Nvidia Bubble: Sell The Rip

Summary

• As Nvidia has skyrocketed towards $1T market cap, investors appear to have started ignoring the fundamentals. As I see no further upside, I am repeating my 2021 Peak Nvidia call.

• Nvidia's growth sustainability, moat, and valuation are exaggerated, making the stock overvalued and potentially a bubble.

• Comparing Nvidia to Zoom's temporary pandemic surge, the current demand for AI hardware (and therefore Nvidia's growth) may not be sustainable in the long term.

• With competition from Intel and other chip vendors being unjustifiably minimized, Nvidia's 98% production gross margin seems unsustainable. Nvidia has no clear AI leadership/moat.

• The valuation is actually not defensible, as a surge to something on the order of $100B revenue is now being priced in. Sell your shares at the current peak valuation.

Justin Sullivan

Investment Thesis

Nvidia $NVIDIA Corp(NVDA)$ has skyrocketed towards $1T market cap after upgrading its guidance. Even following the most optimistic estimates, this means, at the current stock price, Nvidia now trades for an absurd >10x P/S for at least the next half a decade. Simply put, even if those optimistic growth estimates materialize, I see no realistic further upside. Sell your stocks at the top of the bubble (Peak Nvidia).

Analysis

Growth sustainability

I’ve seen some analysts (at least one) call this the most impressive quarter they’ve ever seen. However, this current unfolding of events reminds me very much of a stock which arguably an even more impressive quarter or series of quarters, but eventually turned out to be a bubble (just the same): Zoom $Zoom(ZM)$. As everyone turned to Zoom during the lockdowns, the company saw years of demand being pulled in to just one or a few quarters. Eventually, though, it turned out the company simply had basically run out of new customers, leading to a collapse in growth, and with that investors flocked away and the stock crumbled to prices not seen even since IPO.

Similarly, one could argue that with the current boom of chatbots and AI, since as has been discussed widely ChatGPT has been the fastest growing app ever, companies suddenly realized they need a ton more AI hardware. This is arguably what’s happening.

While, sure, Nvidia has indicated that it does not see a drop-off in demand for (at least) several quarters, and obviously with the likes of Google $Alphabet(GOOG)$ and Microsoft $Microsoft(MSFT)$ broadly rolling out AI products, necessitating more AI hardware to run those services, the comparison to Zoom nevertheless suggests a scenario where the majority of the surplus in demand is already being consumed at these early stages.

Admittedly, there are differences between the pandemic and AI, as AI is expected to continue to increase in importance over time, perhaps even more than people may think (see Tesla: Elon Musk Is Right -Optimus Robot). While indeed the TAM for AI and data center silicon in general is expected to continue to increase over time, which could indeed lead to continued growth over time, the main issue are the parabolic estimates, which seem primarily based on extrapolation of current growth rather than any fundamental insight. That which can be asserted without evidence, can be dismissed without evidence just the same.

Ultimately, though, taking a cautious view means that growth would flatline sooner than what seems to be acknowledged. As case in point, according to the tweet above based on consensus estimates (and which is confirmed by Seeking Alpha data), Nvidia is suddenly seen generating around $100B revenue by 2028. Investors should take such estimates with a grain of salt, and weigh the uncertainty of the extent of continued market growth with other factors such as valuation and competition (see further).

To be fair, many of these recent multi-year revenue revisions are bringing the consensus back to quite similar levels as in early 2022, before the recent downturn. Nevertheless, this just proves the point: both bullish and bearish sentiment gets exaggerated, and mostly the truth ends up somewhere in between.

Competition

Similarly, the extent of Nvidia’s so-called moat has been incredibly exaggerated. To prove this, let’s consider Nvidia’s economics.

According to a respected AI analyst of the firm Cambrian-AI, Nvidia’s latest H100 sells for over $20k. For comparison, a quick Google search reveals that a TSMC $Taiwan Semiconductor Manufacturing(TSM)$ N5 wafer costs $17k. What this essentially means is that Nvidia’s business model is as follows.

Nvidia buys a processed wafer from TSMC for $17k. Depending on the yield (let’s say between 50-100%), Nvidia then dices this wafer into something like 40-80 chips. Then, it sells each of those 40-80 chips for a price meaningfully higher than the price it paid for the whole wafer. That’s a 98% gross margin!

Obviously, such monopolistic prices aren’t sustainable as there are plenty of companies capable of chip design, including both Big Tech and competitive chip vendors such as AMD $Advanced Micro Devices(AMD)$ and Intel $Intel(INTC)$.

At that point, the market dynamics are well understood: competition will lead to both market share as well as pricing power erosion of the incumbent.

Indeed, while people like to tout Nvidia as the leader in AI, I see Intel having already surpassed Nvidia in TCO with its Habana Gaudi chip, then surpassed Nvidia A100 in performance (as well as TCO) with the Gaudi2 chip, and recently surpassed Nvidia yet again with Ponte Vecchio vs. H100, with Gaudi3 also on the horizon which Habana suggested would readily surpass H100.

