Tuesday , November 17 2026
Latest
Home / Energy / The problem restated

The problem restated

Inevitably, my recent post AI is not a flyer drew the two responses that it criticised.  AI, critics explained, really was about to destroy humanity (something that humanity seems far more likely to achieve all on its own) although it may cure cancer along the way.  What few seemed to grasp is that the thing being labelled ‘AI’ is mostly cloud software which requires an insane quantity of computing power so energy intensive that it can never be profitable.  The $500 billion Stargate datacentre in Abilene Texas, for example, requires more than a gigawatt of electricity – that’s about the same amount of electricity needed to power Newcastle-upon-Tyne in the UK, and far more than is available from the local grid.  So that, the proposed datacentres must each install their own dedicated electricity generation – mostly new gas power, but with some nuclear and a lot of diesel back-up generators.  Which is why the proposed projects run to hundreds of billions of dollars and – crucially for anyone foolish enough to have invested in them – are likely to take decades rather than months to complete… and that is a massive problem for reasons I will explain below.

First, however, I want to restate the main issue in my previous post through the lens of thermodynamics (which ought to be the introduction to every economics course).  In particular, we need to understand the real-world impacts of the first and second laws of thermodynamics.  The first law is the law of energy conservation – that energy can neither be created nor destroyed but is merely transformed from one state to another.  Add a small spark to the chemical energy in gunpowder or petrol, for example, and we get a lot of kinetic energy and even more heat.

A large part of what we call human civilisation involves the harnessing of some of that energy to produce useful work… exergy.  Exploding the gunpowder inside a metal tube allows us to focus the kinetic energy to propel a projectile.  Exploding the petrol inside a cylinder connected by a piston and crankshaft via a gearbox to an axle allows us to convert the energy into motion.  For the earliest humans using the sugars locked up in muscle tissue to knap the edge of a piece of flint to create a primitive tool to an army of technicians regulating a nuclear fission reaction to provide a city with electricity, the basic process is the same – converting energy from one state to another to provide and harness exergy.

So far so good, but the conversion of energy comes with the massive downside expressed in the second law of thermodynamics.  This is often stated as the law of entropy… that things break down or move from hot to cold.  When applied to human systems, it means that whenever energy is converted from one state to another it always results in a loss of waste heat – i.e., heat that is no longer available for human use (even if it hangs around the universe at any temperature above absolute zero for eons).  This sets up a perpetual productivity struggle in which we use and develop energy-specific technologies to maximise exergy while minimising the loss of waste heat.

This explains the current ‘productivity puzzle’ that exercises so many modern economists and politicians.  The period following the reconstruction after the Second World War (1953-73) witnessed an explosion in productivity.  But from the 1970s, productivity growth began to falter.  And in the first decade of the twenty-first century, productivity ground to a halt.  And no mainstream credentialed economists could offer any insight into why this had happened.  An interested physicist, on the other hand, would explain it in minutes.

Put simply, productivity means either getting more output from the same input or getting the same output from reduced inputs.  The latter involves cutting waste – which is almost always over-estimated and usually results in unforeseen complications that reverse any gains made.  The former is generally associated with the deployment of technology (in the broadest sense of the term) to apparently increase output.  I say ‘apparently’ because contemporary beliefs to the contrary, technology is something of a middleman here.  It is, in fact, energy which is still doing the heavy lifting.  Technology is merely seeking to increase exergy while decreasing waste heat.  And, importantly, with the exception of heaters, human technologies do a really bad job of maximising exergy.

You might, for example, assume that your car – which can propel you along at 70 mph over hundreds of miles – is a pretty efficient (i.e., productive) technology.  And in a sense, you are correct.  It is a version of a technology which has been around for 130 years.  During that time just about every viable improvement both to the technology and to the driving environment has been made – while few of those making the improvements would have seen it in this way – to maximise exergy and minimise the loss of waste heat.  So, what percentage of the energy in petrol is lost as waste heat in an average car in the 2020s?  It turns out that even today’s optimised vehicles lose between 70 and 80 percent of the energy converted as waste heat.

In many instances, exergy hits a physical limit.  Modern solar panels, for example, are close to hitting the Shockley-Queisser limit.  More often though, technologies hit an economic (affordability) limit long before they reach the thermodynamic limit.  This is why technological (productivity) improvements tend to follow an ‘S’ curve.  Early improvements tend to be cheap and easy but later, as exergy/waste heat limits get closer, the improvements become more difficult and expensive.

