Steve Eisman Warns of LLM Scaling Debate but Stays Bullish on Major AI Stocks

Steve Eisman, the famed investor who predicted the 2008 housing crisis, is sounding the alarm about the growing debate over whether large language models can continue scaling indefinitely. But even as he raises concerns about the technical limitations of AI development, he remains firmly bullish on the companies powering the next wave of artificial intelligence.

Eisman believes the scaling conversation is real — yet insufficient to derail what he calls “the biggest story in markets.”

The Scaling Laws Debate Escalates

For years, AI researchers have relied on a simple assumption: bigger models trained on more data produce stronger results. Those scaling laws drove breakthroughs from GPT-3 to GPT-5. But critics argue the trend may be peaking.

Prominent researchers, including Gary Marcus, warn that simply increasing model size is no longer a guaranteed path to progress. High-quality training data is becoming scarce, marginal gains are shrinking, and compute costs are rising exponentially.

These concerns have intensified through 2025 as model performance appears to slow relative to size increases.

Eisman Still Bullish on Nvidia, Apollo, and AI Infrastructure

Despite the technical tension, Eisman has not wavered on his long-term positions. The investor continues to hold bullish five-year outlooks on major AI infrastructure companies — especially Nvidia and Apollo.

The reasoning is simple: regardless of how AI evolves — whether through larger models, smaller models, or new reasoning techniques — the demand for hardware and compute infrastructure remains enormous.

Eisman has repeatedly said the AI boom is “the biggest thing happening in the markets,” dismissing concerns about corporate accounting for AI investments, including those raised by fellow Big Short figure Michael Burry.

Inference-Time Scaling Emerges as the New Trend

One of the biggest breakthroughs this year arrived on December 4, when MIT researchers introduced inference-time scaling. Instead of building ever-larger models, this method allows existing models to reason longer before producing an answer.

The result: improved performance without requiring the massive compute needed for traditional scaling.

DeepMind, OpenAI, and other labs are already integrating this approach into their next-generation systems. Eisman sees this flexibility as proof that AI innovation is far from slowing down — even as traditional scaling laws face constraints.

Eisman’s AI Investment Outlook

Economic Risks Beyond AI

While bullish on AI itself, Eisman warns the broader economy is weakening. He described the U.S. economy as a “tale of two cities,” where tech and AI spending mask sluggish growth elsewhere.

He noted that outside of AI, the economy is “not even growing 50 basis points,” raising questions about uneven economic strength heading into 2026.

Still, Eisman believes AI will continue dominating market returns — even as the scaling debate evolves.

“The AI story continues,” Eisman said. “It remains the biggest story for markets going forward.”

What This Means for Investors

For long-term investors, Eisman’s message is clear:
The scaling debate may reshape how AI improves — but not whether AI grows. Infrastructure, compute, energy, and hardware remain foundational regardless of the technical path forward.

That logic keeps Nvidia, Apollo, and other AI infrastructure giants firmly in Eisman’s bullish column.