alt Hacker NewsShould this apply only to AI compute facilities or all industries?
I have no real evidence for this but I think a general principle of good governance is that the policy is agnostic to the industries which it affects and robust under any set of conditions, so for instance, the price that electric utilities charge should be enough to pay for expansions, either prospectively or retroactively (via loans secured before the construction, taken out against projected future revenue), rather than subsidizing the expansion with taxes collected by industries that won't be using the electricity, or exercising discretionary control over which types of industrial expansion are accommodated through investments in expanding the energy infrastructure.
Huge expansions of energy production have occurred in the past to meet growing demand for energy from then-emerging industries. And I'm not aware of any past campaigns to stop the various industries that needed energy on account of the allegation that their increase in energy usage would burden other sectors of society. I have concern that the treatment of AI compute facilities is based on public opinion rather than rule of law and consistent protection of private property rights.
Replies
I think it should apply to all industries. But for many (or most?) countries where power generation is privatized in the first place, this shouldn't be a problem that we need to fix. The trickier question might be about the grid, which in most cases is publicly funded.
AI compute is a unique facility: can be placed anywhere, takes in mainly energy, does not produce the same volume or composition of waste, has noise pollution, and needs minimal access.
I am not a regulator but that feels like something that needs new rules
So, the problem in this case is that there's no gain for the country building the datacentre.
If you're in Finland and there are two possible uses for electricity production, let's say, either a steel plant or a datacentre. The steel plant will employ a bunch of people locally. A datacentre will employ a bunch of people in California.
So if you are to build a datacentre, the deal must necessarily be that the R&D for the models that are to run on it must happen locally. Otherwise there's no reason to give them the allocation over the steel plant.
The principle prevents governments from taking actual realities of any specific scenario into account, so it seams bad.
> I think a general principle of good governance is that the policy is agnostic to the industries which it affects and robust under any set of conditions
The trouble with this is that people frequently don't like the consequences of that.
For example, a lot of the costs of a grid connection are non-linear if not totally independent of the amount of power you use. When a tree falls on a power line and a crew has to show up and install a new pole, the cost of that has nothing to do with the average percent utilization of those wires. Likewise, a line that can deliver 100MW to a single customer costs more than one that can handle 100kW, but it doesn't cost anywhere near 1000 times as much or consume 1000 times more land etc.
The result of this is that the "works under any set of conditions" pricing for a grid connection would be a flat monthly fee for the capacity of the line totally independent of how much you use, and a line with 1000 times the capacity would only cost e.g. 10 times as much. But then a typical household electric bill would have a fixed charge of ~$100/mo for a grid connection regardless of how little electricity you use.
Likewise, if you have a high proportion of renewable generation in the grid then the price per kWh during daylight hours when there are clear skies would be essentially zero, but the price on a cold still night in winter would be oppressively high and anyone who failed to turn off their electric heat on that specific day would end up with a $5000 electric bill.
So in most places what happens instead is that the power company charges a fee per kWh which is used to pay line crews and other expenses that don't actually depend on usage and then have a much lower fixed connection charge. And charge a certain amount per kWh during days when supply is sufficient and then use the money to subsidize consumption during shortfalls, even though that actually raises total costs, because otherwise people would scream about prices hitting $500/kWh for a few hours every year or two.
But once you're doing those things, a customer that wants megawatts of power is reasonably going to object to that rate structure, because the few percent inefficiency those things induce is millions of dollars in their case. At which point people start complaining about subsidies, even though not doing those things would be the absence of subsidies.