Tl;dr: The clouds should use their powers to help power the cloud. Atomically.
CAPEX obsessives like myself have focused on the cloud’s vast agglomeration of compute, storage, and networking.
But there is another, less appreciated infrastructure dimension: the cloud is in many ways as much about moving electrons as bits. Without a considerable and constant supply of electricity to power all those building blocks, the cloud quickly dissipates.
While it has a supersized electricity appetite, the cloud is not nearly as voracious as some had feared (hyperbolic power consumption predictions have a dismal track record). And the cloud’s efficiency looks even better compared to the same tasks performed artisanally (on-premises) or – for completeness – IRL.
Proximity to reliable and low-cost power is critical to data center siting decisions, as electricity is the biggest variable cost. Power distribution (and cooling) is a if not the major design constraint inside a data center. And those data centers are surrounded by phalanxes of backup diesel generators in the event of currentus interruptus.
The hyperclouds are acutely aware of their power consumption. They buy a lot of charismatic renewable energy (and spend more money making sure we know about those purchases).
But the not-so-secret dirty secret is the cloud isn’t powered by just solar panels and windmills. Those are intermittent power sources and the cloud doesn’t shut down at sunset or when the wind doesn’t blow (nor for the winter in the case of a new region in Alberta intimated to run on solar power).
All those renewable purchases feed into the broader electric grid, but the cloud that runs non-stop needs reliable baseload power from that grid. Depending on location, that baseload power is a buffet of coal, natural gas, oil, hydro and/or nuclear energy.
A Brief Road Trip
Let’s briefly transform into a travel blog to visit a few prominent data center locales and sample some local energy buffets:
- Washington State: the top destination here (as in so many other ways), and home of Cloud City (Seattle), the cloud’s geographic and spiritual center, generates 8% of its power from solar and wind (only 0.04% of that is from solar, and, admittedly, that seems high), 12% from natural gas (the cleanest hydrocarbon), 4% from coal (the dirtiest hydrocarbon) and a whopping 66% from hydro power (what is bad for solar is good for hydro!). (And that might surpass even my own record for parentheticals).
- Virginia: home of various us-east-1 cloud conflagrations — both real and of the configuration variety — as well as GovClouds galore, generates 1% of its power from solar and wind, 61% from gas and 4% from coal. The cloud cooks with gas! But this cloud epicenter may be facing a power pinch.
- Ohio: the us-east region AWS would prefer you to use, generates 2% of its power from solar and wind, 43% from gas and a whopping 37% from coal. Go Buckeyes! (Little known fact: a buckeye is evidently another name for a briquette of coal).
Actual generation from both solar and wind underperform relative to capacity, underscoring their intermittent nature (which often manifests at inconvenient times). So, despite the picturesque photos accompanying the self-congratulatory press releases, it isn’t all rainbows, solar panels, and windmills.
The Hard Thing About Energy Things
The hyperclouds’ decarbonization efforts have evolved. Initially they focused on offsets (carbon indulgences basically), followed by procuring enough renewable energy in aggregate to match their total consumption. Google has led the way in the shift to renewable data center power, followed by Microsoft, with Amazon bestirring itself more recently.
Actually reducing emissions is proving harder than the press releases promise.
The next envisioned step is “24×7 renewables”, a promise to procure sufficient renewable power to power data centers all the time (but still mediated by the grid).
Because of the intermittent nature of renewables, the costs to converge on 100% renewables start to go vertically asymptotic, as it requires more transmission capacity across a broader area, an overbuild of sufficiently distributed generation capacity and/or vast investments in electrical storage (a capability that simply doesn’t exist today economically or at scale).
But perhaps there are other appetizing options for our energy buffet.
Going Nuclear
Preceding even the Putin-precipitated reappraisal of global energy strategy, nuclear energy was having a moment, even amongst climate activists. Baseload reliability, scalability, carbon-free emissions, and reduced dependence on odious and unreliable foreign suppliers are easy to appreciate.
Nuclear is vastly safer than perceived and comparatively safer than most other forms of energy. And nuclear waste is very manageable. (I am not the first to blame The Simpsons for being anti-nuclear propagandists and await the full investigation into whether they were funded by Putin along with other anti-energy lobbies).
Even the EU (and you can’t spell Butlerian Jihad without E and U), has come around on nuclear power (except of course in Germany, where they would rather burn Ukrainians the dirtiest of dirty coal and Russian hydrocarbons than keep their functional nuclear plants operating, while prevaricating about why).
MAGMAnimity
Which brings us (finally!) to the point of this post: the cloud vendors need to champion nuclear power.
The hyperclouds (Amazon, Google, Microsoft) should use their myriad powers to accelerate the budding nuclear renaissance. Apple and Meta should join the effort, both because they also have large scale, global data center operations even if they don’t offer cloud computing services, and because their participation lets me refer to our new nuclear advocates collectively as MAGMA (Meta, Amazon, Google, Microsoft, Apple – like Larry Page, we’re done with the Alphabet pretense).
