Epistemic status: some errors pointed out and since disentangled in the comments. LLM-use for much of the research. I regret having published this without going through academics online first, and will do so now, and edit the post accordingly. There is surprisingly little research on solar storms' effects on electric grids and solutions to those out there.

Most of civilization's electricity is generated far off-site from where it's delivered. This is because you don't want to be running and refueling coal/gas/nuclear plants inside cities, hydraulic/wind power can't be moved, and solar panels are cheaper to install on flat desert terrain than on cities: 

So in practice this means running power over hundreds or even thousands of kilometers. E.g. here are the Chinese long-distance lines: 

American long-distance lines: 

These are simplified maps meant to illustrate how insanely long power lines get. The true shape of solar storm vulnerability looks like a spiderweb overlayed on population density (see below), which you can visualize on this website. 

The fact that civilization finds it economical to generate its electricity hundreds or thousands of kilometers away from its population centers is rather mind-blowing given the infrastructure involved. For example, the Tucuruí line spans the Amazon rainforest and the Amazon river to supply the Brazilian coast with inland hydropower:

China's Zhoushan Island crossing involves lattice pylons taller than the Eiffel tower and spanning 2.7 kilometers of open sea: 

But electricity traveling through a wire generates waste heat! So much so that if you tried sending electricity exactly as it was generated to buildings thousands of kilometers away, it would all be lost. 

The solution is that for a given amount of power, higher voltage means lower current. 

So the trick is: step the voltage way up before sending power long-distance (reducing the current and therefore the losses) then step it back down to usable levels at the destination. 

The machines that do this are large power transformers (LPTs), which are massive rings of iron submerged in mineral oil and wrapped in kraft paper and copper coils. There are maybe about 16,000 in existence: ~6K in the US, and who knows how many in other countries (public data here is so fucking terrible).^[1]

Transformers at this scale are artisanal creations. They are fragile, expensive, and unique. Think of them as Ming vases or Fabergé eggs. They take 2-4 years to produce, and only a handful of companies can make them. Earth produces less than a thousand a year. They are designed to serve for over half a century. 

LPTs exploit Faraday's Law, which dictates that for each "turn" of the copper wire in the transformer above, voltage decreases or increases proportionally to the amount of total turns there are. Each turn of copper is separated by vacuum-sealed cellulose paper doused in machine oil, which acts as an insulator. The insulator forces the electricity traveling through the transformer to not short-circuit: instead of taking the short route it wants to take (as lightning does), it must take the long, winding passage that inevitably steps its voltage up or down. 

If your grid needs to step voltage X into voltage Y, you need to design your transformer with the exact number and size of turns required to do this: an LPT can't replace another LPT with different voltage specifications, and can't be changed once it's built either.

These things are also absurdly heavy: 200-400 tons, and the size of a small house. Due to exceeding road weight limits, they require custom rail cars (Schnabel cars) for transportation, of which there are about 30 in North America. 

In the US, moving an LPT involves months of advance planning and permits. You have to disassemble your transformer so your railcars can support its weight, empty it of its oil, query each individual overpass, road, and bridge for clearance, and then finally chug along each stretch. The trip can take weeks. 

The journey is so complicated that you need to backchain from it to design your transformer: map out every bridge, rail, and tunnel to retrofit the thing so it doesn't damage rails, can actually fit through things, and doesn't break apart on the way.

You also need to backchain from the specific nonswappable electric magic variables of the grid you're serving like its power rating or impedance, as well as the cooling requirements given local weather. 

This is why manufacturers take 2-4 years to build an LPT: each one is very different from the next.^[2] You cannot swap out one for another. They're absurdly bespoke.

Solar storms can cause LPTs to violently, messily explode 

A coronal mass ejection on the scale of the Carrington Event would be a bubble of rapidly expanding magnetized plasma weighing ~around the mass of a smallish lake, hurled away from the sun at speeds sufficient to reach Earth in ~17 hours.^[3] Only a billionth of the bubble would hit Earth (1AU is a long way to go!) and anyway all the atoms it's made of would bounce off our magnetosphere. So the actual problem isn't the absurd kinetic energy a CME contains (~as much as Chicxulub), but its magnetic field. 

