|
Aardvark DailyThe world's longest-running online daily news and commentary publication, now in its 30th year. The opinion pieces presented here are not purported to be fact but reasonable effort is made to ensure accuracy.Content copyright © 1995 - 2025 to Bruce Simpson (aka Aardvark), the logo was kindly created for Aardvark Daily by the folks at aardvark.co.uk |
Please visit the sponsor! |
Yes, there is an AI bubble.
Obscenely large sums of money have been poured into the creation of cloud-based AI systems that are being integrated into almost all aspects of the computer systems we use.
Obviously, at some stage, the cost of all that capital investment has to be recouped and right now we're left wondering where that money will come from.
Many of the big players, including Google, initially funded their AI expansions using their own cash reserves but now everyone is borrowing heavily and incurring significant debt levels in order to keep up with their competition.
The level of spend has pushed even Alphabet into its first ever negative quarter of cashflow since it whent public in 2004.
Borrowing, bond issues and a whole bunch of "circular funding" structures increasingly point to an industry that is becoming precariously balanced and increasingly vulnerable to sudden collapse.
There is really only one player in this whole game who looks as if they will come out relatively unscathed, if/when the AI bubble bursts.
That player is, of course, NVIDIA.
A huge swathe of the processors being used in the exploding number of AI datacentres around the world come from NVIDIA.
They have become the equivalent of the traders who sold pans, shovels and picks to the miners in the California gold-rush days. These traders got their money regardless of the fortunes of those who brought their wares. It was money for jam with zero risk.
So it is with NVIDIA. They're selling all the GPUs they can make and even if the whole AI industry collapses, they'll still have the cash in the bank even while bond-holders, lenders and investors start shopping for new shirts.
This week I investigated just how smart NVIDIA has been in respect to selling its wares and it became apparent that things are even better than I thought for Jensen and his gang.
For a start, when you buy an AI datacentre GPU the up-front cash is just the start of the revenue that NVIDIA gets from the sale. Apparently, to properly use that GPU, customers must also sign up to an annual contract that provides them with things such as the ability to sign software which operates on the trusted-environment component of those bits of silicon.
Yep, that's right, just owning the silicon doesn't let you run advanced, secure software -- that's an extra annual fee.
I guess even AI companies are being forced into the "you will own nothing and be happy" model of the new world order.
Now imagine an AI company with huge investments in AI datacentres goes belly-up.
Conventional wisdom would be that the administrators who are appointed to recover as much money as possible for creditors would sell off all those racks of NVIDIA GPUs to the highest bidder. Ah, if only it was that easy!
Yes, the administrators can sell the hardware and whoever buys that hardware will be able to use some of the basic functionality but if they want the advanced stuff that are really essential for AI datacentre use, they'll have to go cap-in-hand to NVIDIA to get access to the annual subscription which provides those services.
Now if you were NVIDIA, what would you do if/when someone who's just bought a bunch of second-hand AI GPU gear fronts up at your door?
The easiest thing to do would be to say "sorry, we don't supply this service to anyone other than the original purchaser. You will simply have to buy new GPUs if you want the advanced features".
The most profitable thing to do would be to say "that's fine but there is a one-time up-front fee also involved which is 75 percent the cost of a new GPU. Pay that and then we can also sign you up for the annual subscription needed to access the advanced functions". This option effectively enables NVIDIA to sell their GPUs more than once. They get to charge almost the price of a new GPU without having to do a damned thing and if the new purchaser wants to use those second-hand GPUs in their datacenter they'll have no option but to pay.
The very fact that option 2 could exist effectively means that the actual value of the GPUs that a company buys for their AI datacentres immediately falls to near-zero because the resale value would reflect the additional costs that NVIDIA could charge for them to be re-used.
Any investor or lender who thinks their money is secured against billions of dollars of NVIDIA hardware in a datacentre is going to be sorely disappointed. That security is virtually nill because of the potential for re-use fees and also because of the very short lifetime of these components in today's fast-moving technology. Today's premium AI GPU is tomorrow's legacy chip that is too inefficient and not nearly powerful enough to compete with the newer tech that has replaced it.
There's also another risk to investors in this cloud-based AI...
If just one of the big players crashes and burns, that could release a lot of AI GPU hardware onto the market at knock-down prices. It is almost certain that enterprising companies would buy-up these GPUs and use them to create local-AI servers which could then be flogged to businesses for probably less then they'd be already paying for just six months worth of token fees to one of the remaining cloud-based providers.
When used as part of a local LLM or AI agent, the advanced features enabled by the NVIDIA subscription wouldn't be nearly as important or necessary. This could see a huge surge in the uptake of local AI systems and that would further impact the cloud-based AI providers who are still in business.
The resulting decline in demand for cloud-based AI systems could mean that the failure of a single large player might produce a cascade of other failures in, what has often been referred to as "the bursting of the AI bubble".
Meanwhile, NVIDIA will be sitting in its counting house, counting all its money and at that stage, now that the AI marketplace is flooded with second-hand ex-datacentre AI GPUs, they'll go back to making gaming GPUs and perhaps their own ARM-based CPU systems.
Bookmark this column and let's revisit it in a couple of years.
Carpe Diem folks!
Please visit the sponsor! |
Here is a PERMANENT link to this column
Beware The Alternative Energy Scammers
The Great "Run Your Car On Water" Scam