Saul's Oct 2024 portfolio summary

Investment Analysis Clubs Saul’s Investing Discussions
1

Saul’s portfolio at the end of October 2024.

I’ve abdicated my role in leading the discussions but I’ll still be hanging around as a member of the board.

Here’s a table of the monthly year-to-date progress of my portfolio for 2024. I’ll present them as starting from 100% of my starting value and figure from there.

End of Dec 100.0% starting point

End of Jan 101.7%

End of Feb 125.4

End of Mar 127.2

End of Apr 117.4

End of May 121.7

End of Jun 121.7

End of Jul 105.7

End of Aug 110.7

End of Sep 110.6

End of Oct 116.2

As you can see, October was a good month, but on the last day of the month my portfolio dropped by 2% or it would have been even better.

I was down to five companies at the end of August but I took a small position in Monday (MNDY) in September which put me back at six positions.

Then, this month, I sold out of Transmedics, TMDX, for reasons I’ve expounded on at length on the board What's with Transmedics (TMDX)

I’ve also took a small position in IOT again so I’m still at six positions

So here’s what my postions looked like a month ago (end of Sept):

Axon 29.4%

Nu 25.8%

Nvidia 18.4%

Transmedics 11.5%

Sentinel 7.5%

Monday 7.4%

And here’s what they look like now, at the end of October:

Since August, I’ve been giving them as clean percentages of my invested positions that add up to 100%, excluding the cash part of my portfolio.

Nu 32.3%

Axon 25.9%

Monday 16.4%

Sentinel 9.2%

Samsara 8.4%

Nvidia 8.0%

I’m down to just six positions for now. But I’m an old guy and each month I take more and more money out of my investing portfolio and put it into our **permanently-**out-of-the-market pool . This means I’m constantly investing less and less of our total assets, so its dividing a much smaller pool into five, and if one position gets up to 27% of my portfolio, for instance, it’s a lot fewer dollars than 15% of my portfolio was a couple of years ago, because I’m investing less. Therefore I worry less, and am more comfortable.

What did I do this last month. Well… Here’s approximately what I did in October ! First of all I raised cash at an even faster rate than I had been in recent years. As far as individual stocks

NU – I didn’t buy or sell any this month. It rose about 14% in price and is now my biggest position by a substantial amount

AXON – I sold a bunch this last week at about $443. Why? It was becoming too large a position, over 30%. It closed the month at about $423

IOT – I bought a new position at an average price of about $49.00 and it closed at $47.80. When I had sold out of Transmedics and looked around at my old positions, this one stood out.

S – I added a tiny amount during the month and didn’t sell any. It rose in price about 7% during the month.

NVDA – I sold a large amount of NVDA mostly this last week. It was mostly at $140 and the rest about $134. It was enough to drop my position size from 18.4% to 8.0%. Why? Here’s why. They announced revenue up 122% yoy and up 15% sequentially. Well 15% sequentially compounds to 75% yoy. They announced adjusted EPS up 152% from a year ago, but up 11% sequentially. The trouble is that 11% sequentially compounds to 50% yoy. There’s risk to be some very disappointed people who don’t understand that. Sure there are wonderful new products coming out but these are physical things that have to be manufactured and they can’t keep rising at 122%, more than doubling each year. At 122% per year my calculation would be revenue in three years which would be 11 times what they have now. [2.22 x 2.22 x 2.22 = 10.94] It just ain’t going to happen. I guess I remember all too well what happened with Zoom when the year-over-year results stayed huge for a few quarters but sequential growth started dropping.

MNDY – I doubled the percentage size of my Monday position at about $27.30, early in the month, probably after selling Transmedics. It closed the month at roughly $30

TMDX - I sold out of it my TMDX position in three fairly equal portions during the month at $143, $136, and $130, and discussed my reasons at length on the board. It closed the month at $82. It would take a 66% rise to get back from $82 to my average price of $136. Selling when indicated is important.

As I said last time, I have a lot of confidence in NU, even though it’s a big position. Some people say “Oh, it’s a bank in Latin America. Not interested.” However I find it hard to imagine anyone actually seriously looking into Nu and not investing in it, but that’s just me.