There's a coin flip chance the field is oriented southward, in which case instead of bouncing off our own field like the atoms do, the field would "link up" with Earth's field, riling up the Earth's ionosphere's currents (electrojets). Crucially, we would not know what direction the field is until about 15-20 minutes before the CME hits us. That's because only one operational satellite at the Lagrange 1 point between Earth and the sun has the required equipment to measure this and warn us ahead of the CME. 

The thermosphere (everything right above the Karman line) would convert most of that electrojet excitation into heat, causing the air to rapidly swell. Without thrusters, satellites would deorbit in days or weeks. Also a hundredth of the excitation would be converted into nice auroras. 

What we really care about is that electrojets thrashing around a hundred kilometers above us causes shifting magnetic fields the size of continents to skirt the surface of the Earth. Fields get turned into currents when they go through the earth, which is broadly conductive. Transmission lines are grounded for safety reasons (a copper stake is driven into the ground at either end), and the most conductive stuff around, so current immediately flows through these lines from the ground. Long-distance power lines are hit the most, because the longer the affected line, the more powerful the current. 

Electrojet activity is slow enough that these currents (GICs) alternate the direction of their flow every few minutes, making them, for our purposes, DC rather than AC. Long distance transmission lines, meanwhile, use AC current. (They're three phase circuits, which allows us to avoid having to build a return line.) 

So you end up feeding transformers with AC current corrupted by DC current from a storm-induced circuit running up from the earth itself:

Transformer iron cores depend on the reliability of magnetic flux flowing one way for a fraction of a second, and the next the other, at magnetic strengths lesser than what the core can handle without saturating. For example, the current will fluctuate from 1.5 teslas of magnetic strength to -1.5 teslas 60 times a second (60Hz) for a transformer rated for 1.7T.^[4] A little bit of DC in your AC immediately causes the AC to go out of whack, where the field fluctuates from e.g. -1.2T to 1.8T, saturating the core on half the cycles. A saturated iron core is no better than air generating a magnetic field: it's like teleporting the iron core into space every 8 milliseconds, leaving the copper coils alone for that time with nothing to control the throughput of current. This produces a chaotic magnetic field that swings around the sealed box like crazy, trying to grab onto anything that can contain it that's not the core.

The results are catastrophic: the mainframe of the transformer, the radiator fins, every bolt the structure is made of, suddenly become subject to intense magnetic forces, which are immediately converted to heat, which, well, the transformer is a tank filled to the brim with paper and machine oil... 

Even if a transformer doesn't violently, messily explode, the effects of the metal in it overheating are really nasty. Heat causes the paper the wires are enmeshed in to literally bake, degrading enough to cause coils to become exposed to each other. Indeed, this is why transformers need to be replaced every half-century anyway. Micro-arcs of lightning appear inside the transformer, as currents reach for the quickest point from A to B, meaning coil-to-coil rather than through the coils. Arcs build up more heat, which bakes more paper, which generates more arcs. This is why a transformer can explode in minutes, or, if saved by the bell by an operator shutting down the current, severely degrade itself. Even if only a few hundred LPTs on Earth were to blow up in a Carrington event, thousands more would silently fail in the weeks or months after the CME. 

This is irreversible. You cannot unmesh the paper from the copper wires, because they are too tightly bound together. Once a transformer is dead, it's dead for good. Now you have to order, transport, and replace the LPT into your bespoke grid location, inside a massive blackout, while demand skyrockets.

Fabergé eggs I tell you: 

A blackout whose solution is bottlenecked by our ability to manufacture new transformers would be hell. Refrigerators, water treatment plants, sewerages, hospitals, streetlamps, factories, food processing and data centers are all electric things. 

Shutting down the grid automatically (transformers thrashing around can sometimes trip breakers, though not always) or by operators would later require a "black start" reinitializing of the grid, which can take weeks (catch-22s on power abound in the grid; for example, generators need electricity to begin spinning, same as you need to jumpstart a car). 

