This seems like more than idle confirmation, to along with an AI has legs thread here…
Did you read the whole of it? $8Trillion annually! Do I need to pay McKinsey in order to find out when they predict this level of economic impact? Let’s assume 5 years from now, does that sound realistic? Can that be estimated? Does a number that large need any qualifiers, really.
I’d like to know your thoughts.
Hi brittlerock,
"@intjudo said he/she was in PSTG as he/she felt that it will benefit from the AI boon. All that data has to rest somewhere. "
*) he 
*) “All the data has to rest somewhere”, yes, and specifically/especially for AI it needs to be low-latency data. Doesn’t hurt if the low-latency data costs the same as spinning-disk and $PSTG checks both boxes now.
EDIT: I’d emphasize that if low-latency disk is putting a bottleneck into your AI system, you can either buy more GPU modules and rack 'em up, OR you can buy zero more GPU modules, and simply replace your spinning disk with $PSTG lower-latency storage. An in an environment where $NVDA GPUs are hard to come by, and expensive, folks will do whatever they can to bleed everything they can out of each GPU they do have.
Another factor is networking; AI loads and datacenter in general need low-latency, programmable networking capabilities; for that reason I also have a small position in $ANET and need to learn more about it.
For you history buffs, $ANET’s programmable networking devices were a new-ish thing at the time and they solved a critical/fundamental technical problem $NET had back when it was just starting out.
AMD may do $2B in AI chip revenue next year. Nvidia did $2.5B in AI networking alone last quarter.
I was invested in Arista quite some time ago. I got out with a nice profit when their revenues began to slip. I’ve considered getting back in, but haven’t really done any research. Nvidia also provides some connectivity products. Not sure if these are competitive with Arista or complimentary or just a separate and mostly unrelated offering.
Arista’s management came from Cisco. As I remember the story, Cisco was reluctant to provide R&D $ for the development of programmable switches as they felt it would cannibalize their existing products. Or something like that. Anyway, the founders of Arista left Cisco in order to start their own company that offered competing products. If I remember correctly, Cisco sued them and lost in court.
That was all before SaaS products became a thing. Maybe I will take another look at ANET.
If you all want to read a great article about the AI Value Chain, Ivana Spear is really great.
To be sure, @Smorgasbord1, if you’re purely a momentum investor, AMD isn’t where things are at right now. Then again, neither was NVDA when I invested in Fall 2018, but that’s done alright. I’m certainly not suggesting the same growth for AMD, but I do think they’ll get a piece of the growth that NVDA can’t capture – they’re already selling on allocation – and that will lead to more growth in GPU sales for them.
I love ANET but I’m not sure that AI will be a big tailwind for them. The primary use case for networking and AI, from where I sit, is networking within a data center between your racks of GPUs and other subsystems. For that purpose, Infiniband, where NVDA has a strong play, is a much better option. My $0.02.
Here’s the McKinsey link:
My notes: While Mr. Market is worried that AI is some kind of crypto-mining-like flash-in-the-pan, businesses are looking to adopt AI as part of their long-term efficiency plans.
For instance, McKinsey “estimates that half of today’s work activities could be automated between 2030 and 2060, with a midpoint in 2045, or roughly a decade earlier than in our previous estimates.” Add to that “Current generative AI and other technologies have the potential to automate work activities that absorb 60 to 70 percent of employees’ time today.” and you can see why this huge white collar labor reduction is a big deal.
McKinsey looks at the potential impacts two different ways, coming to basically the same huge impact conclusion. They also provide 4 detailed business examples. It’s worth a skim at least.
If we’re talking about this as a reason NVDA isn’t running higher, I think it’s more that the market is starting to wonder how long they’ll continue to be #1 by a mile. Sure it will take Google or Microsoft or whoever a while to catch up, but Nvidia probably won’t get to be the only (relevant) game in town for much longer.