This is why I often point to the A4 Pacific class locomotives pulling luxury trains along Britain’s east coast main line in the 1930s or the Anglo-French Concordes flying politicians, pop stars and corporate billionaires across the Atlantic in the last quarter of the twentieth century.  It is certainly possible that throwing trillions of new dollars, euros and pounds at these technologies might produce a few extra horses’ worth of exergy and a few degrees less heat loss.  But in an economy which has been gradually slowing for the best part of half a century, would the additional productivity be worth it?

Notice that the A4 Pacific class Mallard and the Concorde were the civilian high point of two different energy sources.  Travelling at 126 mph in 1938, Mallard marked the economic endpoint for steam propulsion.  And because of its cost, businesses and passengers soon settled for slower but cheaper journey times.  In the same way, Concorde was the apex point for oil age technologies, with businesses and passengers soon settling for slower but cheaper Atlantic crossings.  Each energy source comes with its own discrete suite of technologies that exploit it.  And the more potential power (energy density) the energy source, the more exergy (productivity) can be obtained.  Even the relatively small (10mj per kg) difference between coal and oil is the difference between the modern (post-1950) world and the world our Victorian ancestors lived in.

Computing – and particularly software – marks the high point of electronic technology enabled by the development of stable electricity grids in the last quarter of the twentieth century.  Software has an almost magical appeal since the processes which enable it take place out of sight of the end user.  This is particularly true of cloud software which seems to consume a mere flicker of energy for the end user because the action takes place in a datacentre hundreds of miles away.  And this, in turn, has created the illusion that software is free.

This is certainly the business case.  A piece of software, whether a game or a business application, is created once but can be duplicated almost to infinity… which is why the smart money moved from hardware to software several decades ago – the smartphone and the mobile communications network being the last serious hardware update.  In 2025, Microsoft pulled in more than $120 billion providing access to software such as its 365 Office suite, which currently has more than 89 million subscribers worldwide.  Meanwhile, Netflix pulled in $45.18 billion selling access to the same digital video files to its 325 million global paid subscribers.  It is in this context that AI was being sold as the next hypergrowth business… so long as nobody looked too closely at the cost of building and operating the new infrastructure.

The energy cost of AI – at least as it is being proposed in the USA – is far too high for the companies involved to come close to making a profit.  And unlike the dotcom bubble or the railway mania of the 1840s, there is no useful infrastructure roll-out to be repurposed.  Even the proposed new gas power stations are going to become white elephants in short order because, despite the USA’s apparent surplus of shale gas, it only has some 17 to 20 years of proven (i.e., affordable) reserves at current rates of consumption.  Add in all of those AI datacentres and sometime in the next decade we’re going to need a new planet to provide the additional gas.

In theory (although nobody knows how to do it) the potential energy in the nucleus bonds of fissile atoms like uranium – a soda can filled with uranium has roughly the same energy as a 2,000-tonne coal train – might one day allow humans to casually operate the proposed AI datacentres.  But in the absence of any realistic breakthrough in nuclear technology, the more likely outcome is that potential customers cancel their subscriptions once the true cost of AI is (attempted to be) passed onto them; investors desperately (fail to) get their money back; a few of the tech bros (hopefully) get jail time for fraud; and less ambitious DeepSeek-type AI software running on home computers becomes the (far less energy intensive) way of generating video, images and chat for most users, with a handful of businesses running dedicated AI software on their own servers.  At which point, we mere mortals will be left to ponder what we might have done had investors and governments put those trillions of dollars to better use.

As you made it to the end…

you might consider supporting The Consciousness of Sheep.  There are seven ways in which you could help me continue my work.  First – and easiest by far – please share and like this article on social media.  Second follow my page on Facebook.  Third follow my channel on YouTube.  Fourth, sign up for my monthly e-mail digest to ensure you do not miss my posts, and to stay up to date with news about Energy, Environment and Economy more broadly.  Fifth, if you enjoy reading my work and feel able, please leave a tip. Sixth, buy one or more of my publications. Seventh, support me on Patreon.

Check Also

Energy myths

In the real world, any recovery of new oil and gas is a decades-long, multi-billion-pound endeavour