by Travis Johnson, Stock Gumshoe | August 25, 2023 5:05 pm
This is premium content. To view this article (and to have full access to the rest of our articles), sign up. Already a member? Log in.
Source URL: https://www.stockgumshoe.com/2023/08/friday-file-nvidia-amazes-again/
Copyright ©2024 Stock Gumshoe unless otherwise noted.
A private company CEREBRAS has already developed a very large wafer chip which is faster than NVIDIA’S.
The company is building data senders in SAUDY ARABIA with these chips for billions of dollars. Waiting for its IPO
Will be cool to see what the business looks like once they lower the veil.
Excellent writeup!
Thanks, Travis!
Travis, I was looking at the Intel chart. It’s worse than the CSCO chart. On the surface, it tracked very similar (including a double top, which is what CSCO is doing now). After that Intel collapsed, prices till March 2023 were the same as (wait for it) AUGUST 1997. That’s in nominal terms. In inflation adjusted terms the story is far far worse. Total generational wealth destruction.
CSCO is in essence from a chart perspective where Intel was at May 2021. Does that mean it will collapse? I dont know/doesnt have to. But I think your point on buying NVDA at these valuations is a fair one.
Thanks for walking us through the numbers, the outlook and the mindset of all things NVIDIA. Since this company has grown to be my second largest stock position (not counting index funds), your writeup and rationale helped me with reasoning through next steps. I strongly believe in the future of NVIDIA but also know how it feels when the graph suddenly points down sharply as it has with many companies in recent years (Block, Paypal, Twillio, Etsy, Teladoc), just to name a few. They were all crushin’ it until they got crushed. Who knows what the future holds. Bottom line: Know your max. position % for a given allocation so you can sleep well at night and adjust accordingly. Wanting to buy and sell at the perfect time is a gamble that you can’t win consistently.
Travis, thanks again for the research and perspectives. I have a similar experience to you in seeing high flying, that is, high PE companies comes back to earth. I agree that NVDA will do that eventually, which is anybodies guess when eventually is now. MSFT is another company that had to take a seat until valuations equaled growth and new growth kicked in.
The stock that I think is more susceptible today to reversion to earth valuations is AMZN. For 25 years they have been given a valuation for incredible earnings growth for the next 10-15 years. AI possibilities have given them new wings as analyst recite their lead in AWS services being leveraged into AI services for tenants. I believe this will happen, but what degree? NVDA which is growing sequential quarters at incredible annual rates is valued lower than AMZN which is seeing comparatively very modest growth in their businesses.
So we end up with the same question? How does Mr. Market play the high multiple melody before it says take a seat for the next 10 years.
One other note on NVDA. Apparently their CUDA machine learning has been around since 2008. If there is a moat for NVDA it isn’t in the GPUs, its the software underneath the GPUs and the unwillingness of developers to switch to something else. It’s CUDA the DOS for the next 20 years? We’ll see.
One of the challenges with high growth stocks is accepting the craziness of the growth and the valuation even if you expected it. I’ve been committed to NVIDIA for years, during sometimes loopy valuations, in large part because they effectively own the operating system for AI, and because (almost) every AI project developed from 2010-2020, and every new student who studies AI, was working with NVIDIA’s hardware and software. That kind of moat can be extremely durable even if competition arises, because the incentive to learn a new system or try new hardware isn’t there for a lot of those users who are accustomed to what they learned on or what they have previously built on. Maybe that changes because prices for the chips are so nutty now, maybe that opens the door to AMD and others, or to self-developed chips at Alphabet and Apple and Amazon, but those moats usually evaporate slowly. And yet, even though that was what I envisioned being a potential long-term driver… it’s still hard to accept that when it goes really right and the share price goes bonkers as a result.
In some ways, I think of it similarly to Intuitive Surgical — everyone who learned robotic surgery learned on a da Vinci, because there wasn’t anything else, and every hospital has built da Vinci operating rooms… how much better does a new system have to be to make it worth swapping out that equipment and training new doctors? And how long will it take?
With Amazon, there is definitely an aura of faith-based thinking that has persisted for decades. If you build the most capable systems, the best infrastructure, and the strongest brand, it’s worthwhile even if that means you have to roll over all cash flow into investing in building those capabilities and growing that infrastructure and pleasing those customers, because you know, in the back of your mind, that one day Amazon could flip the switch and increase their margins by 10% and make phenomenal profits by taking their foot off the accelerator. That knowing is an article of investor faith in the system and the ethos that Bezos built, and it might fail… but I don’t think so, not with the irreplaceable assets built by Amazon over the past 25 years.