Bear
As I posted on the paid board, AMD came out with an announcement earlier this month on a new AI chip shipping this quarter, and made all sorts of claims for it. Yesterday Nvidia countered those claims:
And now today Intel is claiming their AI chip is great:
This sounds a lot like the “Tesla competition is coming” line we read in the press for years that still hasn’t really happened. The market for AI hardware isn’t just “pick the fastest chip” (though Nvidia still has that and is planing on shipping an even faster chip in 6+ months), it’s all the hardware and software infrastructure around it. Nvidia’s InfiniBand networking for getting data in and out, boards with both low power draw ARM-based CPUs coupled with the GPUs (the “Grace Hopper” series), and, of course, the CUDA software that is the defacto standard and which remains a huge moat.
Add in that Nvidia is getting into AI cloud services in a big way (actually, a few different ways including with partners), see that its forward PE is less than AMD tells me the market just can’t believe how bright the future is for Nvidia.
Nvidia remains production and US export rule constrained. It’ll be interesting to see what Nvidia comes up with for China - we’ve already seen announcements on new fabs in Vietnam, expanding production outside of Taiwan, reducing political risk there too.
Looks like the NVDA deal in Vietnam isn’t for a production fab, it’s for a “semiconductor design hub” – those are chip designers, not factory lines cranking out wafers.
NVDA doesn’t have its own fabs. They outsource to TSMC, primarily. The article quotes an NVDA exec saying they’d be very interested in a 3rd production partner after TSMC and Samsung (which today gets a small chunk of NVDA’s production runs).
Could that third production partner maybe be Intel? Stay tuned!
Meantime, the biggest risk to NVDA that I see in the next several years is the risk that China invades Taiwan. NVDA would be dead in the water.
How likely is that? I don’t know how to quantify that risk, but I can say confidently that the risk goes up significantly if the Ukrainians run out of ammo and missiles because the US can’t figure out how to fund their defense against the Russians. Sorry, I’m probably supposed to keep politics out of this, but to me, this is a clear outcome of the current Washington stalemate on Ukraine continuing.
ActonUp
(and living up to my name…)
Thanks for the correction!
My reason above (eventual competition in GPUs) is part of why the market is reluctant to plug in 30% or 50% growth for the next several years and price NVDA accordingly. But your TSMC concern, @ActonUp, is probably an even better reason the market isn’t more enthusiastic – that is a giant risk if I understand it correctly. Just as bad as if a company only had a single customer or something. For a $1T+ company like NVDA, it’s just almost unfathomable that they wouldn’t have some kind of back up plan. Any thoughts @Smorgasbord1?
Bear
There are a few ways to look at this risk:
• It’s not just Nvidia, but even Apple and other big companies rely on TSMC. According to IBD:
A Chinese blockade of Taiwan — even one that doesn’t spiral into a military conflict — could disrupt over $2 trillion in economic activity by severing the primary source of advanced semiconductors from global supply chains, Rhodium Group estimates. Taiwan Semiconductor’s output — including 70% of all smartphone chipsets, 30% of auto microcontrollers and a dominant share of chips for high-end graphics processing units for PCs and servers — feeds through to $1.6 trillion in annual corporate revenue for its global customers, the research firm says.
What happens to Samsung, Google as well as Apple for smartphones? What happens to everyone dependent on a Smartphone ecosystem?
• Even Chinese companies rely on chips produced in Taiwan. Again according to IBD:
China’s own economy would suffer mightily. Beijing’s Made In China 2025 strategic plan set a decade long goal for domestic output to supply 70% of the chips used by its tech, auto and electronics industries. But Chinese firms only supplied about 7% of its $187 billion chip market in 2021, IC Insights says. Even including domestic semiconductor plants operated by foreign firms such as TSMC, Intel and Samsung, China sourced only 17% of its chip needs from within.
