The Artificial Investor - Issue 27: Are we in a Tech bubble? What Nvidia’s stock price tells us.
My name is Aris Xenofontos and I am an investor at Seaya Ventures. This is the weekly version of the Artificial Investor that covers the top AI developments of the last seven days.
The Most Interesting Story of Last Week: Nvidia stock’s volatility and what it means for AI
The most interesting story of the last few days was Nvidia’s valuation decline by 14% within 5 days, only a few weeks after it became the third company worldwide to reach 3 trillion dollars of market valuation.
Is the AI darling, Nvidia, overvalued? What does it mean about the broader AI sector? Are we in a Tech bubble driven by the latest AI wave? Let’s build it up by looking at the bull and the bear cases individually.
If you prefer to listen to this issue, click here for the audio version.
Powered by ElevenLabs
🐂 The bull case: Nvidia is undervalued
📈 What do valuation maths tell us?
Before we look into the fundamentals of the business, let’s analyse briefly the maths behind the company’s valuation in the way a market analyst would.
Nvidia is currently valued at 50x Price to Earnings ratio (P/E ratio), which means that investors are willing to pay 50 dollars for every dollar of net income of the business. This is in line with the current valuation of its sector, the US Semiconductor industry. This figure is also in line with the median of Nvidia’s Price to Earnings ratio in the last 13 years.
A 50x Price to Earnings ratio means that an investor in Nvidia is willing to wait for 50 years before “getting their money back” from their investment, i.e. they are willing to accept a 2% yield on the earnings of the company (In practice, they are not exactly getting their money back from the earnings, because this would imply that the company distributes 100% of its net income in dividends. Nevertheless, they are getting value, which can be translated eventually into money).
An equity investor in risky growth stocks typically aims for a 10%-15% return on their investment. This investor can get returns in two ways, earnings (yield) and growth (price increase) (see the Gordon Growth model). As a result, this would mean that this investor would need a long-term growth of 8%-13% in Nvidia’s earnings in order to achieve their target returns (i.e. the expected return minus the earnings yield of 2%). The company grew revenues 112% in the last twelve months and is expected to grow by about 80% on an annual basis in the next 24 months. Therefore, it looks like the company’s current explosive growth can justify such a valuation.
Now, let’s also compare Nvidia to the leading US Tech stock index, the Nasdaq-100. Nasdaq-100 stocks grow on average by 13% and the median Price to Earnings ratio is 26x. Nvidia’s nearly double Price to Earnings Ratio (i.e. 100% premium) seems justified by the fact that its growth rate is more than 6 times the one of the Nasdaq-100.
Nevertheless, how reliable are Nvidia’s growth expectations? Is there demand for its products? This seems crucial when assessing the company’s valuation.
💰 The Magnificent 7 breaking the bank
Nvidia's revenue for 2023 was about 61 billion dollars, a 126% increase from 2022. Analysts forecast Nvidia's full-year revenue to reach 125 billion in 2024, i.e. a 125% increase vs. 2023, and 178 billion in 2025, i.e. a 42% increase vs. 2024. (Nvidia’s fiscal year starts in February instead of January, so there is a misalignment by a month in the figures we reference, but we ignore this for now. Also, in case you look into the sources and get confused, the company’s 2025 fiscal year refers to the period of February 2024-January 2025, so it can appear as if it is a year off).
Nvidia’s revenues are largely not recurring, so it needs to find 300 billion dollars in the next 18 months or so, in order to meet market expectations. Where will this money come from? Is there such a high demand for its products?
