Zero‑sum framing of GDP and “arithmetically constrained pies”

Burry suggests economies have fixed, non‑expanding “pies,” implying AI spending must merely reallocate output. In reality, productivity‑enhancing tech can expand real GDP over time, so the choice is not purely zero‑sum.

Overgeneralising from narrow productivity evidence

Several participants lean heavily on one early, specific METR coding study to question broad AI productivity, while acknowledging data are “conflicting and sparse.” Extrapolating from a small, specialised sample to all AI use risks hasty generalisation.

Assuming competition must erase most producer surplus

The escalator analogy implies rival firms’ AI adoption will always fully pass gains to customers. In practice, sectors with network effects, IP, or scale economies (cloud, chips, platforms) can maintain supra‑normal margins even with widespread adoption.

Big Tech executives

Leaders at Microsoft and Nvidia argue massive AI capex is rational because AI will materially raise productivity and expand addressable markets from ads to labour itself; near‑term ROIC compression is framed as an investment phase. Sources: Dwarkesh–Nadella interview · Nvidia CEO on $3–4T data‑center spend

Mainstream labour economists and policy bodies

OECD and similar institutions see AI as raising both automation and task creation; they expect reallocation and skill shifts rather than sudden mass unemployment, and emphasise education, retraining, and social insurance over protectionism. Sources: OECD Employment Outlook 2023

AI accelerationists / techno‑optimists

Some researchers and investors believe scaling plus recursive self‑improvement will soon produce AGI‑level systems, driving explosive growth that easily repays today’s trillions in capex; from this view, the main risk is under‑investing or over‑regulating. Sources: Epoch on scaling & distributed training · Nvidia CEO on AI augmenting large shares of GDP

AI capex boom vs historical tech bubbles and infrastructure cycles

Analysts compare today’s AI data‑center buildout to railroads and dot‑com: very high capex relative to current revenue, with risk of overbuild and stranded assets but also potential long‑run productivity gains. Sources: Fortune on AI capex & GDP · Network World on $1.1T data‑center capex

Mixed evidence on AI’s productivity impact, especially for coding

Besides METR’s negative result for experienced OSS devs, other studies and vendor reports find gains on simpler or greenfield tasks, suggesting AI’s productivity effect is highly context‑dependent. Sources: METR RCT · InfoWorld summary

Enterprise software and SaaS remain a minority of total IT and GDP

Even with strong growth, enterprise software (~$900B) and SaaS (~$250–400B) are small versus total IT (~$5T) and global GDP, which limits how much AI software alone can directly justify multi‑trillion hardware spend. Sources: Gartner enterprise software market share · Gartner IT spending forecast

Early evidence on AI and jobs: exposure high, displacement slow

Research finds many occupations exposed to AI, especially white‑collar roles, but aggregate employment and labour demand have not yet fallen; adjustment may be gradual and uneven across groups. Sources: OECD Employment Outlook 2023

Competing narratives from industry leaders

While Burry emphasises ROIC compression and bubble risk, Nvidia’s Jensen Huang and Microsoft’s Satya Nadella publicly argue that AI infrastructure is akin to power plants or factories and will unlock trillions in value if productivity gains arrive. Sources: Nvidia blog, “AI is infrastructure” · Nadella interview with Dwarkesh