The AI All-You-Can-Eat Buffet Is Ending with Gary Marcus | The Real Eisman Playbook Ep 62

| Stock Investing | June 01, 2026 | 78.4 Thousand views | 48:29

TL;DR

Steve Eisman and AI researcher Gary Marcus warn that the AI investment boom faces a reckoning as scaling laws break down, the ecosystem reveals dangerous dependency on VC-subsidized unprofitable labs, and infrastructure investments risk becoming stranded assets.

Infrastructure & Supply Chain Bottlenecks 3 insights

Hyperscaler capex reaches staggering heights

Google plans to double capital spending to $180 billion in 2026 while Amazon commits over $220 billion, driving demand for chips, networking gear, and construction.

Power becomes the binding constraint

Data center expansion faces physical limits on electricity and water, creating winners in gas turbines (GE Vernova), nuclear energy, and electrical infrastructure (Eaton, Quanta).

Memory and CPU shortages drive chip rally

Memory chip makers like Micron see soaring prices amid shortages, while CPU demand surges alongside GPUs, pushing semiconductor stocks to 16-17% of the S&P 500.

🔬 The Scaling Crisis & Technical Pivot 4 insights

Scaling laws hit a wall in late 2024

Major AI companies realized that simply adding data and compute no longer delivered proportional improvements, resulting in GPT-5's disappointing delay and underperformance.

Neuro-symbolic AI replaces pure scaling

Progress now comes from hybrid approaches like Claude Code that combine neural networks with traditional symbolic programming rather than just bigger models.

AGI hype deflates amid failed predictions

Despite Microsoft CTO Kevin Scott's whale-chart predictions and Sam Altman's January 2025 claim that conventional AGI was achievable, systems still hallucinate and lack reliable reasoning.

Trillion-dollar infrastructure bets face obsolescence

If medium-sized models plus symbolic AI prove sufficient, the $2 trillion invested in massive GPU data centers may generate stranded assets and unpayable loans.

⚠️ Economic Vulnerabilities & Market Risks 4 insights

The SaaS apocalypse destroys software moats

AI dramatically lowers software creation costs, eroding competitive advantages for traditional subscription companies and crushing private equity portfolios acquired between 2018-2023.

Circular dependency on VC-subsidized labs

An estimated 50% of hyperscaler revenue growth depends on Anthropic and OpenAI, private companies that remain deeply unprofitable and rely on continuous venture funding to purchase cloud services.

Private credit faces refinancing crisis

With software valuations down over 50%, PE-backed companies face impossible refinancing conditions, forcing equity injections or walkaways while retail investors flee private credit funds.

Data center projects face community backlash

Massive AI data centers consume extraordinary amounts of electricity and water, triggering local opposition that threatens project timelines and the entire expansion narrative.

Bottom Line

Avoid the 'bigger is better' GPU infrastructure narrative and exercise extreme caution on software and private credit exposed to AI disruption, as the ecosystem's survival depends on continued VC subsidies to money-losing AI labs that may soon dry up.

More from Steve Eisman

View all