It’s true, though, that in terms of profit or cash flow, Amazon has pretty much never been reasonably valued. I think it’s more likely that they grow into that investor faith than that investors give up and push the price down to the point where it is valued like “regular” companies, but it’s certainly not guaranteed. I’m more comfortable with Amazon’s valuation than with Apple’s, personally.
Thanks for the follow and perspective. I own both NDVA and AMZN because of their strengths and no apparent change in Mr. Market’ perspective in the near future.
Two other companies, which I’ll call 2nd derivatives, are Arista Networks(ANET) and Cadence Design Soln (CDNS). They seem to have or are building enabling moats for 1st derivative companies like the hyper-scalers and chip makers.
ANET is a fast growing, pick and shovel company, that provides high speed routers switches and routers for cloud and databases. The risk is the Ciscos and other such companies catch up or pass them. But for the moment they seem to provide a better solution.
CDNS is one of those niche companies, that to me, is surprisingly needed by even the best companies. They provide the design software the enables design engineers to build the next gen chips, systems, etc. Since they are at ground zero they participate in a lot of the new developments whether AI, bitcoin mining, gaming, etc. Again the risk is customers decide to bring that in house.
Which brings this question to mind. When does a company decide the CapEx and manpower to bring something house is better than simply buying or renting? Does back integrating enhance or detract from the core business? NVDA relies on TSM to make the GPUs. C omaníes rely on CDNS to provide best in class design tools and they focus on designing.
it is interesting to see when and where companies provide synergistic solutions to seemingly totally self-reliant mega-tech.
Travis, ANET was pitched a lot in 2019. What is your take today?
Big beneficiary of capital spending boom in data centers — if they need to build out more AI capacity or build new data centers, they’ll need more Arista switches. Hugely reliant on a couple of the “titans,” as I recall, maybe Microsoft and Meta? That’s been good news lately, but was bad news last time those companies pulled back.
Haven’t looked at the valuation recently.
Travis, Thanks for the quick reply. Wasn’t aware they were reliant on a couple of major customers. Disclosure: a very minor position:.
Here are some metrics I follow and readers can make own judgment on valuations
5 yr Price AGR: 19%
5 yr Earnings AGR: 28%
5 Yr Revenue AGR: 22%
4 yr FCF AGR: -1%
FCF/EV: 0.85%
This is really a do you believe the growth will continue story.
If you have time, what is your take on CDNS?
Have never looked at CDNS in any detail, but it pretty consistently pops up as an interesting “picks and shovels” play for the fabless semiconductor world… particularly with big companies like Apple and Alphabet moving to design their own chips.
I have sold 50% of my holdings and have a limtit to sell another 25% at $600. I also have buys in at $275 and $185. I’m expecting a rollercoaster. $NVDA is projecting annualized growth of 23.5% from now through 2030 and that sounds really good except…they don’t make anything. $TSMC makes it all and $NVDA is 52% of their production. $TSMC is looking at 9% growth in the same period which assuming is all production and not product pricning would mean that if all that growth went to $NVDA then 9%/0.52= 1.3% which is < 23.5% and oh some of that growing manufacturing has been promised, contracturally, already to people like $APPL. Also to get these types of chips produced you can't switch over capacity from lesser technolgy manufacturing as it doesn't work. Also $ASML is projecting 12% growth from increased sales and pricing power and their machings are required by $TSMC to produce these chips. Something doesn't math.
Don’t think NVIDIA is anything like 52% of TSM’s production… more like 5%, I expect. Apple is TSM’s biggest customer, something like 20-25%. NVDA is looking into diversifying their supply a bit, perhaps with Samsung, but I don’t think they’ve done so yet.
Didn’t like leaving that at a guess, so I was digging around a little bit — most of TSM’s top customers are disclosed (other than Apple, which never lets suppliers disclose anything), so we know that the ten biggest over the last couple years have very likely been Apple (by far the largest, 20-25% of the business), MediaTek, AMD, Qualcomm, Broadcom, Nvidia, Sony, Marvell, STMicroelectronics, Analog Devices, and Intel. And they also make chips for major companies like Renesas and Amazon.