Looking a bit further, since Falcon Shores in 2025 is expected to be produced on “angstrom-era” process technology (presumably 18A), which is when Intel expected to regain process leadership, it should be able to further extend its leadership, as Nvidia is expected to adopt N3, which will be an obsolete node at that point.

Overall, even if AI demand will skyrocket to a $100B market over the next few years, it certainly isn’t so sure that all of this will be captured by Nvidia, never mind at 98% kind of gross margins.

Valuation

Valuation is where things get tricky, as the future as discussed is highly uncertain. One of the primary issues is that investors tend to use the metrics they prefer, explaining away any worries about inflated valuation.

As case in point, one argument in favor of Nvidia’s $1T valuation is that the PEG value is actually very low. The issue here is that Nvidia is coming up against some very weak comps ahead, which naturally inflates the growth rate for one. In addition, as growth more or less flatlines under the most realistic (although not as acknowledged) scenario, this simply means that this PEG valuation metric will skyrocket in a year or so as the earnings growth component from this metric collapses. So no, Nvidia is not cheap.

Secondly, the main argument is that due to the growth in earnings estimates, this actually offsets or even more than offsets the increase in stock price, and therefore the (more traditional) P/E valuation at worst has remained the same.

I don't share this mindset. First, given the semiconductor downturn over the last year, many companies have seen a decline in earnings and therefore a bump in their P/E. Since investors have been more or less banking on this downturn remaining fairly short in duration, a higher P/E ratio isn’t necessarily a sign of a valuation multiple expansion, but rather a somewhat educated guess about the more normalized earnings power of these companies. For example, Intel saw its P/E increase to something like 50x before transitioning to not even being profitable anymore.

Secondly, while growth companies may enjoy higher valuations, there still has to be some realism. On one hand, the market certainly isn’t infinite, which means that a scenario of nearly infinite growth (>$100B revenue) should not be assigned the same probability as for example growing from $1B to $2B revenue, which therefore should be reflected in the valuation multiple. In other words, 40x or 50x multiples can definitely be considered too high even if a company is currently reporting or guiding to very high growth rates – again referring to the Zoom case study.

And secondly, even neglecting any historical precedents, these kind of >20x P/S and >40x P/E ratios are, simply put, very high. Since these metrics are already forward-looking valuations, this means that after the current spike in demand over the next year, Nvidia would still have to double its revenue from there for the valuation to make sense.

As argued, while that may actually be in line with where the consensus estimates have been revised to for the next half a decade, it is risky to bank on such sustained high growth, and even then, this means the most optimistic scenario is simply Nvidia growing into its current $1T valuation. In the worst case, Nvidia never will.

Simply put, I see no upside, and therefore no investment case to be made. Hence, I am repeating my 2021 thesis of Peak Nvidia: Nvidia Stock: Investors May Want To Avoid Or Sell. If one has any shares in Nvidia, one should take profits at its peak valuation.

Peak Nvidia

At the time of my previous Peak Nvidia call, Nvidia was trading around $200, half of its current value.

As discussed, unless my call that revenue may flatline sooner rather than later is wrong, then $200/share indeed seems a reasonable valuation, as in my view revenue would have to double again for the current valuation to make sense. This doubling of revenue could easily take another half a decade, a decade or more, or perhaps it may never happen.

Admittedly, though, at that time I likely would not have expected that within two years from that time Nvidia would already be generating this high kind of revenue, approaching Intel in revenue given the latter's downturn. I also would have expected some fiercer competition from Intel, but the company has had some delays with its first-gen Ponte Vecchio product. On the Habana product side, the product as discussed is a viable, cost-competitive alternative for Nvidia products.

Nevertheless, the risk of competition as discussed remains very real. For example, if Nvidia would lose a similar amount of market as Intel to AMD on the CPU side, then billions of dollars of revenue are at stake. This scenario certainly isn't unreasonable given Nvidia's incredible gross margin on one hand, and the trends with regards to process leadership on the other hand: TSMC Stock: Losing Leadership. This same kind of position is indeed what led (or at least contributed) to Intel's market share erosion.

Investor Takeaway

I am repeating my 2021 thesis of Peak Nvidia for three reasons. First, the sustainability of the revenue growth is arguably strongly exaggerated, which I argued by comparing to the Zoom case study. Secondly, Nvidia’s moat is also strongly overstated, and the amount of competition is being unjustifiably neglected. Thirdly, the valuation as well cannot be justified.

While my previous Peak Nvidia call was at $200/share, or around half its current share price, as discussed Nvidia would have to grow its revenue to something like $100B for the valuation to somewhat make sense, never mind for the company to become investible. As there is no evidence that those lofty estimates will really materialize (that first argument about the growth sustainability), I would argue that a $200 price would indeed be a much more reasonable price target for the stock.

As the stock is therefore overvalued at best and a bubble at worst, investors should consider selling their shares.

Source: Seeking Alpha