Software Investor Warns About Flawed Billion-Dollar AI Valuations

In July 2007, Citigroup CEO Chuck Prince told the Financial Times that as long as the music kept playing, his bank had to keep dancing. Four months later, he was out of a job. This week, Venky Ganesan of Menlo Ventures borrowed that line for a long post on X that passed 500,000 views within days. His message: venture capital is in its most disorienting stretch in years, and the people dancing have good reasons to keep going.

Ganesan does not claim the market is broken; he says some of his portfolio companies are growing faster than anything he has seen in his career. He also says companies with no revenue, no product and a founding team that fits in a conference room are raising billions at valuations between $10 billion and $50 billion. Both statements hold at once, he argues, and anyone who applies a single framework to both will go crazy.

He turns to two ideas, neither his own. The first is reflexivity, the concept George Soros has developed since the 1980s. Weather does not care what you think of it. Markets do. Prices change what participants believe, and those beliefs change the prices. While that loop runs, Ganesan writes, it looks exactly like progress.

How The Loop Works In AI

Ganesan lays out the mechanics with a disclosure: Menlo invests in Anthropic. By his account, investors watched a frontier lab climb from a $4 billion valuation to $18 billion, then $60 billion, $180 billion, $380 billion and now something near a trillion. They concluded that this is what a neo lab looks like. The next lab then gets priced off that path instead of off anything it has built. A markup in the following round becomes the proof. In his words, valuation stops being an output of the metrics and becomes the metric. Investors no longer discount cash flows. They discount the last round.

Soros makes one point that people skip, Ganesan notes. Nobody can know when or how a reflexive process ends. Investors only know that it does.