MAGMA have immense power of the purse, with around $400 billion in cash and short-term investments among them, and a demonstrated willingness to make big, long-term investments.
MAGMA are practiced at the power of public persuasion (and have legions of lobbyists for political pleading). They frame what the future looks like every day, and are more trusted than government or the media (ok, we’ll ask Meta to quietly don a headset and sit out any PR efforts).
MAGMA have a proven power to design, deploy, and operate complex infrastructure at scale. They spent over $150 billion on CAPEX last year. One of the challenges with nuclear power is building capacity efficiently and economically. The status quo approach to building nuclear plants is trending in the wrong direction (seemingly a reverse learning curve), and MAGMA can apply engineering chops and cash to that conundrum.
Last but not least, MAGMA have a superpower that has been remarkably absent in energy strategy: they can do math. Math, even just basic arithmetic, turns out to be critical to planning large-scale, long-lived, long lead-time infrastructure.
The much-touted transition to clean energy is looking more and more like a haphazard and unplanned transition to scarce and expensive energy. Energy innumeracy has been particularly acute in California, New England, and Europe, with Germany performing the worst on this existential test. They all face soaring energy costs (even before this year), risk of blackouts and rationing, and varying levels of vulnerability to Vladimir.
The resurgence of coal consumption in developed countries, even before the Ukraine invasion, is the most damning indictment of the current “transition”.
Treating natural gas as if it were as bad as coal or oil was an epic mistake, neck and neck with shutting down fully functional carbon-free nuclear plants. And then there are retro-medieval absurdities like burning wood, because, hey, it is renewable. Noble intentions and pieties that fit on a bumper sticker are not enough – energy demands realistic planning and execution.
Energy is fundamental to civilization and an advanced civilization requires abundant energy. Abundant energy can be clean and abundant energy lets us accelerate the move to clean energy, as well as do things like prodigious domestic polysilicon production for solar panels (it comes from slave labor-friendly Xinjiang today), desalination, moving heavy industry to hydrogen, and the general electrification of everything.
Being anti-carbon has been conflated with being anti-energy, and vice versa. That must change. If you’re really for energy scarcity across the board, at least be honest that your vision for the future is one of Malthusian austerity (and let’s see how much popular support that garners).
We need abundant, heterogeneous, and reliable energy. The cloud needs it, along with the rest of civilization.
Not Because It Is Easy, But Because It Is Hard
MAGMA can do much more to support their own energy needs, and the wider world’s. That means not just tweeting platitudes, writing easy checks for charismatic albeit intermittent renewables, and indulging in Patagonia ad envy, but doing some hard things. Specifically:
- Stand Up for Nuclear Power – MAGMA should take a strong advocacy position for greatly expanding our use of abundant and carbon-free nuclear energy with the public, their customers, politicians, and policymakers. They need to support all clean energy, not just fashionable renewables.
- Stop the Shut Down of Functional Nuclear Plants – MAGMA should apply all their financial, political, and persuasive muscle to keep existing functional nuclear plants open and reopen recently closed plants. The world is fortunately waking up to the fact that closing nuclear plants leaves us with a dirtier, less reliable, less abundant, and despotically dependent energy system. MAGMA should keep the pressure on in California, New England, New York, Belgium, Canada, Japan, Korea, the Netherlands and anywhere else there are recently shuttered plants or more plants scheduled for shutdown. Germany, which this year has had to repudiate every strategic decision of the last three decades except closing their nuclear plants, is hinting they might be open to possibly considering not closing down their last three plants. Seemingly it is not a no-brainer in the face of an impending cold winter and collapse of their economic model.
- Build New Nuclear Plants – The best time to build new nuclear plants was 20 years ago. The second-best time is now. MAGMA should help finance and accelerate the building of new plants (and they’re welcome to collaborate – northern Virginia might be a great place to start). They can make procurement commitments as they do for solar and wind, as well as invest directly in new plants in conjunction with utilities or by themselves (the utilities are both too timid and playing with too short a stack of chips to really take the lead here – the parent companies of the top four US utilities annual CAPEX spend is about what AWS alone spends). The new generation of small modular reactors look especially attractive to add to the grid or even potentially deploy alongside data centers (let the grid be the source of backup power). MAGMA can also push on licensing and regulatory reform, which today is either designed to ensure nothing happens or is so poorly designed that nothing happens.
The cloud companies tell us climate change is “catastrophic”, “the greatest challenge”, and “the world must act now”. Yet MAGMA’s efforts pale relative to both their statements and their breadth of credibility, resources, and skills. They can do more than toss some more renewables into a flawed energy system and spray paint themselves green.
If not MAGMA, who? The political class are unable to rise above performative Manicheanism (plus math is hardly their forte). The utilities’ skills at depreciation schedules and appeasing utility commissions don’t lend themselves to bold but necessary shifts in strategy.
And it turns out championing nuclear isn’t even that bold a move. Public opinion is already there. Even in Germany (except for the bewildering political class).
“Do the other things, not because they are easy, but because they are hard” – JFK in his “We Choose to Go to the Moon” speech.