I have kept a permanent safety fund out of the market that I could live off for several years if necessary, and I feel everyone who does not have a secure regular source of income should do the same. I have gradually added to it over the last several years, moving some funds gradually from my investing pool to my out-of-the-market pool. Given our advanced ages, my wife and I probably have enough to live for the rest of our lives with our out-of-the-market pool, with a little left over for our children. I add a little to our out-of-the-market pool almost every month.

I have learned long ago that sticking with great companies wins out in the end, and beats market timing, even though living through the 2021/2022 decline was very difficult.

FINISHING UP

Let me remind you first, that I have NO IDEA what our stocks will do next month. I’m terrible on predictions. I try to focus on the businesses of our companies.

When I take a regular position in a stock, it’s always with the idea of holding it indefinitely, or as long as circumstances seem appropriate, and never with a price goal or with the idea of trying to make a few points and selling. I do, of course, eventually exit. Sometimes it’s after months, and sometimes after years, but I’m talking about what my intention is when I buy.

I do sometimes take a tiny position in a company to put it on my radar and get me to learn more about it. I’m not trying to trade it and make money on it, I’m just trying to decide if I want to keep it long term. If I later do decide that it’s not what I want, I sell it without hesitation, and I really don’t care whether I gain a dollar or lose one. I just sell out to put the money somewhere better. If I decide to keep it, I add to my position and build it into a regular position.

You should never try to just follow what I’m doing without making up your own mind about a stock . First of all, you may have a completely different financial picture than I have. Different age, different income, different assets, different debts, different expenses, different financial and family responsibilities, etc.

Besides, in these monthly summaries I’m giving you a static picture of where I am currently, but I may change my mind about a position during the month. In fact, I not infrequently do, and I make changes in the position. I usually don’t announce these changes until the end of the month, and if I’m busy or have some personal emergency I might not announce them even then. And besides, I sometimes make mistakes, even big ones! Don’t just follow me blindly! I’m an old guy and won’t be around forever. The key is to learn how to do this for yourself.

THE KNOWLEDGEBASE

Since I began in 1989, my entire portfolio has grown enormously. If you are new to the board and want to find out how I did it, and how you can try to do it yourself, I’d suggest you read the Knowledgebase , which is a compilation of my “words of wisdom”, and definitely worth reading, (a couple of times), if you haven’t yet. It’s on the panel to your right.

I hope this has been helpful.

Saul

76 Likes
2

Hi Saul,

I have a question about your Nvidia decision. Jensen Huang has said that the new ‘Chip’ Blackwell is in full production and it’s sold out for the next twelve months. I know better than to argue about your numbers and logic. Although I agree, I think their will be some disappointed investors this quarter, given the probability of a significant decline in eventual YoY revenue growth, as effected by the sequential rev growth you’ve mentioned. The typical run up in share price to all time highs immediately after ER followed by a fade down 20-30% in the following month, to be then followed by a return to the all time highs before then next ER is bound to change. I’m not so sure of how this will specifically change.

Despite your very reasonable arguments regarding this ecentual decrease in YoY Rev growth, as a result of the sequential declines you’ve outlined, I believe most investors see the medium and long term, 1-3 years as Nvidia’s transition to more and more inference, as something not to be missed.

Do you have any rebuttal to what I have said here? Please, I’m seriously overweight and a little nervous.

Best

Jason

28 Likes
3

Jason, please remember that I am no tech expert at all, and don’t really know what a lot of the words even mean. I’m just doing what I 'm comfortable with based on the sequential growth, and how impossible a continuation of 125% growth each year is.
Best,
Saul

36 Likes
4

You’re right to be concerned about slowing growth, but it’s still going to be high. Here’s free TMF article:

If Nvidia hits management’s Q3 revenue goals, its revenue growth will fall to 80% year-over-year growth. That’s still impressive growth, but it is slower than what it has put up, which hurts Nvidia’s chance of posting a repeat.

For the 2026 fiscal year (ending January 2026), Wall Street analysts expect 43% year-over-year revenue growth, which is very impressive considering Nvidia’s size, but likely not enough for the stock to triple.

