Nvidia will be fine, but minor cloud companies could be in trouble.

We are entering a period of time where multiple physical limitations hit the AI build out supply chain:

– Energy generation

– Grid capacity

– Data center space

– Copper

– Helium, Bromine, Neon

– Sulfuric Acid

– Rising cost on LNG and oil

– Copper

– Aluminum

– Gallium and other rare earth

And just to make it funnier, consumer strength is waining and inflation is still an issue. I believe consumer never trully recovered since covid.

Each element here is required for continued build out. World economy is scrambling to close the gap at break neck speed. But it is evident that it wasn't catching up and recent geopolitical changes have made situation worse.

I'm not saying that AI will slowdown because of the demand or that technology is not real. I personally believe that next wave with physical AI, AI industrialization, AI empowered industrial and military drones, robotaxi and robotics. All that definitely have more clear ROI map. Demand is there and more to come.

What I am saying is that… physical limitations of manufacturing could make this whole endeavor pricey and increasingly selective.

In many ways Nvidia have been frontrunning many issues with strategical focus on R&D and M&A. Biggest example is how Vera Rubin is 10 times more energy efficient than Blackwell. They are moving toward photonics and they have placed bets all over the physical AI space to pounce once those theme find traction.

And that's the crust of my case. If a platform that Nvidia will deliver after Vera Rubin continue its tradition of technological miracles on annual cadence, older systems could rapidly lose their economic relevance. It's not that the hardware is no longer unusable, but just the substitute could be that more technologically advanced that keeping them on could be losing money.

I've seen anecdotal evidence that some data centers already have been selling their Hopper GPUs to make space for newer technology. I don't know how wide scale is the issue, but I can presume it is growing trend rather than shrinking.

And… for all those datacenter minors, who spend billions of borrowed money to get the hardware that is rapidly obsoleting, it presents a unique problem that I don't see an elegant solution to.

Michael Burry posted last year that hyperscalers are using accounting tricks to depreciate hardware over larger periods of time to inflate their revenue. And at the time I dismissed the argument, because the counter argument that newer data center hardware has a longer lifespan made sense to me. But now I think I get what he was talking about.

If costs on everything are rising and the technology is rapidly advancing to be more cost efficient, the hardware will become obsolete not because of its technical lifespan, but because the brightest minds of our time are working overtime to push the tempo.