EV Vs AI: Clash of the Titans

 A little about me: I am a transportation-fuel agnostic. I believe there is room in transportation for any number of fuel solutions. I haven’t (and won’t) go into the highly politicized carbon reduction topic. I will only say that my interest in alternate fuels developed long before the public push towards carbon reduction and proponents of any technology were hell-bent on pushing their specific solution as doctrine. It started as a desire for clean air (I have family with asthma) and energy independence. The clean-air topic speaks for itself, and I think the country’s cyclical fuel pricing challenges (particularly since the Strait of Hormuz-Gate incident) adequately illustrates why energy independence is a good thing.

In my long and storied career (ask others for stories), I have seen the ascension, decline and re-ascension (and decline again, in a few cases) of no fewer than 5-alternate fuels: Biodiesel, Electric, Ethanol, LPGas  and Natural Gas. All of them faced a rigorous adoption battle, and all worked particularly well in certain/very specific applications, until they didn’t. In the Fleet world, it appears that what steers alt-fuel adoption is upper management commitment and simple financial constraints. Don’t get me wrong; domestic corporations and their fleet people are happy to make purchasing decisions that offer positive environmental impact, but those decisions are easier to justify if they can also save money.   

In a recent Clean Cities board meeting we were having a casual discussion that posed questions regarding EV adoption timeline. Being the only big-truck guy on the advisory board, I was asked when I thought EV big-trucks would own some actual market share. My response was quick and candid: “I’m not sure they ever will”. My answer was based on knowing the demands placed on trucks running in municipal and utilities operations (my wheelhouse for 30+ years now); how limited battery storage, significant battery weight/packaging restrictions and high charging cap-cost all fight positive outcomes.

Later, after reading about yet another plan for a regional datacenter (this will be proposal number 10 in our area), I realized that my response was only a partial answer at best. Datacenters and EV chargers have this in common: they are both very power-hungry. So power-hungry, in fact, that it would almost appear that at least the AI server-farm guys are taking their cue from the Matrix movie franchise. Of course, I always wanted to be an organic battery when I grew up, so it’s all good. I am only half kidding, Mr. Anderson.

Let’s look at the current demands (intentional pun? You decide):

An AI worthy (hyperscale in the vernacular) server farm requires approximately 1kW of power per sq/ft to run at full load. This same hyperscale server farm, right now, occupies about 100,000 sq/ft (2.3 acres). Multiply these power requirements by gross sq/ft and a typical hyperscale server-farm will require 100 megawatts of power to operate at full load, give or take. A side note is that recent AI deployments favor business functions, so the highest-demand usage patterns will handily fall into utility peak-demand periods (can you say brownout?).

Trucks are easier to calculate load using unit-based consumption. Running commercial level “Fast DC” chargers (360kW units – typical heavy truck fast chargers), a 50-truck class 7-8 charger installation (modest count for a commercial depot) will consume about 19.5 megawatts at full load. My own agency happens to run a little over 400 class 7-8 trucks, so that would be 150 Megawatts or so all in. So, if all the data-center plans actually result in10-hyperscale server-farm installations, by the time you add-in (just our agency’s potential) 400 trucks, you have the possibility of a 1,150 Megawatt load at any one time.

To put this into perspective, an average conventional power-plant can produce about 500 megawatts of electricity at maximum capability. That means you can only run 5-hyperscale data centers, or charge only 1,250 class-8 trucks per powerplant, at full load. Not much. For perspective, that same 500 Megawatt power plant would run about 420,000 houses, give or take. So those 10-server farms and electrification of this 400 unit heavy truck fleet (in this case) can consume the entire output of 2¼ utility-grade powerplants, alternately powering over 945,000 homes.

Pragmatically, recent forecasts for fleet electrification and datacenters have been aspirational at best – the figures are steered more by marketing releases and letters of intent than committed projects. But if it does come to pass, how do we propose to feed all the server-farm, EV, manufacturing, business and residential electrical needs? Regions of the nation’s electrical grid are already strained with existing load. Even our region experienced rolling blackouts trying to hold up the grid during a significant power failure in Texas back in 2021.

I’ve read that there are several legislative proposals (various state and federal dockets) demanding that all AI companies provide their own power. I’d ask how? A 100 megawatt solar array occupies about 500 acres. A wind-farm doubles that space requirement. Parcels of land this large are rarely available anywhere close to metropolitan areas. Also, neither can produce power 24/7/365, so you must figure conventional electrical generation in somehow; combustion turbines (natural gas or diesel), conventional diesel internal-combustion powered generators or steam-turbines (natural gas or coal fired). There goes any Green hope.

In the end we may have to fundamentally re-think either our ambitions or our methodology. The logical choice favors economies of scale by sourcing the heavier electrical demand out to a power utility.  Maybe they reexplore nuclear power – clinically “clean”, but we are culturally phobic about breached reactors and nuclear waste disposal (two-mouthed genetic mutants are ideally suited to be politicians though!). Or maybe we spin off the lower-load residential generation to neighborhood electrical cooperatives. In this scenario burning fossil fuel will be in the supply chain somewhere (again, the Green challenge).

These solutions don’t sound fun or easy, but AI operators would stand a chance of getting electricity to do their deed; but only because server farms are stationary. Fleets operate mobile assets, and as they are mobile, their depot is often moved. Leases expire, capital projects develop or the company’s business forces location changes. Next thing you know, the assets need to report across town. This is more common than you may realize; I’ve seen this location restructuring happen more than 10 times since I’ve worked in major fleet, so figure an average organization will move its inventory once every 3-years.

It is very difficult, time-consuming and costly to remove, transport and reinstall 100’s of 1,000’s of dollars’ worth of chargers. And at the power levels needed for even modest commercial EV deployment, it is unlikely that any facility your management suggests has access to the required electrical service. So now even more money is required to establish electrical service before installing the chargers, assuming the local grid can support that load.    

The historical pattern in technological deployment has been this: Human ambition overshoots practical limitations, expectations are scaled back and the technology deploys in a smaller scale organically; speculative failure, the way of capitalism. But in this case, with the extreme pressure from both sectors to get market share, I wonder if this paradigm will hold true. If the two factions decide each of their interests are too important to scale, we may be fixin’ to be witness to a clash of the Titans on an epic scale.