That said, and despite mostly making physical things, there is room for growth to continue at high levels. For instance, the new generation of AI chips, Blackwell, are priced higher than the previous generation, Hopper, for instance the price per chip of B200s are around 60%-70% higher than the H200s, which are still selling out. That means even for the same volume of chips, margins and profits will rise. The faster chips also increase demand for ancillary products, such as networking.

And Nvidia isn’t just a GPU company, they actually make more money on networking hardware alone than second-best competitor AMD makes on all of its AI hardware. Nvidia is also moving into software and “As A Service” offerings, but those aren’t large enough to move the needle, at least now.

I do need to get a better hanlde on TSMC’s capacity. For instance, is Blackwell additive to Hopper at their fabs, or is it replacement? The fab process for Blackwell (Custom-built TSMC 4NP) is somewhat different than for Hopper (TSMC 4N), but I don’t know if the existing fabs are easily modified for the new chip, or if TSM has to build mostly new fabs for the new chip.

And, for how much longer will Nvidia sell the old Hopper chip? It’s slower than it’s relative price discount, but at a lower price point there might still be demand for it given Blackwell volumes will be limited for a while.

Here’s an article comparing the two chips on a technical basis:

27 Likes
5

@Smorgasbord1 (or others) - What’s the risk NVDA’s innovation cycle outpaces its customers’ refresh cycle?

Given the cost of data center buildouts, is there a time where customers might keep working within existing compute rather than running toward the next generation? At some point, couldn’t the theoretical threat of falling behind technologically be less than the real threat of Cap Ex-ing yourself into oblivion?

16 Likes
6

My thinking is that with the new chips costing 60% to 70% more but performing 250% to 500% better/faster, people will want to build out with the new chips to save money on data center infrastructure (less racks for same compute). There’s also the power efficiency per compute factor as well.

What was surprising is that Blackwell hasn’t Osborned Hopper at all. That shows that companies are seeing value is getting AI compute today, rather than wait a year for something faster.

Some people question the ROI on AI infrastructure. Microsoft, I believe, is monetizing its build out even for its own use in the CoPilot line of services. Add to that Azure and then AWS for AI cloud services for companies that want some AI but don’t want to put out the Capex, or don’t have the expertise to build that out. Is Meta making money on its AI investment, and even if not, does the future for it for them not look bright?

I believe at this point any slow-down in AI Capex is at least a year away, as long as the economy holds up.

27 Likes
7

Pertinent to this, I have read/heard industry folks saying that it appears the gpus etc seem better thought of as OPEX, rather than capex. Their reasons for this are the very rapid improvement cadence of both AI (and hence its appetite for more powerful compute), and also the fact that the gpus wear out at a pretty solid clip. Part of that is the heat they generate (and will be presumably lower once liquid cooled racks are the norm). But a great many datacenters are already built, and cannot be easily converted to liquid cooling, so the DC operators are stuck with them. This may end up being a much more durable ‘build-out’ that past tech waves.

13 Likes
8

My 40th year in Chips and I’ve never heard of one “wearing out”. Some types of Flash Memory can wear out, fans absolutely wear out, but not a GPU that I’m aware of.

19 Likes
9

Hi MF

No doubt I’ve mischaracterized what happens with the chips (racks?, packages?, something else?) and I will look for the sources where I read that. It had to do with the higher levels of heat that the gpus generated, and the end result (according to what I read) was that they degrade in capability unexpectedly rapidly. They further stated (suggested?, implied? via the general context of the article?) that new generation data center designs that feature liquid cooling don’t have this problem. But I will be the first to admit I am a total poser when it comes to semiconductors, and that there must be far more nuance to it all Cheers

9 Likes
10

Hi MF

I looked around in my browsing history, did some quick searches on other sites where I might have read this, and came up with a blank. So, I headed over to Nvidia’s developer forum and found this:

It deals with a different model, but the topic is the same.

I have read in several places (sorry, I don’t have any links at hand!) that increased power consumption by and large directly correlates to 1) increased compute power and 2) greater heat generation (at least so far), and hence the strong push to incorporate liquid cooling into datacenters.

