CharlieKey takeaways from your favorite podcasts
← 11 November 2026a16z

Mike Volpi on Why AI Infrastructure Will Outlast the Hype Cycle

Mike Volpi argues the AI buildout resembles the late-90s internet boom: a correction is likely, but the underlying infrastructure will stay useful, and the cycle is more promising and consequential.

Watch on YouTube ↗

Charlie's summary

  1. The AI infrastructure will be fully used even if a correction comes; the open question is who owns it

    Mike notes the 90s debt bought fiber that proved valuable later, but was operated by others after the debt-laden players went bankrupt. He expects an economic correction among the roughly 200 neoclouds with weak balance sheets. Still, the buildout will be “fully utilized, if not more.” He is more optimistic than for the 90s, since the product is intelligence rather than communication and markets are more efficient.

    ▶ 1:01
  2. GPUs are outliving their expected depreciation, which makes the buildout easier to finance

    Hopper chips were assumed to have a five-year life and rent at about $1.80 an hour. Four years on, Mike says, they are not defunct and rent for $2.50 or three dollars. A longer asset life helps financiers whose models assumed a shorter one. Mike calls it an unintended consequence of AI being good at software. He and Guido caution that abundance could change this, but scarcity in chip manufacturing persists.

    ▶ 10:47
  3. Cisco's M&A worked by making enough bets to get outliers and never overpaying

    Mike says roughly a third of Cisco's acquisitions worked very well, a third were mediocre, and a third failed. Because the team did many, a few produced venture-like returns, and about 50% of revenue came from acquired sources by 2000. Deals were paid mostly in equity and rarely exceeded 4–5% of market cap. His rule: if your stock moves more than the price paid in a day, “nobody will care.” He applies this view to Nvidia's acquisitions.

    ▶ 14:50
  4. Financing an industry's structure is futile because scale businesses consolidate

    Mike argues Nvidia funding many small customers cannot change the shape of the market. Industries “find their own shape.” Hyperscaling is a scale business, so there will be about three big players, like airlines, however many you fund. He still sees the funding as well-intentioned and useful to startups who get capacity. Guido defends it as a way to avoid depending on three or four customers.

    ▶ 19:26

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.