Most industry experts speculate that nobody other than Apple has been more than 10% of TSM’s business… but the top ten customers do usually account for about 75-80% of TSM revenue.
I don’t know their specific accounting policies, but my assumption would be that TSM’s charges to the chip designers make up most of the “cost of goods” in the income statement for the fabless companies, and most of the revenue for Taiwan Semiconductor itself (that might not be true, these are tight relationships and I expect the customers spend pretty heavily to cover their share of building TSM’s capacity, and probably have to buy their own raw materials too). COGS for NVIDIA over the past four quarters has come in at a little less than $12 billion, and TSM’s annual revenue has been in the $72-75 billion range. NVIDIA may well be their fastest-growing customer, and by that measure they might even be over 10% now (NVIDIA COGS are about 15% of TSM’s revenue), but over the past few years NVIDIA has probably, going by estimates of industry publications, generally been a top ten customer for TSM but not an overwhelmingly important one, at roughly 3-5% of the business.
Incidentally, AMD has probably been at a similar level — their COGS over the past four quarters is about the same as NVIDIA’s, and they have likewise been pretty fully dependent on TSM for their fabrication. The last estimate I saw was for 2021, when Digitimes estimated that AMD was a 4% customer for TSM, and NVIDIA 3% (Apple 26%).
Sorry rushing a comment in when I didn’t have time and not enough depth plus many typos, 17.3% not 1.3%. It isn’t all of $TSMC capacity but rather the capacity and growth for the AI chips. Present Nvidia A100 GPU is 7nm tech as is Googles TPU. Apple uses 5nm and calls them 4nm for A16 and the a17 is supposed to be 3nm. 7nm at this time can only be reliably done by Intel, Samsung and TSMC. 4nm and 3nm only be done reliably in large quantity by TSMC as Samsung yield appears to be 25% and Intel isn’t saying. Global Foundries threw in the towel and there don’t appear to be any real new threats unless you believe chinese propaganda. Next generation H100 is supposed to be 4nm and subsequent iterations 2nm but none of those plants are on line. Even ASML has stated the physics based limit is 2nm, with their EXE system, without a new technology not in production anywhere yet. Intels 5nm miraculously became a 3nm “equivalent” technology with marketing and layering tricks. Intel claims they have tech for 2nm and 1.8 nm but having tech and having production are two different things as Samsung has found out with the 4nm while claiming they also have a 2nm. It might be believable if they could reliably produce significant yields of the 4nm.
The $100B of fab expansion that TSMC has been concentrating on these small scales, 4nm and 3nm in AZ but in a smallish fab and having issues coming into production. There are plans for 2nm in Korea and Taiwan. I don’t remeber what the Japanese fab scale is but I believe at this time 4nm only. These small nm chips are a very small part of total unit sales but large part of revenue and profit for TSMC. The Nvidia profit is driven by these chips and their production limitations. The % I was provided for capacity were for these small nm scale chips only.
I hope I’m wrong but with what I’m being fed I don’t feel like I am.
Good insights, thanks — it’s true that NVIDIA is a big chunk of the “leading edge,” but they also say TSM is agreeing to boost production pretty steadily, so at least for now things seem to be holding. I don’t know how it will evolve, and you cold be right that the capacity isn’t there for NVIDIA and others to make all the chips they want.
It’s true that TSM’s capacity in Arizona is being pushed out a year (mostly because of the lack of workers), though companies in this business seem to get past the impossible, somehow. I do remember people saying that 10-12nm was the theoretical limit for chipmaking, but they kept getting smaller. Innovation tends to find a way through in this space, even if I don’t really understand the details of semi fabrication.
https://www.extremetech.com/computing/apple-reportedly-buying-all-of-tsmc-3nm-wafers-for-2023-thanks-to-intel more complexity.
Great discussion about NVDA! I, too am willing to ride the roller coaster, with about 10% of my portfolio. I have owned it since 1999, taken my initial investment off the table twice, and holding on for the rest of the ride (up, as well as probably down again along the way).
Robert Anderson
TSMC Candidly Explains Why It Can’t Keep Up With NVIDIA’s Red Hot AI Chip Demand | HotHardware https://hothardware.com/news/why-tsmc-cant-keep-up
Interesting, I didn’t realize it was the chipset packaging that was the choke point. Thanks!