Artificial Intelligence – Critical Test

Q3 26 results and GTC are going to be crucial.

• Following renewed jitters around AI financing on Friday, the coming week with GTC Berlin and Meta’s Q3 26 results are going to be crucial when it comes to restoring confidence that the AI build-out will deliver on the promises that it has made.
• The problem has been caused by the general increase in the cost of borrowing driven by government deficits, fears of Anthropic growing more slowly than expected, and mounting delays in the rollout of data centres.
• Almost half of all data centres appear to be suffering from some sort of delay related to permitting, protests or shortage of electricity, the latest example of which has occurred in the UK.
• Here, Microsoft has cancelled its 50MW deployment with Nscale, as the data centre will not be connected to the electricity grid before 2030, some three years later than originally expected.
• While the market is having a case of heebie-jeebies over the problems of the data centre rollout, there is a silver lining to this issue for those who have data centres already in operation.
• The price that a data centre operator can currently charge for capacity is currently at supernormal levels and offers a 5-year return of between 45% and 85% each year.
• This obviously will not last, and as more capacity comes online, the rate is likely to fall substantially to more normal levels.
• However, this is increasingly looking like it will take longer than expected as new capacity is delayed, creating a substantial benefit to those who have capacity available for sale now.
• This is not really the neo-clouds but the existing hyperscalers, where they should be able to charge higher prices for longer as everyone else struggles to add capacity to the market.
• This is most likely to be of incremental benefit to xAI and Meta, as of all of the hyperscalers, these two are likely to have the most incremental capacity to sell.
• However, for the moment, delays are being seen as a negative, as the neo-clouds have so much debt and need to raise so much more that changes in the timing of revenue could cause them financial distress.
• As the neo-clouds are the most financially fragile, they are the companies that everyone is looking at for signs of the AI trade starting to crack.
• This is a very good reason why none of these companies should be public, because their financings would then remain behind closed doors and would expose the market to fewer paroxysms of panic.
• When it comes to the fundamentals, very little has changed other than the interest rate and sentiment, as all of the companies that have reported in the last month are still sold out at least 12 months in advance.
• Hence, 2027 still looks good, although with 25GW+ supposed to come online in 2027, it leads me to think that it is then that we will start to see short-term data centre prices normalise.
• Consequently, I think that this reporting season should see a confirmation of strength from those involved in the AI trade, and any shortfalls are likely to be due to capacity constraints rather than weak demand.
• Furthermore, I suspect that Nvidia will put on a confident show at GTC Berlin this week, which should help the market regain some confidence in the AI sector.
• Hence, I remain pretty comfortable with my positions in Samsung and Meta, and Nvidia continues to look interesting to me.