The Watt-Bit Spread

“Every company is an energy company, and if it’s not now, it will be soon”

-Deloitte brochure I had on my desk 10 years ago

There is a fundamental disconnect today between the cost of a watt and the value of the bits created by those watts. I call this the Watt-Bit Spread, and it's never been higher than it is today. This disconnect is intractable, but not hopelessly so. To understand better, it is helpful to put it in the context of manufacturing theory. There are two fundamental ways to run a manufacturing process, and they are centered around what you are optimizing for: minimizing overhead to protect profit margins or maximizing throughput to increase revenue. The former approach is exemplified in a lean manufacturing model where the latter better aligns with the theory of constraints (see Eliyahu M. Goldratt’s The Goal, 1984).

The AI Imperative = Maximize Throughput

The size of the potential AI market creates an economic imperative to plug in as many GPUs as possible, as quickly as possible. Building the largest and most sophisticated AI models will create a moat (not dissimilar to what occurred with commercial cloud) making it difficult for future competitors due to the technical and capital advantages of the largest first movers. As has been reported extensively, the key constraint these companies face today is access to power - see great reporting from Jennifer Hiller (WSJ), Brad Plumer (NYT) and many others. What results is a high value for power that can be delivered within lead time to build data centers and deploy servers. In other words, power delivered in 2027 is worth significantly more to these customers than power in 2030. But the market for this power reflects something quite different.

The Utility Imperative = Minimize Overhead

Utilities operate under an altogether different paradigm that does not currently reflect the economic realities of the digital world. To a typical electric utility an electron delivered in 2027 is valued the same as an electron delivered in 2030. Utilities do not differentiate between these two products. So here we find a critical market inefficiency where the value of a product in a given year to the customer is not being reflected in the pricing or revenue opportunity for the seller. It’s not that utilities have no motivation to respond to the market, it’s simply the way they sell their product (via tariff) and the manner in which they are regulated (minimize risk for existing rate payers) results in market inefficiency.

The response of many data center operators to this conundrum is to simply dispense with the utility all together and pursue generating their own power. While this can solve some short-term problems it seems to overlook the fact that the companies providing the gas service are also, in fact, utilities constrained by the same economic realities discussed above (I will defer a discussion of nuclear for now as, beyond a handful of potential restarts, new nuclear is not relevant for deployments this decade). In other words, there is no simple fix for this problem. What is needed is a new regulatory paradigm that enables utilities to capture some of these excess rents being offered by the market to enable them to accelerate investment in their systems to capitalize on this load growth opportunity.

The scope and scale of the AI challenge is a grid scale problem and will only be solved with grid scale solutions. Fortunately, we have technology that is ready today (e.g., grid enhancing technologies) and technology that will be ready soon (long-duration energy storage, superconductors) to help address this challenge. But the technology itself is insufficient unless we focus heavily on how to alter the commercial incentives that hold back investment and innovation. Nevertheless, I’m optimistic that the economic incentives on both sides (datacenters and utilities) are sufficient to catalyze action that has the potential to transform the electric grid. While it is certainly a challenge, the demand for electricity stemming from AI represents the best opportunity we have to expand and modernize the grid to the benefit of all electricity consumers.