AI Energy – Two Birds, One Stone.

Going behind the meter should solve a lot of problems.

• Oracle’s problems with rolling out AI capacity and making money from it are once again topping the news cycle in another indication that going behind the meter is probably the best way to solve both the power problem and the public resistance that it has caused.
• Oracle will be forced to provide $7bn in collateral to ensure that its Wisconsin data centre can have adequate power, which could cost it another $100m a year in interest charges.
• This in turn has caused the debt markets more consternation, who have responded by pushing the credit default swap (CDS) to an all-time high of 2.03% (203bp).
• The CDS is the insurance premium that a market participant is demanding to insure the purchaser against Oracle failing to honour its debt obligations and redeem its bonds when they are due.
• Before the huge AI build out began, Oracle’s CDS hovered around 50bp, some 75% below where it is today in a clear signal that the debt markets are uncomfortable with Oracle’s proposition.
• This makes sense because at the moment, Oracle is able to generate around $10bn per GW, which RFM Research has long concluded is not an economic proposition.
• Furthermore, Oracle is borrowing heavily to finance this build-out, explaining why the CDS keeps on creeping up.
• This particular problem is related to the supply of energy, which is both increasingly scarce in the USA (unlike China or the Middle East), but it does have a relatively straightforward solution.
• These installations all consume vast amounts of power and can easily consume all of the power that a single power station can produce.
• Hence, it begins to make sense for data centre operators to also build the electricity generation capacity they need behind the meter, where both natural gas and nuclear power are popular options.
• This is rapidly becoming a better option as opposed to relying on local utilities, who are increasingly having a hard time meeting demand
• A large part of the resistance to data centres being built is the perceived impact of the price of electricity on local residents, despite the economic benefits that such an installation can provide.
• Hence, one solution is to offer residents cheap (or even free for a while) electricity in return for being allowed to build a data centre in the location.
• Given the scale of power consumption, this would not have an enormous impact on the economics of the data centre, and I suspect would go a long way towards quelling any dissent with regard to data centre build-out.
• The problem then becomes supply of gas turbines (long waiting list) and the licensing of small modular nuclear reactor (SMR) designs, which also takes a long time.
• The net result is that data centres are going to take more time to come online, meaning that compute capacity will be constrained for longer, which is a good sign for pricing.
• Hence, if it can, Oracle should try and renegotiate its compute supply agreements because, as far as I can tell, they are both uneconomic and way below where the spot market currently is.
• The real winner here is OpenAI which, if the contracts are watertight, has secured compute at very good prices, meaning that it will be able to offer its services at much lower prices.
• We are already seeing signs of data centre operators moving to build their own power and given its potential to dispel dissent, I suspect it will continue to increase in importance.
• Oracle fits into my theme of companies that will thrive as a result of AI rather than be destroyed by it, but not until it can show better economics from the compute contracts that it has.