Well, it looks like the answer is yes. Nvidia's main growth engine is its datacentre infrastructure business, which accounts for more than 85% of revenues. The main clients are the hyperscalers or the “Magnificent 7” as they are called by analysts: Amazon, Facebook (aka Meta), Google (aka Alphabet), Microsoft, Apple, Tesla. We will also add X.AI into the mix, Elon Musk’s recently-launched AI business. All companies have announced they will continue to invest heavily in AI infrastructure with annual capital expenditure amounts ranging from 10 billion dollars annually for Apple and Tesla to 50 billion dollars for Meta, Google and X.AI. The aggregate figure reaches 450 billion dollars for the next 18 months. One one hand this is not only for AI and only for Nvidia, but on the other hand Nvidia has more clients. So, this should be more than enough to cover Nvidia’s revenue projections.
The fact that hyperscalers are investing in AI infrastructure doesn’t mean it will all go to Nvidia. How strong is Nvidia’s market position?
⚔️ Nearly a monopoly
Nvidia is the clear leader in the AI chips market, as its market share is 65%, three times the one of the next player, Intel, which indicates a monopoly. This is reflected in the superiority of Nvidia’s gross margins, which show that the company has the luxury to demand higher prices for its products. Nvidia’s gross margin was 75% as of July 2024, which compares to the 45% average gross margin for the Semiconductor industry.
Nevertheless, a current market leadership doesn’t necessarily mean that it is sustainable. How strong are the entry barriers in Nvidia’s sector?
There are a number of entry barriers stated by most market analysts, such as i) high capital requirements to operate in the sector, ii) economies of scale that lead to better margins, iii) strategic partnerships along the value chain that take a long time to replicate, iv) patents, and v) brand. However, our favourite entry barrier is the one mentioned the least: Nvidia’s software and developer ecosystem.
Nvidia has developed a platform for parallel computing (called CUDA), which allows developers to use Nvidia chips for high-performance computing tasks. On top of that, Nvidia provides a comprehensive suite of developer tools and libraries (e.g. Nsight and CUDA-X) that enhance performance of applications running on Nvidia hardware and are widely used in AI and data science. This creates a strong software ecosystem that is difficult for new entrants to replicate.
In addition, the widespread adoption of Nvidia's chips and machine learning applications has led to a large community of developers familiar with its tools and libraries, which results in a network effect that makes it challenging for competitors to attract developers to alternative platforms.
🐻The bear case: Nvidia is overvalued
♟️ Strategy is the art of making choices
When asked, most of the hyperscaler CEOs, such as Zuckerberg and Sundar, stated that they are making the AI infrastructure investments simply because “it is too strategic for them” or “they cannot afford not to” or “there is more to gain than to lose”. They don’t expect any ROI in the near term.
On one hand, who can blame them? If you have a cash cow sitting on a high pile of cash and you are not able to do acquisitions, then how do you add value to the shareholders? There are three alternatives: i) you buy back shares, ii) you issue dividends or iii) you invest in new technologies. And they are doing all three (plus, they have invented an alternative M&A structure to bypass regulatory scrutiny - see Google/Character AI, Microsoft/Inflection, etc.).
On the other hand, how long will this last for?
The hyperscaler’s capital expenditure plans for the following two years are sufficient to cover Nvidia’s projected revenue growth for the same period. However, if the revenues don’t come, they will have to redirect their capital somewhere else. Sequoia’s David Cahn talked about a 600 billion dollar gap in AI, caused by the fact that there are significant investments made in infrastructure that still haven’t been matched by revenues generated by the software companies that use this infrastructure.
Many talk about AI infrastructure investments as being the “demand side”, when looking at the world from Nvidia’s perspective. However, infrastructure spend is actually the “supply side”when looking at the world from the end client perspective and the software businesses serving them.
🌍 The AI Cold War’s impact
One should not forget that we are in the middle of a Cold War between The West, including the US, Canada, Europe and the UK, and The East, including China and the Middle East (with Taiwan caught somewhere in between), where each side is “attacking” the other with export bans, domestic subsidies, etc. This naturally affects Nvidia’s growth prospects in China and Taiwan, its second and third largest markets globally that are responsible for 39% of the company’s revenues (vs. 44% in the US), and in the Middle East, where it recently had its first large-scale deployment.
