Key points in this equity note:
- Nvidia’s earnings tonight after the US market close with analysts expecting 65% y/y revenue growth is the biggest short-term equity risk event. Recent volatility highlights the big anticipation and positioning ahead of the earnings release.
- Analysts have aggressive revenue projections with low variance mimicking the crypto boom period 2019-2021.
- Capital expenditures from Baidu and Microsoft are the only ones that show a bit of AI bonanza while the rest of the technology companies are not showing an AI boom in their capital expenditures for Q2.
- The three biggest risks to Nvidia’s outlook are 1) in-house AI chip design by Nvidia’s largest customers, 2) less excitement of LLM models such as ChatGPT as performance improvements stall, and 3) supply chain bottlenecks feared to not be resolved until the end of 2024.
Nvidia earnings are biggest event risk in equities
Nvidia is undoubtedly the biggest short-term event risk in equities, except for the new highs in the long-term bond yields, reporting FY24 Q2 (ending 31 July) tonight after the US market close. Analysts are playing it safe estimating $11bn in revenue for Q2 up 65% y/y which basically reflects Nvidia’s own guidance on 24 May at $11bn +/- 2%. The confidence band that Nvidia provided three months ago is evident of Nvidia being quite sure they will hit this revenue figure and thus we have to assume that their order pipeline at that point in time was quite predictable. Nvidia has also not changed their guidance so investors have to assume Q2 revenue will be closer to the guided range.
The chart below shows analysts guidance over the next seven quarters which basically shows that analysts expect Nvidia’s growth to go back to the crazy years of the crypto bull market that drove demand for GPUs at a record pace. Prior to the crypto boom years Nvidia’s revenue had much more q/q volatility. Revenue growth q/q had a standard deviation of 14.5% using data since 2013. The expected standard deviation on revenue expected over the next seven quarters is just 3.4%. This different is significantly different on a test of variance (also called F-test) indicating a high probability of being wrong.
Revenue estimates have continued to climb for the subsequent quarters since Nvidia’s blowout guidance in May and on Monday the shares closed at $469.67, just below their peak in July, in what was a 8.5% rally in a single day highlighting the attention and positioning in Nvidia shares leading up to the earnings release. Yesterday, Nvidia shares were down 2.8% against the previous day’s close and down 5.1% from the opening price. Volatility is clearly excessive before the earnings release. Nvidia shares were the second most traded single stocks instrument yesterday by Saxo clients with around 58% of transactions being buy orders, so a tilt in favour of a positive surprise and a sign that retail investors are still buying on a daily weakness.
If analyst estimates hold for FY25 (ending 31 January 2025) then Nvidia is valued at a forward free cash flow yield of 2.5% which much less than the historical 3-6% level before the crypto boom years during the pandemic.
Nvidia share price | Source: Saxo
Microsoft’s capital expenditures are the only AI boom town
Yesterday’s earnings from Baidu was quite interesting to watch because according to this recent FT article, China’s largest technology companies such as Baidu, TikTok, Tencent and Alibaba were massive buyers of Nvidia’s GPUs. As we have highlighted in our previous notes it has in fact been the Chinese geographical segment that has increased the most in the previous quarter confirming this hypothesis. We were therefore expecting to see a massive increase in Baidu’s capital expenditures. Why, you might ask? Because AI research and development requires a lot of expensive GPUs and would normally be depreciated and not expensed (it would cause a too big hit to accounting profits). Baidu’s capital expenditures were $371mn in Q2 up from $189mn in Q1 and $279mn in Q4 2022, so Baidu did confirm an increased spending. But to call it a boom would be an understatement as capital expenditures at Baidu were running around $300mn quarterly in the period 2020-2021.
Tencent, another key Chinese technology company, actually reported lower capital expenditures in Q2 compared to Q1 and is investing half of what it used to just two years ago. Alibaba, being classic non transparent, does not show its capital expenditures, so this figure is hidden for investors.
