff this year, though his base case if it fails is an orderly sale (Microsoft is the obvious buyer) rather than a systemic collapse, since the technology and the 900-million-user base retain value.
3. The SaaS-pocalypse is repricing enterprise software and banks. Fear that AI models can write code cheaply enough to disrupt software vendors has hit stocks like ServiceNow (down 25% in a month against a roughly flat Nasdaq) and pulled bank shares such as JPMorgan (down 4.5%) on fears of AI-driven payments disruption. Mallaby is sceptical the disruption is that fast or that clean: incumbents with entrenched relationships and regulatory moats, such as JPMorgan, may integrate AI into their own operations faster than AI-native challengers can break in.
4. The productivity paradox: no one yet knows if a boom is real. Patterson stresses that productivity is a statistical residual, calculated only after GDP is fully decomposed, so claims of ‘the biggest productivity boom in years’ are premature — the data will not be clear until sometime next year. In the meantime the US risks a ‘jobless expansion’: companies redirect budget from headcount to AI spend without necessarily cutting existing staff outright (Walmart’s CEO said AI could keep global headcount flat for years). Patterson watches hiring-intention surveys, weekly and continuing jobless claims, sentiment, and credit-market stress as early markers of whether this stays benign.
5. AI’s effect on interest rates and inflation is genuinely ambiguous. One camp, echoed by Fed chair nominee Kevin Warsh, argues AI productivity will be disinflationary and can justify cutting rates now; Fed Governor Barr counters that AI could instead raise capital demand and push rates up. Patterson argues against pre-emptive rate cuts based on a productivity story that has not yet shown up in the data, preferring the Fed’s qualitative regional intelligence over a single GDP print.
6. India’s AI summit exposed a stalled global governance agenda. Eighty-six countries attended, but the momentum that produced AI safety institutes and the Bletchley Park, Seoul and Paris summits after ChatGPT’s 2022 launch has ‘completely fizzled’ in 2025–26 — a casualty of Trump-administration impatience with regulation and an intensifying US-China race dynamic. Mallaby, drawing on a Foreign Affairs piece co-written with Sebastian Elbaum, frames the trade-off as an ‘AI trilemma’ between economic security, national security and societal security, arguing the US is currently sacrificing societal security (safety testing, red-teaming, guardrails against misuse) to maximise the other two. His two proposed fixes: a tax-and-credit regime that funds private-sector safety research, and a strengthened AI Safety Institute with real veto power over dangerous model releases, modelled on the FDA.

Context

Mallaby’s OpenAI-specific bubble framing here is the same thesis he later develops at greater length in Sebastian Mallaby on OpenAI's Cash Crunch, the AI Bubble Debate, and the China AI Race — this earlier Spillover conversation is where the 50%-chance-of-financing-cliff call and the $660bn loss projection first surface in the wiki’s coverage, before the OpenAI-specific numbers hardened further in subsequent episodes.

Related

• Sebastian Mallaby — host; his OpenAI-bubble-not-AI-bubble thesis anchors the discussion
• Rebecca Patterson — host
• Sebastian Mallaby on OpenAI's Cash Crunch, the AI Bubble Debate, and the China AI Race — Mallaby develops the same runway/bubble thesis
• OpenAI's Leaked Financials, SpaceX's Valuation, and the AI Bubble — the numbers behind the bubble debate