AI Boom or Bust: Where Smart Money Is Moving
AI Bubble? Research Firm Says the Hype Has Already Peaked
Artificially low interest rates drove billions into artificial intelligence, but independent research firm MacroStrategy Partnership warns the boom has already hit its limits.
The firm, led by former UBS strategist Julien Garran, argues the AI surge isn’t just another bubble — it’s potentially 17 times larger than the dot-com crash and four times bigger than the 2008 housing collapse.
Cheap Money, Bad Allocation
Their analysis leans on economist Knut Wicksell’s principle that capital is best allocated when corporate borrowing costs stay above nominal GDP. Instead, a decade of ultra-low rates distorted the system, funneling trillions into misallocated assets: AI, housing, real estate, NFTs, and venture capital.
LLMs at the Breaking Point
The research also casts doubt on AI’s technological foundations:
- Task completion rates at one software firm ranged from 1.5% to 34%, and even the higher results couldn’t be reproduced consistently.
- Corporate AI adoption is already slipping, according to U.S. Commerce Department data.
- Expensive training cycles are yielding smaller gains: GPT-3 cost $50M, GPT-4 cost $500M, and GPT-5 — delayed despite a $5B price tag — showed little improvement.
With no clear competitive moat, escalating costs, and weak commercial applications, MacroStrategy sees diminishing returns setting in fast.
Risk of Recession Ahead
The firm warns this could trigger a reversal similar to the dot-com bust in 2001. If data-center spending and wealth effects stall, the economy — already slowing — may tip into recession. Policy options for the Fed or the Trump administration could prove limited, raising the risk of a drawn-out reflation effort, like after the early ’90s S&L crisis.
Where to Invest Instead
MacroStrategy advises:
- Overweight: Resources, India, and Vietnam
- Underweight: AI and platform stocks
- Go Long: Gold miners (GDX), short-dated Treasurys, volatility (VIX), and the yen versus most currencies
