As part of our ongoing investigative series, “Hidden Wonder Valley” into Kevin O’Leary’s proposed $70 billion AI data centre in northwestern Alberta, The Energy Mix reporter Jody MacPherson has been talking with a wide range of experts. 

Jody spoke with Amory Lovins, RMI co-founder and world renowned energy analyst, by telephone on September 16 to discuss Alberta’s 2% levy on data centre projects seeking to connect to the provincial electricity grid.

The levy was announced after the Alberta Electric System Operator (AESO) said  in June that it “cannot possibly connect” all the proposed data centre projects, at least in the short term. AESO said it can handle up to 1.2 gigawatts (GW) for new large-load data centres but the 29 projects waiting in the wings were looking for more than 16 GW. Projects bringing their own power are exempt from the levy.

This interview has been edited for length and clarity.

The Alberta government recently announced a levy on new data centres that want to connect to the grid, but it will not apply to off-grid projects. What are the issues when you go down this path to attract more data centres?

Well, I assume, like Texas, there are strong commercial forces wanting to build more gas-fired power plants so they can sell more gas. Of course, if you take economics seriously, you have to look at the whole competitive landscape. There are ways to do AI with much less electricity. 

For example, a Duke University study showed that a tiny bit of flexibility in running some of your AI activities goes a long way in freeing up existing utility assets, which would be suitable for all but perhaps tens of hours per year. In North Carolina, if you crank down your load by half of 1% over the year, during a small number of hours when the grid is under stress, that would free up enough capacity to run all the proposed US data centres from existing assets.

Another thing you’re betting against is the speculative nature of data centre projections, not due to the fundamental uncertainties about AI business models, although that’s a big issue, because currently the AI companies are earning in the order of US$50 billion a year, but they’re going to need many, many times that to repay the investments they’re making.

It’s not clear whether or where they will get that extra revenue, but I’m talking about the tendency for developers who might have a piece of a deal to behave as if they had a full deal in order to try to attract capital and get something built, hoping somebody will need it and pay for it.

And that’s such a common speculative pattern, at least in the U.S., that many experts believe most of the proposed U.S. data centres will not get built. Then, of course, once they’re built, they would need to keep running for several decades to repay the power supply investment, let alone the financial investment. So that’s an awful lot of stuff to be betting against.

And then there’s another factor that makes it an even worse bet, and that is if you were fully confident that a particular AI data centre will get built and will thrive and will pay its electric bills for decades, you have to compete with all the other ways to provide that electricity.

Generally, the cheapest one is to save electricity somebody else is using inefficiently—what I call [pdf] “megawatts for efficient use” or “flexawatts for timely use.” If you had exhausted all the cheaper megawatt and flexawatt opportunities, well then you have to compete with resources like wind and solar, which unlike gas, have a constant price warranty once you build them because there’s no fuel and therefore no financial risk from fuel price.

There’s also a lot less to go wrong. It’s a more reliable source, it builds much faster and therefore has less financial risk. The price in the U.S. of gas-fired power plants has doubled to tripled in the past couple of years. Typically you don’t have that kind of risk in solar and wind.

This May, China added 93 GW of solar power and 26 GW of wind power. That means they were adding nearly 4 GW each day, or 4 billion watts of solar and wind capacity. To put it another way, in the first half of this year the world added more direct current (DC) gigawatts of solar power alone than the world has alternating current (AC) gigawatts operating nuclear power over decades.

In Alberta, one of the complicating factors is that the government is making renewables almost impossible to get built. I read a report yesterday that estimates [pdf] almost 40% of the land in Alberta is not allowed to have any solar, or wind because of “pristine viewscapes.” What are the risks we’re facing by not allowing renewables to power data centres?

That sounds like what President Trump is trying to do. One risk is of course that they will kill the data centre industry they’re hoping to create because data centres depend on reliable, affordable, and above all, fast power supply to run on.

This is what’s showing up now in the U.S. where the fossil fuel industries, which have very strong influence politically at all levels, are trying to kill renewable power. The result is that the power supplies available from gas, or even slower costlier nuclear, to run new data centres will be too little and too late to compete with other places.

It also means you’re getting locked into perpetual fossil fuel generation which will become unsustainable in all respects in the coming decades. 

It doesn’t sound like that free market approach appeals to the current government of Alberta. I would respectfully suggest applying conservative economic principles because there’s a lot of history showing that gives better results than picking winners and losers. 

What you described sounds like what the Texas legislature tried to do earlier this year. They are very strongly influenced by the natural gas industry and they passed new fees, taxes, and severe restrictions. If you built a wind or solar megawatt, you had to build a gas megawatt to go with it, as if it required that sort of backup, which it doesn’t. There are ways to keep the grid balanced and stable as it becomes renewable and they’re all cheaper than fossil fuel backup. 

But anyway, the threat didn’t happen. It was pulled off the floor by the Republican leadership in Texas because they started to hear from the many business leaders in Texas who are making a great deal of money from having made Texas the national leader in wind power and now in installing solar power as well. They pulled ahead of California, which is the world’s fourth biggest economy. 

