Andreessen Horowitz's Own Data Shows the AI Bubble Warning Signs

By Daniel Okafor

Retail investors held $192 billion in leveraged technology ETFs at the end of August — roughly double the level of two years earlier. The number appears on page 13 of Andreessen Horowitz's latest "State of the Markets" report. Brett Arends, writing his ROI column for MarketWatch, argues the firm presented that figure as bullish evidence. He reads it as a warning.

Andreessen Horowitz manages over $100 billion in assets and is a major backer of AI and technology companies. Its report carries headlines such as "Tech is the everything cycle," "We keep underestimating the capex to meet demand," "Revenue growth is unprecedented for speed and scale" and "Surprise! Data centers may be lowering your electricity bill." The report cites market data from Bloomberg, Citadel Securities and other reliable sources.

Arends' verdict is blunt. "The numbers don't tell quite the hubristic message implied by the headlines," he writes. "Quite the reverse."

What does the report actually say?

The bullish framing sits on data points that cut both ways:

• Retail investors have poured a record $192 billion into leveraged ETFs that use derivatives to deliver two or three times the daily performance of technology stocks, up or down. Most of that money targets chip makers and other technology companies.
• The big hyperscalers have already spent at least $1 trillion on data centers and other AI hardware, and are preparing to spend $4.6 trillion more over the next few years.
• That spending has already exceeded the hyperscalers' annual cashflow, and then some.
• Nearly all AI demand comes from 1% of superusers, who together outspend the entire top 10% of users by a factor of eight.
• AI prices, measured by price per token of usage, are already plummeting — before the next $4.6 trillion of data centers comes online.
• The share of U.S. households paying for AI has reached an all-time high of 2.2%.
• Among those paying subscribers, average monthly spending has reached $31.
• Technology companies now account for about 60% of the market value of the world's top 100 firms. That share has doubled in 10 years and quadrupled in about 15.

Arends asks what happens if the 60% share doubles again in the next decade. "So that technology companies account of… er… 120% of the market value?" he writes. "Good luck with that."

Why does the framing matter?

The column's core argument is about who reads the data. Arends notes that most people working in finance are young. The Andreessen report refers to the past five to 10 years as "the long term" — without irony, in his reading. He argues that a 28- or 35-year-old investor has not yet lived through a bear market and learned "that luck isn't the same as genius, that leverage is a double-edged sword, and that trees don't grow to the moon."

Half the report, he writes, shows sudden upward spikes in spending or earnings. "To those with long memories, such things rarely end well." He adds that if such spikes always kept growing, "capitalism would be broken."

Is this time different?

Arends compares market manias to a Scooby-Doo episode. The names and details change, but the story stays the same: the monster chasing everyone turns out to be the amiable old guy from the opening scene. "If so, I have some very exciting leveraged technology ETFs for you," he writes.

He traces the pattern back through the 1800s railroad boom, the South Sea Bubble of the 1700s and the Dutch tulip mania of the 1600s, and quotes Ecclesiastes: "there is no new thing under the sun."

What should investors watch?

Timing the peak is impossible, Arends concedes, citing the New Testament line that nobody knows the day or the hour. But he draws one operational lesson from the report's own framing. "When the bulls start reporting that massive stock gambling by the retail public, using levered ETFs, are a reason to be bullish, you have to start wondering if the end is nigh."

The implication for ordinary investors is straightforward: the same document that promotes the AI boom also documents the leverage, the concentration and the thin consumer demand beneath it.