Nvidia’s largest US customers are Amazon, Microsoft, Google, Meta, and Dell. These companies excluding Dell do also capitalise and depreciation investments in GPUs while for Dell the GPUs are expensed over the cost of goods sold with GPUs estimated to be roughly around 20% of PC and laptop sales. Dell’s Q2 revenue was $20.8bn down from Q1 and way less than the $26.4bn in Q2 2022. So it is definitely not Dell that is pushing Nvidia sales higher. It is in fact only Microsoft capital expenditures that exploded in Q2. They were lower for both Meta and Amazon, while the growth at Google was tepid. The chart below shows the total capital expenditures for Baidu, Amazon, Microsoft, Google and Meta. It does not show an explosion in investments.
If we assume that Nvidia’s estimates of demand is correct then the only explanation for capital expenditures figures not aligning well with estimated Nvidia revenue is that the capital expenditures mix has changed dramatically in favour of AI chips under overall restraint on technology spending. Another curious observation is that Nvidia’s revenue is expected to hit $74bn in FY26 (ending 31 January 2026) which would be a significant portion of the capital expenditures spent by the world’s largest technology companies, something that seems a bit too optimistic given the more muted outlook on AI from Microsoft in their recent earnings release.
Hang Seng continuous futures | Source: Saxo
For the Chinese companies it makes no sense, so here the only plausible explanation must be that these AI chips are brought by other Chinese companies on behave of the largest Chinese technology companies. With the US putting stricter export controls on semiconductor chips to China there is an incentive to circumvent these restrictions in clever ways.
One supporting factor for Nvidia’s GPU sales is the gains in Bitcoin this year which have made Bitcoin mining more profitable again increasing the demand for GPUs.
Key risks for Nvidia
While the long-term prospects for Nvidia are undoubtedly positive as various AI systems will continue to grow into the future increasing the demand for AI chips there are short-term risks to the outlook. These are critical to be aware of as an investor.
- In-house design from leading technology companies – just as Apple cut Intel out of its smartphones with its own in-house designed M1 chip designed specifically for the needs of the iPhone, so could other technology companies do on AI chips. Google is already far in those efforts and Tesla has said that it is also working on its own AI chip for self-driving cars. The incentive is of course better control but also because of Nvidia’s 46% net profit margin indicating large savings for companies such as Microsoft, Amazon and Google. Is it doable? Yes, because Nvidia is an IP company which has outsourced its manufacturing to TSMC. This is also why Nvidia was so keen on acquiring Arm from SoftBank which was turned down by regulators as the acquisition would have led Nvidia controlling the chip industry and limiting technology companies in pursuing their own chips.
- LLM excitement will soon disappear – indicators are already suggesting that user numbers on ChatGPT are declining and it is a very costly affair for OpenAI to run ChatGPT. As Sam Alt man, co-founder of OpenAI, has already said, large language models (LLMs) which are behind ChatGPT are not the road to artificial general intelligence (AGI) and that bigger models are also not the future. This could mean that as soon as the five largest technology companies have trained their foundational models the need for more AI chips will be more tepid going forward. The “model collapse” theory, which says that as generative AI models create more and more content on the Internet their future training will be more and more polluted and stuck in a local minima leading to no further model improvements, is another potential risk as it could lead to significantly less excitement in LLM models over time as their performance hits a ceiling, or even worse, declines due to polluted datasets.
- Supply chain problems – according to the semiconductor industry it could prove difficult to meet the expected demand from AI chips with bottlenecks not being fully resolved before the end of 2024. That could become a key constraint for Nvidia meeting those current expectations priced by investors.
- The last risk is just good old competition with AMD getting closer and closer to Nvidia in terms of product offerings.
The equity notes below highlights some of our previous perspectives on the AI boom and Nvidia.
Peter Garnry
Head of Equity Strategy at Saxo Bank
The information in this article should not be interpreted as individual investment advice. Although Hugo compiles and maintains these pages from reliable sources, Hugo cannot guarantee that the information is accurate, complete and up-to-date. Any information used from this article without prior verification or advice, is at your own risk. We advise that you only invest in products that fit your knowledge and experience and do not invest in financial instruments where you do not understand the risks.