The Load

Monday 29 April 2024

Well known photo used on the cover of Led Zeppelin IV, originally sourced from “Reminiscences of a visit to Shaftesbury. Whitsuntide 1892.”

Cold Eye Earth won’t be joining the latest wave of worry in energy circles: that AI is going to push global power demand to unsustainable levels. The fundamental proposition of energy transition is to bring everything over to the efficient platform of electricity. So, if you’re worrying about AI and crypto demand why aren’t you worrying also about transportation, air-conditioning, and industrial demand because all that too is coming to the grid. The proper stance for transitionists is to simply say, bring it. Indeed, a more productive way to worry is to fret that we won’t build enough wind, solar, storage, and nuclear power to handle all the load growth that’s coming anyway, and as a result, natural gas will continue to exploit the shortfall. Worry about natural gas, not Microsoft’s GPU demand.

The world currently uses over 29,000 terawatt hours (TWh) of electricity each year, and growth is going to take that figure to at least 37,000 TWh by 2030. We already face a challenge covering that growth with wind, solar, and storage so it’s understandable to be concerned. But there’s simply nothing special—or new—about data center demand growth. According to a report this year from the IEA, data center demand may double from 2022 to 2026. Scary? Not really. Demand doubled over the previous four years too.

Chart from the IEA Electricity 2024.

Again, your baseline expectation should be that global electricity growth is going to grow faster than ever before, now that we propose to call upon the grid heavily as the primary platform for economic growth. AI, crypto, and data center growth may add another 400 TWh by 2026. So what? Total global demand is going to grow by by over 25% off a huge base, advancing by at least 8,000 TWh from 2022 to 2030. And the more successful we are at rapidly moving cars and industrial processes to the grid, the faster we will need to deploy new clean power capacity to handle it all. Last year, the world put 14 million new EV on the road, representing new power demand of 42 TWh. (14 million X 3000 kWh average annual consumption per global vehicle / 1,000,000,000 = 42 TWh). But we’re still early in the EV adoption game and the world has over 1.5 billion vehicles, the vast majority which are still running on petrol.

Data center growth simply doesn’t scale as a notable problem compared to the challenge we face at the global level. Instead, this is a story that is a very big deal at the corporate scale, with enormous implications for the economics of providing AI services to the public, and the required capex to be undertaken by Meta, Google, Microsoft and Amazon. When the stock of Meta was rocked this week on the announcement that its AI infrastructure spending was set to soar, you can begin to see how the AI energy-growth story slips out from its appropriate lane in the business world and winds up in the wrong place. Put another way, humans are wired to be sensitive to small scale stories which are often more relatable. Hardly anyone knows the world is on course to add 8,000 new TWh of demand to the year 2030, so going from over 400 to 800 TWh globally to support the cloud sounds like alot, to most people.

When you read breathless stories about US data center demand growth, just remember that this is a 20 year old story, with several big waves already behind us. And finally, it must be remembered that US based technology giants are themselves prime movers in the deployment of wind and solar, with most of these same companies having contracted to build their own utility-scale clean electricity projects already, often far in advance of states and municipalities.

In each of the previous transitions, observers became alarmed at the new dependency being formed on coal, and then on oil. If we want to make sure we compete with fossil fuels effectively, we will need to not only meet all the new demand coming onto the grid, and then some, but we shouldn’t complain about it either. We shouldn’t gate-keep who or what is coming to the grid. We should instead welcome the demand, as demand is required to keep driving the cost curve downward. Build. Build. Build. Deploy. Deploy. Deploy.

US natural gas demand growth has been so strong, for so long, that it’s about to overtake oil as the top energy source used by the United States. Over the past year, Cold Eye Earth has devoted at least a quarter of its content to the natural gas problem. A few points. Natural gas has not only been growing strongly in the US for more than a decade, but it continues to grow at a fast pace. Yes, natural gas along with wind and solar helped kill coal. But at some point in time, that “win” fades, and fades rapidly, and you find yourself with a whole new problem of emissions growth. And apart from natgas growth in industry, we’re still not suppressing natgas growth even in the power sector because the replacement of lost coal capacity is still not entirely met by renewables. Result: natural gas consumption reached 33.6 quadrillion btu last year compared to oil’s 35.4 quadrillion btu (quads). Notice also the big leap that started in 2018, when natural gas began to lift from 28 quads.

The chart is interactive on desktop devices at the Cold Eye Earth website: mouse over it for data.