2024 June CAD

Back to archives

0.26% MTD
0.18% YTD

Caravel Capital June 2024 Partner Letter

Dear Partners,

For the month of June, Caravel returned +0.26% vs. +1.09% for the benchmark (S&P 500 +3.59%; SPTSX -1.42%). June brings Caravel’s Q2 and YTD performance to +1.45% and +0.18%, respectively. While these levels remain modest and well below our targeted returns, we are pleased with the turnaround this past quarter and excited about the setup in H2. Our number one goal, and one of the few we try to keep at the forefront of our process every day, is capital preservation. We believe this is especially important now, when there are many shiny new things to chase in the market. More on that below.

The runaway train that has been the S&P 500 continued barreling through the first half of the year, posting an impressive +15.3% total return. This performance has however, as we and many others have noted, been propped up by the largest few stocks with the most exposure to the AI theme that first emerged in earnest in Q4 2022. To put this in perspective using two observations:

  1. The top 10 largest stocks in the S&P 500 are, of course, collectively only 2% of the index’s membership. But they make up nearly 40% of its value and accounted for nearly 80% of its performance in the first half. Importantly, this is counterbalanced by the fact that they constitute over 30% of the index’s total earnings.
  2. Said differently, the cap-weighted S&P 500 outperformed its equal-weighted counterpart by more than 3x in the first half of 2024.

This dynamic has caused many active managers to underperform, as it is difficult to hold such concentrated positions in just a few stocks and still tell your clients they are investing in a diversified product with a straight face.

Let’s be clear, this is not an exercise in sour grapes. We hope our clients have ridden the AI wave elsewhere in their portfolios even if it means answering some tough questions the next time we speak. We would, however, like to try to assess where we are and where we might be going next.

There was a fascinating article published in June by David Cahn. Cahn is an investor at Sequoia, one of the world’s leading venture capital firms. We encourage you to read the entire article, but will summarize some of his most salient points. He poses a seemingly simple question: where is all the return on the large S&P weights’ feverish investments in AI going to come from? This is framed as “AI’s $600B Question,” so named based on a straightforward set of assumptions:

  1. Nvidia (NVDA) will exit its fiscal 2024 with annualized datacentre revenue of $150 Billion.
  2. Someone will be spending $300B/year on datacentres to which NVDA provides GPUs.
  3. That someone must generate $600B of annualized revenue from AI applications to justify $300B of annualized expenditures on data centres.

Assumption #1 is fairly vanilla, as it lies squarely within the realm of consensus estimates for NVDA. Assumption #2 is also rather uncontroversial, but it is staggering. GPUs are roughly half of the cost to build a datacentre, which suggest that someone’s $300B (NVDA’s $150B / 50%) datacentre spend will not only have to hold, but grow rapidly for the AI semiconductor stocks to maintain their high growth and valuations.

Assumption #3 is where things get interesting. NVDA’s revenue is some other organization’s expenditure, as are the other (non-GPU) costs to build and operate a datacentre. Nobody is spending $300B (and growing) per year without some (very high) expected return on investment. Cahn assumes another 2x markup such that the provider of the AI compute can make a 50% gross margin, or a 100% gross ROI ($300B / 50% = $600B of revenue). If anything, we find this to be conservative. For example, Google’s gross margins are near 60% and Microsoft’s are closer to 70%. This suggests that Microsoft’s hurdle rate might be closer to a 200+% ROI than the 100% assumed by Cahn, but let’s just use his numbers for the sake of argument. Also implicit in this assumption is that the GPUs will have a very short useful life (~1 year), such that what Microsoft or Google buys from Nvidia is closer to OPEX than CAPEX. This could well prove to be true given the rate of competition, Moore’s Law, and other factors.

Netflix has ~270 million paid subscribers. Spotify has ~240 million. These are mature software companies with excellent offerings. But, for the sake of argument, let’s assume the ‘someone’ spending $300B a year on AI infrastructure gets to 500 million users. $600B of annual revenue means that someone is able to generate $1,200 per year for each of these users on average, ostensibly through subscriptions or advertising. While it is not difficult to envision a world AI tools would generate $100 per month of value, we’d argue that we are not there yet. Remember, $600B is the implied number coming out of this year. As Cahn points out, OpenAI, the market leader in generative AI applications, just surpassed $3B of revenue despite already having 100 million users. In summary, the rate of improvement and, most crucially, perceived value of AI applications must grow at a rate much faster than AI spending itself in order to close this gap. Given that NVDA’s stock has risen 920% over the past seven quarters, this is a very high bar.

The market seems to have already sniffed out some of these concerns. So far in July, a soft inflation print has put a September interest rate cut back on the table in the US. This has prompted a (still early) rotation out of megacap tech and into some of the areas of the market that have had less shine so far this year. This is what makes us optimistic about the back half of the year. While we won’t try to call a top in the AI trade and certainly wouldn’t short it, we are excited about the prospect of some capital finding its way into the many solid opportunities we have already identified. Think single-digit multiples, structurally advantaged competitive paradigms, 20+% growth rates, proven business models, and rock star management teams. If we hadn’t rambled so much already we’d get into specifics, but perhaps that had best wait until next month.

Thanks for reading.

Best,

Jack and Glen

Monthly Performance (net of all fees)

JanFebMarAprMayJunJulAugSepOctNovDec YTD
20241.74-1.70-1.260.930.240.260.18%
2023-3.42-.95-0.11-0.07-3.192.221.57-0.222.06-0.762.211.180.32%
20221.151.02.93.10-1.61.82-1.61-0.33-8.490.06-.090.68-7.5%
20213.403.993.751.271.301.540.221.514.893.700.501.2030.78%
20200.41-.20-1.91.741.662.251.263.131.100.572.043.1515.02%
20191.721.793.131.151.35-0.75-1.54-1.340.04-1.45-2.571.392.76%
20186.364.810.950.71-0.85-1.072.501.693.530.670.02-0.1820.58%
20170.270.050.350.251.391.451.770.123.273.6113.961.9631.51%
20161.593.301.53-0.825.67%

Risk vs. Return Comparisons Across Indexes

Month Return YTD Return Volatility Sharpe Sortino Beta Best Month Worst Month Annualized
Caravel0.26%0.18%8.36%0.911.271.0013.96%-8.49%11.88%
S&P 5003.59%15.29%16.22%0.71.060.0912.82%-12.35%14.49%
S&P/TSX-1.42%6.06%13.58%0.340.40.0910.79%-17.38%8.61%

Growth of $1000 since inception

Commentary Archives