Source item

Quoted context

These are the quoted author’s words.

As it relates to the $CRWV IPO on Friday, a solid deal is one where the stock is up 15-30% from its IPO price on the opening day of trade. The issuing company is happy at not having underpriced the deal too much in raising capital. Investors receiving the IPO shares are happy for being compensated for the risk of investing in a company with no public market history. It is a win-win for both the buyer and seller. On the other end of the spectrum, a stock that closes below its IPO price is what an underwriter wants to avoid at all costs. I was an investment banker in the early 1990s but these rules of thumb still apply today. Coreweave on Friday closed at its exact IPO price of $40.00 to the penny. The CEO also said in an interview that without the support of $NVDA, which is already a 6% investor and 15% of Coreweave revenues, buying shares in the IPO, the deal “wouldn’t have closed.” What is more troubling to me is the news story that $MSFT, a 62% customer of Coreweave, turned down a $12B option for more capacity. OpenAI reportedly took this capacity but Microsoft generates a lot of free cash flow to fund their spending while OpenAI loses a tremendous amount and will need to raise capital in the future to fund their capex desires. This tough IPO also follows the statement on 3/25 by the $BABA Chairman on US datacenters who said, "People are talking about $500 billion, several hundred billion dollars. I don't think that's entirely necessary. I think in a way, people are investing ahead of the demand that they're seeing today." This in a vacuum does not bode well for how Coreweave is likely to trade this upcoming week or related AI names. Having said that, the reaction of the market to April 2nd reciprocal tariffs is likely to be the biggest driver of stocks this upcoming week. Near-term, I continue to believe that until there is a killer consumer app for PCs or smartphones that drives inference demand, AI capex spending by mid-year will be going through a digestion phase due to a slowdown in training demand. The largest AI models are already training on all the relevant available data on the internet. In addition, I believe that foreign companies have bought as much AI chips as possible in front of potential upcoming export controls. This has likely overstated demand in the near-term. Spending is already forecasted to slow to the mid-single digits sequentially during the the first half of 2025 versus mid-teens sequentially during 2024 for the largest AI capex spenders. Also capex depreciation may need to increase given the useful life on existing AI infrastructure is depreciating more quickly. Over the past several years, most large internet players have increased the depreciation life of their compute infrastructure to 5-6 years. This has shifted costs from the income statement to the balance sheet which has benefited EPS. But with Nvidia now on a yearly cadence for introducing new chips, this is probably too high and will need to come back down. As the Nvidia CEO said on 3/18/25: “I said before that when Blackwell starts shipping in volume you couldn’t give Hoppers away… There are circumstances where you are fine, not many.” This is an added risk to earnings guidance for the largest hyperscalers as we go through 2025. For those who over the near-term want to stay tangentially invested in the AI theme, I believe $CSCO is in the early stages of a refresh cycle. Enterprises need to upgrade their networks for the AI workloads generated from all the infrastructure spend over the past two years. Cisco organic product orders accelerated for 4th consecutive quarter in their January quarter. Total product orders were up 29% y/y with 3 of 6 largest webscalers up 100%+. Trading at a 16x PE on CY25 EPS estimates, the valuation is also compelling compared to the S&P at 21x. Longer-term, I still believe AI capex spending is likely to double by the end of the decade from current levels driven by inference demand. Inference demand should dwarf training demand ultimately. During the internet infrastructure buildout, Cisco revenues went up 15.5x over six years from the introduction of the first mass internet web-browser in late 1994, NetScape Navigator, till the peak of the internet infrastructure buildout. Cisco became the most valuable company in the world at one point. Nvidia revenues by comparison are up “only” 6.6x over the past 2 ¼ years of the AI infrastructure buildout since the launch of ChatGPT in late 2022. Also the biggest AI spenders today are some of the most profitable companies in the world versus the unproven business models of the companies during the internet buildout. To be clear, I am not saying that this is what I expect from Cisco this time. It is just a historical reference to compare Nvidia’s revenue growth as the key enabler during the current AI investment cycle so far to what Cisco’s growth was as one of the key enablers during the historical internet infrastructure buildout. In summary, the key will be to remain intellectually flexible and data dependent as we go through 2025. As Niels Bohr, the Nobel laureate in Physics, said: “Prediction is very difficult, especially if it's about the future!”

DanielTNiles ↗