In this episode of PING, APNIC Chief Scientist Geoff Huston reflects on the tremendous economic consequences of the surge in capital investment in Artificial Intelligence (AI) and the associated Data Center (DC) build out. This not involves so much money, it represents a tenth or more of the US annual Gross Domestic Product (GDP) and has significance outside of the application of AI to problems in the world. As technology consumers we’re already used to the idea the cost of basic inputs like RAM, storage devices, the GPU cards which we need to run high speed gaming computers are both in short supply and gone up in price significantly. But the economic effects of the surge in AI go beyond the IT sector. They affect all of us fundamentally. Having a speculative activity like this represent so much of any economies activity means that as it ebbs and flows, or even worse crashes, the same amount of economic activity is removed from other productive investment. Some of the AI money is hypothetical, set into the future on the premise the investments being made now play out, but some of it therefore is highly risky, money which won’t actually exist when the time comes, but the debt side, the cost of raising all this money, spending all this money and time and effort building out the DC. That debt will exist, and will either be recovered in a fire sale of assets, or be distributed around as pain in the pockets of investors who put their money into the proposal. A problem for almost all of us, is that the investment decisions here have been taken out of our hands: Although our apetite for risk may vary, the decision around investment is made mostly on our behalf, by the banks, the investment companies, the “institutional investors” which may seem remote, but when you consider the Australian Superannuation sector represents $4.5 trillion Australian Dollars, and is the fourth largest investment pool in the world, for a nation of only 27 million people, our exposure to this risk is a lot larger than you might think. The same is true for the pension funds of Canada, of the United Kingdom, the European Union, Asia and especially the investment industry of the USA, because the AI companies have joined or are on the brink of joining the pool of investment which is held to be “normal” and a required part of a balanced portfolio. To build and operate a modern AI focussed DC demands a huge investment in construction, in power engineering, in air conditioning and related systems. This investment is so potentially profitable that it can command all the inputs, all the labour required to build it, and afford to pay ahead of the rates normal construction sector projects would expect. It’s contributing directly to a supply chain shortage across the civil engineering sector, and in unrelated construction activity like housing. The diversion of electricity supply to power a DC means that electricity prices for domestic consumers are being pushed up, and the social consequences of where these DC are built effects their immediate surroundings in terms of the amenity and viability as a place to live and work. As Geoff points out, these systems are so big that in the USA, they cannot co-exist in the same state as a competitive sector: They have to be sited independently without a comparable DC investment, lest they overburden the utllity supply systems. This isn’t something the worldwide economy sat down and decided to plan out, it’s emerged into the world with probably unstoppable momentum. The problem exists everywhere, all at once, and the problem is getting larger, as the investor side gets more concerned of their risk of exposure as the cracks in the model emerge. Who “wins and loses” here goes beyond the nameplate on the DC, or the name of the AI company. We’re all standing to suffer economic shocks, as this AI craze works its way through to the inevitable other side of the bubble.