Also China is a manufacturing leader in various hardware components that are critical for the AI supply chain, such as circuit boards. Disruptions to this supply chain can cause delays in Nvidia’s revenues and increase in production costs that impact profitability.
🔪 Death by a thousands cuts
While it is not easy to think of a single challenger semiconductor player capable of threatening Nvidia in the near term, there is indeed increasing competitive tension.
Hyperscalers are responsible for the vast majority of the AI server revenues and literally all of them are working on their own AI hardware: Microsoft and Google have partnered with Arm, while Meta partnered with TMSC and Samsung, years after Apple partnered (also) with TSMC and Amazon acquired Annapurna Labs for the same reason. Even OpenAI is doing it with TSMC.
At the same time, one cannot ignore the emerging startups that challenge Nvidia’s dominance and have collectively raised nearly 3 billion dollars of funding:
Groq has raised over a billion dollars and specialises in AI inference chips optimised for speed of pre-trained models
Cerebras Systems has raised 720 million dollars for its Wafer-Scale Engine (WSE) chips designed to accelerate inference on deep learning models
Skild AI has raised 300 million dollars for its general-purpose AI models specialised in robotic use cases
Celestial AI has raised 370 million dollars for its Photonic Fabric technology that addressing AI latency and bandwidth issues
SiMa.ai has raised 270 million dollars for its high-performance, low-power AI chips for edge computing applications, such as IoT
Etched has raised 120 million dollars for its specialised processors that are optimised for transformer models, which is the architecture of all leading LLMs
What could slow investment ROI, deglobalisation and competitive tensions mean about Nvidia’s revenue and gross margin expectations in the long run?
🤥 Numbers don’t lie
Let’s go back to the valuation maths, in the same way as we started.
We said that Nvidia investors would need to see a 8%-13% long-term annual earnings growth for their investment to be worthwhile.
Let’s assume that Nvidia’s competitive pressure results in Nvidia’s gross margins converging to the mean (of its sector), i.e. to the 45% range, in 2-3 years time.
Also, let’s assume that the growth of hyperscalers’ AI capital expenditure halves after 3 years, as they redirect the capital in other areas due to the slow GenAI ROI.
This combination of lower margins and lower growth would result in a 16% annual earnings growth at that point, which is very close to the 13% upper end of the investor hurdle we mentioned earlier.
This may not happen exactly in 2026 as we don’t have a crystal ball, however, taking a mid-/long-term view, it is certainly a plausible scenario.
🔮 Conclusions
If you own Nvidia stock, then you are probably safe in the near term, let’s say 1-2 years. Revenue growth is secured and gross margins are well protected. On the other hand, in the medium term, the AI revenue gap and increasing competitive tension are likely to put pressure on the stock.
This is probably a good indication of where we stand with the AI market overall. Relative stability is expected in the next 12-18 months, effectively during the “ROI grace period”. Once this is over, unless we have seen a real impact on SaaS or Cloud revenues, or another technological trend S-curve has started, we are likely to see a correction. Time will tell.
🍽️ Fun things to impress at the dinner table
Honey, I’m home! 1x’s new humanoid robot, Neo, can put in the dishwasher and help with frying eggs. We are hoping for more complex tasks in the future….
See for yourself. Honor’s AI technology enables people with high myopia to see clearly. For now only available in foldable phones, but we should see it soon in smart glasses.
A for creativity. Companies lost money, so it’s not a fun thing. However, sometimes criminal creativity is worth its merit. The FBI caught a fraudster who used AI to create a song and then used AI bots to create thousands of fake streaming accounts that listen to the song in order to generate ad and loyalty revenues. Genius.
See you next week for more AI insights.
Great article! For calculating the expected returns, especially in the bear case and for short/medium term time periods (1-2 years), I would also take into account the PE contraction if the company does not fulfil market expectations
Thanks. Good reference point to judge situation with facts.