Doug Kass: My 'Deep Dive' Into Nvidia
This is how a generalist thinks about the 'most important' tech company extant.
You've reached your free article limit
You've read 0 of 1 free Pro articles.
I understand my investing weaknesses.
One of those deficiencies is in analyzing technology.
I could "lie" like other fugazzis who parade in the business media and seem to be an authority about everything — and, regardless, of body of knowledge, can answer any question lobbed at them! — when, in reality, they memorize sound bytes and have a knowledge base miles long but only inches deep.
So, take my following comments with a grain of salt — recognizing I know "sh-t from shinola." But it will give you a sense how, as a generalist, I think about Nvidia's prospects and share price.
I would love to see the following analysis of Nvidia NVDA ...
Turn Nvidia's quarterly data center revenue into AI chip/board unit sales (H100, etc).
How many data centers worth of parts is this? If it is a half a data center worth of parts, you feel good about Nvidia having a pretty long run.
But, if it is 100 data centers worth of parts, where is everything going?
The lead time on these data centers is years. They take a long time to build. It is really hard to even find power and water which they consume a ton of. These are new DCs that need to be architected very differently from existing capacity.
Then on top of Nvidia, include the parts Advanced Micro Devices AMD and Intel INTC are selling, in addition to the parts some DC owners have built/architected on their own (Tesla TSLA , Alphabet (GOOGL), Microsoft MSFT). How many total data centers per quarter have parts been procured for?
Then the other way of looking at this is if the parts sold are consistent with vendors closer to the end of the food chain whose stuff does not end up in inventory or triple ordered or being hoarded. Are server sales and networking sales and power sales (data center air conditioners, etc.) consistent with the chip sales in terms of the amount of final capacity they represent.
I am totally curious, this is neither a plus or minus point on Nvidia, I am just curious if the parts they are consistent with data center capacity that is actually being put in place right now? I am under the impression that people are having a really hard time adding incremental DC capacity. It is complicated, and it takes a long time to get these things up and running…
But I have ZERO idea what this analysis would show… not that I expect anyone to do it especially when the only thing that matters is figuring out their wafer starts and how the next quarter looks.
In thinking about my question above, someone sent me the below article. It does not directly answer the question, and frankly leaves me more confused.
I know lead times for Nvidia parts started coming in last quarter too. Normally with a semi, when lead times come in, you run for the hills. It could simply be that lead times were so stretched, that they have awhile to come in until supply and demand is balanced.
But as this happens, it is a somewhat artificial boost to sequential trajectory that provides a fair bit of the upside. Eventually it normalizes. Lead times were 8-11 months, now they are 3-4 months, and if they continue coming in at the same rate, they should be in balance by the end of this quarter.
Other aspects of the article are interesting including the fact that some are now sitting on excess parts and trying to offload them, as well as the underlying points about demand shifting to different use cases and delivery methodologies:
Buyers of Nvidia's highest-end H100 AI GPU are reportedly reselling them as supply issues ease.
At the same time, Coreweave just raised another $7.5 billion, so the money keeps flowing. The structure of these deals, however, continues to blow my mind. It feels like a combination of Enron and the CDO craze during the housing bubble in 2008. Same guys (e.g. Magnetar) are involved too, amongst others. At least mortgages were collateralized by homes, which appreciate in value as opposed to depreciate, and you also know exactly what their use case is and that there is real end demand for them.
This thread is interesting:
Then something that didn’t seem to get much attention, the FTC seems to be poking around into how all this is structured from a competitive perspective. But the tweet following the FTC press release I think is wrong. The FTC could care less about financial reporting. That is for other government agencies that also are poor at doing their jobs.
Bottom Line
At the end of the day and on a purely short term basis, Nvidia's shares got cheaper on a multiple basis after earnings and guidance. The question longer term remains: Is the use case there or not, and to what degree?
That is another thing I do not have the answer to and I am on the sidelines on.
At the time of publication, Doug Kass was short TSLA (S).
This article was previously published in Doug's Daily Diary on TheStreet Pro.