Even in the unlikely event that Beijing quickly prevailed and seized TSMC intact, there would be no escaping massive economic fallout for China and the world. The U.S. would presumably bar companies like Applied Materials, KLA Corp. and Lam Research from providing support and components for their equipment. In short order, the world’s most exacting manufacturing processes would grind to a halt.
TSMC does have a facility in Arizona, but it’s not nearly ready to take over all Taiwanese output. I think the Vietnam move is part of Nvidia building up a backup plan. I suggest reading the IBD article I linked above for a detailed discussion. I’m as far from a geo-political expert as one can get (I did visit Taiwan in the 1980s though), so my advice here is worth nothing. I do think if China invades Taiwan all stock in all the world’s markets will be impacted, most heavily.
I’m going to reiterate that last point. If China invades Taiwan, the source of a huge amount of chips by percentage, the entire tech sector collapses. Period. Software needs hardware and hardware is chips. Game over.
China’s economy which assembles these tech gadgets? Over.
In AI, NVDA will eventually see competitive threats that are meaningful. I’m extremely confident that “eventually” is considerably longer than the street is assuming.
NVDA is my highest conviction.
That’s right, @MFChips and @Smorgasbord1 . If China invades Taiwan, there’s a lot more disruption than just to NVDA. The entire tech sector locks up, and the world probably drops immediately into a sustained recession.
Despite the headlines over the past many years about China’s increasing threat to Taiwan, this hasn’t been a particularly central risk to the West simply because there was tacit understanding across the board that it was in our collective self-interest to stand alongside Taiwan.
That tacit agreement seems to be in question right now, which is why the risk is dialed up. I still expect that somehow we’ll pull ourselves back from the brink of yet another “own goal” that we seem to be prone to these days.
NVDA and AAPL are two of my highest conviction positions, and I will certainly pull back on those and others if I see us going over the cliff.
As to Bear’s question about how NVDA could end up with this kind of dependency, it’s a complex question, but my “too simplistic but directionally right” answer is that re-creating somewhere else the highly developed ecosystem supporting chip manufacture that Taiwan has built up since the 1960s will take decades of sustained government effort, coupled with enormous private sector investment.
Last I checked, we don’t live in a world where that can happen.
The Biden administration efforts along those lines are a small start against a very large problem. They probably won’t get to do more on that before 1/20/2025; and whether this will be a priority to the next administration is anyone’s guess.
ActonUp
This board disallows politics. Geopolitical discussion is still politics. Besides, there is no meaningful geopolitical threat facing Nvidia as an outcome of the way they source their products.
To your last sentence, really?
The question Bear brought here is not off limits. I think you are stretching the rule here in this case.
I think we can quickly turn this completely non-political. There are many more risks related to being dependent on a single supplier than just the geopolitical risks. Maybe they could decide they don’t want to sell to NVDA for some reason (maybe not likely). Could they double prices, though? The supplier could itself have some sort of supply issue. Fraud or regulations could hit the supplier. Maybe the supplier could have an earthquake damage all their factories.
Ok maybe none of these will apply to TSMC, but a single point of failure is always a risk.
Bear
Bear, 100% correct. TSMC has never raised their prices to take advantage of their market position. I owned them for some time and that was my primary thesis and they won’t do it.
And I’ve never understood how they industry allowed for fab after fab to be built in Taiwan with plenty of threats such as hurricanes and other natural disasters. There are multiple factories on different power grids etc. But nearly all high performance chips are built within close proximity
Broadcom, Qualcomm, AMD, Xilinx, Apple, Tesla, all computing, networking and the data centers are all deeply affected if there’s a hiccup. And this has been true for years. Getting a second source in place is pretty much not done for the high performance products.
I would say the risk is the same to all SaaS companies with a few minor exceptions such as IoT, where the growth would be hit hard but not likely eliminated as their hw is not performance driven. Datadog, Snowflake, Net, etc all run on HW and anything that does availability of new HP hardware will hit software growth hard.
