Watt-Bit Spread (part 2)

Having determined in my prior piece on this topic that there is a spread between the value of a bit and the cost of a watt the next question is what is the economic cost of this spread? This cost shows up in two distinct ways 1) the economic value being lost for power that is undelivered that could have gone to the production of more bits and 2) the power that is delivered but where the excess economic rents are not being allocated appropriately and thus creating further market distortion.

Uncollected Rents = Economic Loss

The deployment of GPUs and growth of AI will be constrained by access to power. Let's just assume that the demand for compute by 2030 equates to 50 GWs, and the available supply of power is only 30 GWs. In that scenario, 20 GWs of investment potential will be lost, which is the equivalent to ~$500B of foregone capital investment in digital infrastructure not to mention the associated profits and lost investment from power infrastructure to support that load. Therefore, failure to invest in delivering power on a time horizon that meets the demand for GPU deployment results in significant economic loss.

Collected Rents = Inefficient Capital Allocation

I noted in my prior piece that utilities are losing out on the potential to capture excess rents (read capital that would otherwise be invested in grid expansion). As noted above, some of these rents are being lost entirely. But what of that which is being collected? Perversely, that economic benefit is accruing to landowners with preferable load-side interconnections. In other words, the value of the power delivered in a given year is reflected in the cost of a piece of land, not the delivered cost of power. So, landowners are receiving excess market rents that would otherwise go to utilities to invest in their system.

Quick aside on Cloverleaf. At this point you might say wait a minute, isn't Cloverleaf's business model that of buying land to capture these excess rents that you just said should go to utilities? Not exactly. Cloverleaf is actually focused on the first point which is how do we work with utilities to rapidly expand access to power for land that otherwise would not have it. In other words, we are working to expand the overall size of the pie to increase the available deployment of digital infrastructure to the benefit of both utilities and the digital industry.

Solution = Advanced Grid Tariff

What is needed is a new form of tariff that enables rapid investment in expanding grid capacity and cost recovery targeted at those customers most willing to pay. I call this an Advanced Grid Tariff (AGT), credits to Katie Fehrenbacher for naming this tariff for me. This is not dissimilar to green tariffs that have been deployed to encourage the development of renewable energy, geothermal, etc. However, the incentive mechanisms in this tariff are aimed at delivering firm and or flexible grid capacity into the market as quickly as possible leveraging grid enhancing technologies, storage as transmission, and ultimately enabling longer term capital investment and expansion of rate base for electric utilities. This would better align the willingness to pay for access to power for those interested in investing in digital infrastructure. This would also benefit local communities in two ways. One it accelerates capital investment that will directly result in jobs, economic growth, and expansion of the tax base. Second, it would insulate these same communities from rate increases that might be associated with utility investment in grid infrastructure.

I know some utilities will read this and say wait a minute. I've been talking to XYZ data center company and when I put in front of them the full cost of infrastructure necessary to serve their load they seem to get cold feet. Does that mean the size of the watt/bit spread is not as large as you say? Or that the demand for AI is overhyped? Absolutely not. But it does reflect a further inefficiency in the market that invites a new reality in terms of risk and capital allocation that will be necessary to unlock the full potential of this market. I will discuss this in a future post.