While I don’t understand all the details, and while it is no doubt incorrect to say the chips ‘wear out’, Nvidia seems to suggest that the overall assembly will be negatively and permanently impaired by the heat (and dust) that is part and parcel of a data center.

This was behind my earlier post to the effect that they might be better thought of as OPEX, rather than capex. Though of course that is a blurry line in any event.

I’d like to hear your thoughts, given your decades in the industry.

Cheers cm

5 Likes
11

On a recent BG2 podcast, CEO Huang talked about re-use of older Nvidia chips for inferencing. The clear implication is that these older chips are still working just fine and when replaced with Nvidia’s newer, faster chips, can be re-purposed for less time-critical tasks.

18 Likes
12

Higher voltage can get you faster clock speed, which means more computations per second. But higher voltage raises power by the square of the voltage change. The power dissipates in the form of heat.
However, too much heat damages the transistor junctions within the component. And dust blocks the escape of heat from the surface of the components.
So … why spend on water cooling? Same reason your internal combustion car engine is (usually) water-cooled. Water conducts heat better than air does. You run water (more likely antifreeze, because it won’t boil at 100C) through a system where the heat is, and push it through a “cooling tower” where air is blown past the water to pull heat from the water. In your car, the cooling tower is a radiator. In a nuke plant, it is a real tower of an oversize radiator. In a data center, it is in between. This costs more money, but the extra power/speed can get you higher computational throughput. It used to be that only gamers spent the money. Now data centers do too.

14 Likes
13

All chips have a rated maximum “junction” temperature and designs are done to NOT exceed this maximum. Those making boards and subsystems with Nvidia chips at this point are well-funded, highly skilled engineering teams (even SMCI) and they know not to exceed this temperature. TSMC guarantees in their models the same, as this is a key Reliability metric for many applications that I am involved in (think national security applications that rely on many years of operation).

There may be other components that fail in the system (or wear out), but I can definitely say that digital chips built on CMOS processes (every GPU) do NOT wear out or degrade in any way with age.

So, we can absolutely remove the case where customers will “wear out” their NVDA chips anytime within the window of their expected lifetime. They will be long retired due to performance of cell phone chips overtaking them long before then.

25 Likes
14

Thanks for the information. I think I garbled some factoids from the interwebs and wove it into a hopeful fantasy. Cheers

5 Likes
15

You would see this voltage game played in graphics cards, but you do NOT see this in Enterprise computing, which relies on extreme measures of engineering to assure that the chips are not over-heated or over-stressed as that would result in reliability being significantly reduced.

You will not see Dell, HP etc building product lines and playing games with increasing voltage. This market is very different than gaming. Games like that are NOT played as the big players will not stand for it. Extreme measures of testing are done prior to deployment.

11 Likes
16

Correct me if I’m wrong, but just to be clear, silicon based GPUs are not dissimilar from SiC chips used in inverter applications with respect to the bathtub curve. The specific parameters of that curve would be defined the “window of their expected lifetime.” I’ll take your word for it that cell phone technology will outpace the expected failure timeframe of GPU failure.

Separately, it seems to me that repurposing older chips is unlikely. First off, exactly where will the units housing the older chips be placed? As understand it, physical space in the datacenter drives power requirements, ventilation design and installation, even the cable length for connecting the boxes together to work in harmony. It just seems really unlikely that datacenter design would incorporate unused space for the future placement of repurposed machines. And second, even if there’s life remaining in the GPUs, I expect that a significant amount of there expected lifetime will have already passed and they will be that much closer to their point of failure.

If I’m right, and the machines with the older GPUs are not repurposed, at least not at the same site where they are being surplussed, what happens to them? Is there a secondary market for those machines? At some point they will need to be dispositioned, is there a salvage value or do you have to pay someone to take them off your hands? Who disposes of them? Where are they disposed of? Is there an investment opportunity in high tech waste products?

2 Likes
17

Ya’ll are sooooo far off topic here, and on a Saul thread? (shaking head) :stuck_out_tongue_closed_eyes:

23 Likes
18

I agree with dlbuffy. This thread has wandered way off topic and into minutiae. Let’s close that discussion. Thanks,
Saul

18 Likes