READ THE ORIGINAL
Published in full and free by Oaktree Capital Management. This page indexes its arguments; it is not a copy.
THE ARGUMENT
Marks distinguishes bubbles that leave useful infrastructure behind from bubbles that leave nothing, argues that current AI valuations are less extreme than the 1998-2000 comparison implies, and holds that the question in the title cannot honestly be answered until years afterward.
What it actually claims
He splits bubbles into two kinds, which is the memo’s main contribution: inflection bubbles, built on technology that ultimately delivers, versus mean-reversion bubbles, financial fads that leave nothing behind.
On the central valuation question he argues the historical comparison cuts the other way — today’s leaders trade cheaper than the 1999 leaders did.
- P/E ratios for Microsoft, Cisco and Oracle in 1998-2000 were much higher than current ratios for Nvidia, Microsoft, Alphabet, Amazon and Meta
His concern is located in the financing of the buildout rather than in the equity multiples.
- roughly $5 trillion of data center buildout
- against about $350 billion held collectively by Microsoft, Alphabet, Amazon, Meta and Oracle
He cites private-market pricing as the clearer symptom of excess.
- Thinking Machines raised $2 billion at a $10 billion valuation with no released product, later valued at $50 billion
He notes how narrow the market’s dependence on one theme has become.
- AI stocks accounted for 75% of S&P 500 gains, 80% of profits, 90% of capex
He grounds the downside in what actually happened to earlier inflection bubbles, where the technology survived and the shareholders did not.
- RCA lost 97% after 1929
- aviation stocks fell 96% by 1932
Figures are the author's own, as cited in the memo.