Steve Eisman warns LLM scaling debate could reshape AI but refuses to sell his biggest bets

Steve Eisman, the famous investor who predicted the 2008 housing crisis, is voicing growing concerns about the debate surrounding LLM scaling while standing firm on his bullish bets on major AI stocks. The tension reveals a nuanced view: worries about technical challenges don’t diminish confidence in industry leaders.

The Scaling Laws Debate Intensifies

The artificial intelligence industry is grappling with questions about how far large language models can improve by simply making them bigger. Traditional scaling laws have long suggested that performance improves exponentially with compute power and data. Critics now argue these assumptions face practical limits.

Researchers point to challenges including limited availability of high-quality training data and diminishing returns on model size increases. Gary Marcus and other prominent voices published arguments that “scale is all you need” is becoming dead as an approach to achieving artificial general intelligence.

Eisman’s Bullish Stance on Hardware Leaders

Despite these concerns, Steve Eisman highlighted Nvidia and Apollo as two stocks providing exceptional five-year investment opportunities in April 2025. His confidence in these companies reflects belief that hardware infrastructure will remain essential regardless of how AI research evolves technically.

Eisman has called the AI trade “the biggest story in markets” and argues that infrastructure investments will be required whether models scale traditionally or adopt new approaches like inference-time scaling. The investor disagreed with Michael Burry‘s concerns about how tech companies account for AI infrastructure spending.

Inference-Time Scaling as the New Frontier

MIT researchers introduced inference-time scaling on December 4, 2025, enabling language models to reason longer before generating answers. This technique reallocates computational resources during inference rather than requiring entirely larger models during training phases.

DeepMind, OpenAI, and other leading labs are actively developing inference-time scaling methods announced throughout 2025. The approach suggests AI performance might improve through different mechanisms than traditional scaling of model parameters.

AI Investment Category	Eisman Position
Hardware (Nvidia, Apollo)	Bullish for 5+ years
AI Infrastructure	Essential regardless of scaling approach
Scale Debate Impact	Concerned but not deterred
Nuclear Power Value	Bullish on supporting AI energy needs

Concerns About Economic Disconnect

Eisman has warned that the U.S. economy resembles “a tale of two cities,” where AI and tech spending masks weakness elsewhere. The broader economy is “not even growing 50 basis points outside of AI” according to his October 2025 assessment of GDP projections.

This economic concern doesn’t diminish his confidence in AI companies themselves but highlights his view that AI investments are concentrated in a narrow set of winners while the rest of the economy stagnates significantly.

“The AI story continues” and remains the biggest story for markets moving forward despite technical debates about scaling methodologies.

— Steve Eisman, Big Short investor and AI market analyst

What Does This Mean for AI Stock Investors?

Eisman’s positions suggest a practical distinction between two concerns. The scaling debate reflects genuine technical questions about AI development paths and efficiency, but these questions don’t eliminate the requirement for massive infrastructure investments.

Whether AI improves through larger models, better training data, or techniques like inference-time scaling, the underlying hardware, chips, power infrastructure, and capital equipment remain critical. This logic supports his continued bullishness on companies like Nvidia and infrastructure plays that profit from AI buildout regardless of which scaling approach wins.

Sources

• CNBC – Big Short investor Steve Eisman’s latest views on AI trade concerns
• MIT News – Inference-time scaling breakthrough in large language models
• Business Insider – Eisman’s bullish positions on Nvidia and Apollo