They did this not because of environmental goals, but because Texans tend to be good at building things and making money, like Albertans. And the renewables entrepreneurs managed to defend their industry from this extraordinary attack to try to basically outlaw their successful industry in order to give more business to the defeated competitor.

Let me give you an example of how quickly technology is moving. I went to Sparks, Nevada, which is less than an hour’s drive east of Reno, to see something developed by a new firm called Redwood Materials, well established as a world leader in battery recycling, founded by one of the original Tesla car pioneers, JB Straubel.

They had many hectares of recyclable batteries of all shapes and sizes, including a lot of electric car batteries from cars that aged out or crashed or had a defect. The stuff was sitting out in the desert and it was being profitably recycled, but the profits weren’t as juicy as they wanted just now because the prices for things like lithium, nickel, cobalt, manganese, and copper graphite had basically crashed, because so many people believed the panic fermented by the fossil fuel industries about how we’re going to run out of these critical materials. That turned out not to be true.

So although Redwood could still make money, they started to think about getting another 5-8 gigawatt hours of batteries coming in every year. So what they did was lay 20 megawatts (MW) of photovoltaic solar modules on the ground. They just graded the desert. They laid down the panels. No concrete, no racks. Then they wired them to about 800 used electric car batteries and hooked them all together with software and power electronics that made them into a single giant battery. Even though the individual batteries had many different sizes, chemistries, voltages and so on, they blended that all together electronically into a 62 MW/hour battery.

The result was 12 output AC MW of 100% solar power, cheaper than the eight US cent per kilowatt hour grid power. Soon four modular data centres were built right on the site. And all built in four months.

Whoa, four months?

Yeah. Now it takes about 18 to 30 months to build a data centre, if all goes well. So, you don’t need to guess or debate whether a particular data centre will get built and will flourish in order to make a big bet on building a giant power plant to run it. Instead you can wait until the data centre is mostly built and then build something like what Redwood did. They’re now designing versions tenfold bigger. You could probably build it by competitive bid. It doesn’t need any transmission. Everything is right on the same site. It may not need any approvals. And again, it’s cheaper and more reliable.

We’re actually finding that modern renewables win not only on cost and speed, but also on reliability and resilience over fossil fuel generation. 

But even more interesting is the state of South Australia, which is currently three quarters run on sun and wind. They have no hydro and they have nothing that would be called a baseload power station. At the moment they are running within the state normally only a single gas plant, combined cycle, amounting to 2.4% of their generation. And in certain circumstances they might have two such units, so 5% of their generation from gas.

They intend by the end of 2027, just over two years from now, to be 100% running on solar and wind and that will be not only reliable but also so cheap that they now have 37 large companies wanting to come build factories there. We have more and more examples that renewables can run large, difficult, heavy industries very reliably with superior economics. And anybody who doesn’t realize that and discounts them as a potential way to run Alberta data centres is not paying attention. 

In terms of the water usage for cooling, is there progress happening on that? 

Well, wind and solar power don’t use any water. With the minor exception that in the Sparks solar microgrid I described, Roomba-like robots come out of their little houses in the night and wander around scrubbing desert dust off the solar panels, and then the water is recovered and reused. But that’s a trivial use compared to what gets evaporated by cooling towers in the front of the power plant.

So yes, if you, if you are in a water-stressed area, renewables have that further advantage. They are safe, they emit nothing. They are extremely reliable and resilient. They vary in output, but probably more predictably than electricity demand. Oh, and by the way, they’re portable. If the folks at the Redwood plant decided they would be more advantageous to move their solar microgrid somewhere else, they just load it on the truck, take it somewhere else, set it up again. There’s no concrete and there’s nothing in the ground. 

It seems that technology improvements are happening so quickly that the way that we do it today might be different in six months time.

Well, it’s true that the renewables and especially the batteries keep getting relentlessly cheaper. A year ago people were projecting that good stabilizing batteries like the giant ones used in about 50 places in Australia now would cost maybe US$150 per kilowatt hour of capacity. The latest Chinese bid was $51 per kilowatt hour a couple of months ago and still falling. It’s a little like buying a computer. You know it’s going to get better and cheaper if you wait. But if you need it now and there’s a good business and operational case for it, then you buy it now. If you can avoid buying a gas-fired power plant which will, you know, probably keep getting costlier over time, that’s immediate. And more importantly, by providing prompt power in months, not in years, you have a much better prospect that your data centre will actually get in early enough to be competitive and successful.

Thank you, you’ve answered all my questions.

You’re welcome. And I’ll add one little thing I forgot to say. South Australia is running completely on solar and wind, which are variable. The reason that works is that the grid is stabilized by giant batteries, including the first one built by Tesla. They actually turned it on a few days before it was supposed to start. Immediately one of the big coal fired power plants a thousand kilometres away failed, and the giant battery just turned on, leapt into action and solved the problem almost instantly.

It’s now turning out that the batteries are valuable not only for storing energy and shaving peak loads, but also for providing voltage stability, frequency stability, and keeping the grid stable when things go wrong. They do that much faster and more precisely than you can do with rotating machinery that we used to think was necessary to provide physical inertia. We now do it with virtual inertia. That’s the best news.