Deepseek’s Silver Lining?; Indonesia Unleashes Ocean Sand Miners; Book News
and more about the human and environmental costs of renewable energy and digital technology—and how we can do better.
Deepseek’s Silver Lining?
The tech world is panicking over Deepseek, the Chinese AI company that has apparently produced a chatbot that’s as good as ChatGPT and other well-known models, but that runs on far fewer computer chips, making it much cheaper. This may all be terrible news for stock traders, tech companies, and America’s geopolitical position. But if Deepseek really does run as efficiently as advertised, it could have at least a small upside for the Earth. If it’s really possible to train and run AI systems with fewer chips and less energy than heretofore believed, that means we might be able to reduce both the technology’s carbon emissions and appetite for critical metals.
As I explained (in more detail) back in October:
AI mainly requires metals for two purposes: for the computer chips and associated machinery on which its algorithms run, and, much more significantly, for the energy that powers those chips and machines.
The chips in question often include small amounts of gallium and germanium. (China dominates global supplies of both, and throttled exports of them to the US in December. Another reason it would be good to lessen our reliance on large numbers of chips.) Each chip holds only a small amount of each metal, but the data centers that house them can contain tens of thousands of chips.
Lots of other metals are required. Tin is used to solder chips into place on circuit boards. Copper is used to network chips together, and carry power to them and the other machines that underlie them. A single data center can require more than 2,000 tons of copper for its internal workings. Other equipment in data centers requires tantalum, rare earths, and other minerals. And much more is coming. Amazon, Meta and Microsoft have declared they intend to collectively sink some $220 billion into chips, data centers and other AI projects in 2025 alone. Cheered on by President Trump, several other companies recently pledged to invest up to $500 billion in American AI infrastructure.
Significant as all that is, AI consumes far more resources indirectly, in the form of the energy infrastructure required to power it.
A single data center complex can consume as much power as a small city. As a result, according to Bloomberg, “Microsoft’s emissions were 30% higher last year compared to 2020 while Google’s emissions were up 48% compared to 2019, in both cases largely due to AI.” The energy demand from AI could increase tenfold by 2026, according to the Harvard Business Review, outstripping the annual electricity consumption of Belgium. At this point, most of that electricity comes from carbon-spewing fossil fuel power plants. In the US and elsewhere, coal plants that had been slated for shutdown are being kept online to meet the voracious appetite of data centers.
Carrying all of that electricity will require oodles of new copper cables. Supplying power to the world’s data centers could require 2.6 million tons of copper by 2030.
All of those chips running in all of those data centers eating up all that power could easily add up to environmental calamity. But we might be able to at least blunt the damage if ChatGPT, Gemini and the rest of the pack can be made to run more efficiently. If Deepseek can do it, presumably OpenAI, Google and all the rest can as well. Now, they may have to.
It all puts me in mind of how fuel-efficient Japanese cars snatched market share from US automakers in the wake of the 1970s oil crisis. Before that, American carmakers were happy to crank out gigantic, gas-guzzling vehicles, because why not? Gas was cheap. The issue wasn’t that more efficient technology didn’t exist. It was that American carmakers had no incentive to get more efficient. Ditto today’s big AI companies. As long as they’re raking in billions of investment dollars to support their gigantic resource-gobbling systems, why bother trying to make them more efficient? Deepseek might have given the AI world a much-needed kick in the ass.
This Week in Sand: Indonesia Unleashes Ocean Miners
My first book, The World in a Grain, was about the deadly global struggle for sand. That’s right, sand. It’s a lot like critical metals: a little noticed but absolutely crucial component of the global economy, the pursuit of which is wreaking havoc around the world. We need sand to make concrete, glass, and silicon chips—and incredibly, we are starting to run out of the stuff. So much so that we’re tearing up beaches and riverbeds all over the world to get it, and in some places criminal gangs are stealing and even killing for it. Hundreds of people have been murdered over sand in recent years.
So this news caught my eye: In Indonesia back in the early 2000s, intensive ocean sand mining completely destroyed 26 islands. Most of the sand was sold to Singapore, which used it to artificially build up its own territory. The environmental damage was so severe that Indonesia banned exports of sand. But now, Djakarta has changed course, and industrial-scale sand mining for export is set to begin again soon, to the horror of environmental groups, politicians, and fisher folk whose livelihoods are threatened.
Book News
Thanks to everyone who packed Upstart and Crow bookstore for my conversation with fellow author Chris Pollon last week!
I also did a cross-Canada radio blitz, appearing on CBC radio shows from Nova Scotia to Vancouver Island. For a taste, here’s me discussing Power Metal on Vancouver’s North by Northwest.
More News Worth Knowing
🏭 Rise of the “Clean” Energy Megaproject
🏆 Ten Prize-Winning E-Scrap Recycling Projects
⛏️ Congo Rebels’ Gains Could Boost Illegal Mineral Trade
💰 India Will Invest Nearly $2 Billion into Critical Minerals