Predictions in energy and computing

Is it a good idea to make predictions about future energy consumption?

Is it a good idea to make predictions about future energy consumption? Or data center energy? Or efficiency improvements? Or anything that is more than a few years into the future?

In startups, there’s a common understanding that financial projections more than 12 months out are basically worthless. That doesn’t mean you can’t model costs – those are under your control – but when it comes to revenue or profit figures, customer numbers, or even more tenuous metrics like clicks, downloads, installs, those are more uncertain the further into the future they go.

Even mature public businesses typically only issue earnings guidance for the next quarter. Maybe it’s because predicting the future is hard?

Coming back to environmental predictions, one of the key texts I had to read at the beginning of my Environmental Technology MSc was the 1972 publication, The Limits to Growth by Matthews et al. This uses simulations and models to predict the future based on the trends of the time, with the famous prediction:

The basic behavior mode of the world system is exponential growth of population and capital, followed by collapse. As we have shown in the model runs presented here, this behavior mode occurs if we assume no change in the present system or if we assume any number of technological changes in the system.

The Limits to Growth, Matthews et al.

I found this interesting because, of course, my reading it many decades after it was published obviously meant that the world had not ended! Technological change has indeed been able to make a huge difference to the outcomes of “business as usual”, not least the invention of widespread computer technology.

The Limits to Growth is not even the most famous of these types of fundamental errors of prediction. Malthusianism must surely take that title, but they are all in the same category. It’s worth comparing to the IPCC reports and UK Government Net Zero plans because The Limits to Growth said that technology change would have no impact on stopping the inevitable collapse, whereas modern reports say that technology change will be crucial to avoiding those problems.

That is not to say that we can just continue to do as we have in the past. We can’t just keep burning fossil fuels, or failing to measure water consumption in areas of water stress, or ignoring the poor working conditions where key minerals are being mined. But making dire predictions so far into the future that they lack credibility is not helpful.

Extrapolation only works if all variables remain constant. Maybe it’s possible to adjust for certain changes here and there, but it’s difficult enough to model the past behavior of complex systems. Making predictions more than a year or two into the future is almost impossible, even when you don’t have unexpected events like a pandemic to deal with.

Will Moore’s Law finally end? We thought that was happening, then the industry started shifting to a completely different CPU architecture. ARM chips on the desktop and in the data center are growing rapidly. Offloading software development to the cloud will improve efficiency. Centralization is good for taking advantage of economies of scale, but moving workloads closer to users will also have benefits. Processing data on a single CPU might be reaching the end of the innovation cycle, but we’ve only just started to rearchitect and split workloads to specialist environments. Can you predict how that is going to play out?

This presents a challenge for those working on large infrastructure projects, particularly energy systems that have lead times measured in years, such as building new power plants or constructing transmission lines. Organizations like National Grid produce Future Energy Scenarios to help them plan how the energy grid will need to change over the coming years. Building out too much capacity means wasted resources, but failing to deal with unexpected demand might be worse – delayed projects, brownouts, or system failures.

There is no easy answer, so I try to remain skeptical of predictions. When I see a new estimate, I first think about whether the prediction fits the general direction I was expecting. Ranges are helpful because an exact number is rarely needed, so long as it is within a reasonable range. Anything that is an order of magnitude different from where we are today, or is going in a completely different direction than expected, is where I dig into the details.

The near future (a few years) tends to be quite similar to today. The far future tends to be very different, so is impossible (and pointless) to predict. It’s the mid-range that causes all the problems.