Nvidia’s $200bn Problem: Its Best Customers Want Its Job
Google’s chip ambitions just wiped billions off Nvidia’s market cap. This is what tech betrayal looks like in real time.
Nvidia has spent the past two years selling shovels in the AI gold rush. Now the prospectors want to open their own shovel factories.
The stock was off almost 5% in afternoon trading on Tuesday, wiping almost $200 billion from its value amid reports that Google is in talks with Meta to supply its own AI chips for data centres from 2027.
That sounds boring until you remember that Google is both one of Nvidia’s biggest customers and one of the few companies on earth with the scale to take a real swing at it.
The report, from The Information, says Google is pitching Meta on a multi-billion dollar deal for its tensor processing units, or TPUs. Up to now, TPUs have been a kind of captive species.
Developers access them through Google Cloud, neatly confined to Google’s own data centres. The new idea is very different: let other giants plug TPUs into their own racks, on their own turf.
According to the same report, Google is not stopping at Meta. It has started pitching other cloud customers too, and internally the company claims it could realistically capture as much as 10% of Nvidia’s annual revenue.
Nvidia holders did not need a spreadsheet to understand what that implies, with the stock tumbling as much as 7% in a torrid morning session, before clawing back some of the losses.
Advanced Micro Devices, which was meant to be Nvidia’s main challenger, also caught a hit, dropping more than 8%.
Underneath the price action is a tension that has been building for a while. Nvidia’s biggest customers are also its most credible future rivals. Google has TPUs.
Amazon has Trainium and Inferentia. Microsoft has Maia. These chips started life as strategic insurance policies against supply shocks and pricing power. Now they are quietly mutating into products.
The shift is already visible. Amazon just finished a huge data centre buildout that includes renting out half a million of its custom AI chips to Anthropic. Google has its own deal with Anthropic.
OpenAI has reportedly tested Google’s silicon as well. The message is simple. Even the most AI-hungry labs do not want to be tied to a single supplier forever, no matter how fast Jensen Huang walks on stage in his leather jacket.
For Nvidia, this all lands on top of a different line of criticism. The company has been accused of inflating demand through what sceptics call circular AI trades: invest in customers, sell them chips, and let the market extrapolate. Michael Burry, of Big Short fame, has publicly bet against Nvidia and compared the current AI craze to the dot-com bubble.
