Mike Volpi is right that Nvidia cannot stop the AI correction
By Bram de Vries · Reporting from Amsterdam ·
On the a16z podcast, technology investor Mike Volpi warned that Nvidia's attempt to finance scores of specialized AI clouds cannot override market consolidation or prevent a painful debt cleanup.
On the a16z podcast, veteran technology investor Mike Volpi laid out an uncomfortable truth. Funding fragile startups to maintain customer diversity is commercial futility. Volpi managed corporate development at Cisco during the late 1990s internet boom. He pointed out that Nvidia is currently backing around 200 specialized NeoCloud providers. Nvidia does this for an obvious reason. No hardware supplier wants to depend on just three buyers. But as Volpi noted, scale industries always consolidate. Capital can defer economic gravity, but it cannot repeal it.
Capital Can Delay Gravity, Not Cancel It
In financing trade assets, I recognise the pattern Volpi describes. During the fiber optic boom, telecom operators laid cable on borrowed money. When overcapacity met financial reality, those firms collapsed. The debt wiped out equity holders, yet the physical fiber remained in the soil. Solvent buyers eventually acquired the assets to serve commerce for two decades.
Today, the buildout is compute rather than glass fiber. On the podcast, the hosts pressed Volpi on whether today's GPU buildout faces a similar reckoning. Volpi's answer was clear. The physical infrastructure being installed today will be fully utilized over time. But it will not be owned by the debt-saddled balance sheets currently holding it. Nvidia's artificial support of specialized clouds merely delays the day of reckoning.
The Scarcity Premium on Aging Silicon
The counterargument raised during the episode rests on asset depreciation. The hosts observed that older chips like the Nvidia H100 still rent for $2.50 to $3.00 per hour instead of depreciating. In normal technology cycles, older silicon becomes obsolete. Opponents of the crash theory argue that infinite demand for AI tokens makes compute assets uniquely immune to traditional write-downs.
That view mistakes short-term bottlenecks for permanent structural immunity. Volpi correctly identified why old chips still command high rental prices. The cause is extreme scarcity in advanced semiconductor manufacturing at foundries like TSMC. Scarcity masks poor balance sheets. As newer chips arrive with superior power efficiency per watt, the economics of older cards will deteriorate rapidly. A server rack drawing massive electricity to run inefficient calculations loses its market value the moment efficient supply catches up.
Corporate balance sheets exist to absorb risk, but capital markets eventually demand a return. Trying to subsidize 200 separate cloud buyers to prevent an oligopoly is like subsidizing 100 airlines to prevent carrier consolidation. The market will clear weak balance sheets regardless of vendor financing. When the correction arrives, equity will burn, debt will restructure, and the GPUs will pass to operators who can run them profitably.