New Zealand

In 2001 a solar storm hit New Zealand and permanently bricked a transformer in Dunedin. There are only a few hundred large transformers in NZ, and NZ is exposed to Southern electrojets; figuring out how to not permanently brick transformers became a priority of Transpower, the (single!) national grid operator. This took the form of 72 GIC sensors installed, a detailed model of the entire transmission network and ground conductivity maps of NZ, and a plan to disconnect the key locations which would most concentrate GIC risk in the most vulnerable transformers. When a storm came in 2024, the grid was successfully rerouted and partially shut down, protecting most kiwi transformers while continuing to serve consumers. NZ is the best prepared country on Earth for a Carrington event, and even they estimate 13-35% of their transformers would be at risk from permanent damage if one came up. (But unlike most grids, they acknowledge the problem!) 

America has 8 grid operators, not one, and 3,000 utilities you'd have to negotiate GIC sensor installation and data collecting with if you wanted to do what NZ did in the US (the utilities are who owns the hardware in any case). But in principle we could do what the kiwis did, and it would avert a lot of damage. 

Québec

Québec, like New Zealand, is exposed to the electrojets at the poles; but it also is sitting on a large plate of precambrian rock that works as an insulator, forcing magnetic fields to "bounce off" the ground rather than being absorbed by it, affecting power lines and transformers on the surface all the more

In 1989 a solar storm hit Québec and half-cycle saturated all its transformer cores, driving its LPTs to draw an immense amount of current from the grid, tripping breakers everywhere and leading Québec to blackout in 90 seconds. It took 9 hours to reinitialize the grid, during which the region was in the dark. 

The reason the Quebec grid shut itself down so fast is because it works as a funnel with all power coming from a concentrated few lines: 

And those SVCs (machines which maintain voltage along the length of these power lines) were tuned to react to harmonics by tripping a breaker and shutting down the line, putting more load on other lines, causing more breakers to trip, etc. until all the long distance transmission lines were down in 2 minutes. 

This is poor design, actually: the Québec grid puts all its eggs in one basket, which for threats like ice storms (such as this one in 1998) makes it much more likely to fail. By contrast, a dense spiderweb grid like the American Eastern seaboard's could lose hundreds or thousands of transmission lines to natural events and be mostly fine: there are dozens of redundancies for each line, which would immediately reroute power. The American grid has a strong immune system for fighting off transmission line failures; Québec does not. 

But for GICs? The grid shutting itself off is perfect, since it immediately cuts LPTs from their power source, saving them from being fried. Indeed, a transformer near a nuclear reactor in New Jersey was permanently destroyed due to the same 1989 storm that left Québec's transformers intact, because every grid failure kept rerouting power to every other line, and transformers were exposed to GICs during the whole duration of the storm. The American grid is vulnerable to auto-immune disease, since a grid thrashing to keep power running through a solar storm is a grid that maximally damages its transformers. This failure was deathly quiet: no alarms blared during the storm, since the grid wasn't designed to detect or defend against GICs. The transformer just quietly failed the next day. It's likely, given precedent in South Africa (where the 2003 Halloween solar storms caused 12 transformers to brick themselves in the weeks and months after the event), that the 1989 storm also cut the lifespan of dozens of American transformers by half or more, and we just didn't think to detect it since transformer "blood tests" were not in vogue yet.^[5]

Québec swore to not let this happen again, and installed series capacitors—devices that block DC but let AC go through—on all their lines. This was an expensive measure, because capacitors are not cheap, and require majorly overhauling lines. They also have to be implemented universally, since a protected line will only place more load on unprotected lines, making vulnerable transformers even more liable to fry. A utilities engineer in 2021 claimed "if the 1989 storm happened again today, I believe Québec would not lose power." Québec is "[...] confident that our network would survive the anticipated worst case GIC." 

Silver bullet

SolidGround (whom I'm going to shill a lot, they didn't sponsor me I swear) is a nifty ground neutral blocker invented by a company founded in the 2010s (Québec set up protections for their grid in the 1990s, which is why they had to go a far more expensive route than SolidGround). 

An ungrounded transformer is a hazard; but also, the grounding cable is exactly where GICs come from. What SolidGround does is sit on the neutral and provide a capacitive blocking path parallel to the normal metallic ground path. If a certain threshold of DC is detected within the ground, SolidGround kicks in in a matter of seconds, and closes the default ground path to replace it with the capacitive blocking path. Capacitors block DC power but let AC through: the transformer would still be grounded, but would be invulnerable to GICs. SolidGround adds a third path which they call a "spark gap", is confidential IP, and can provide a brute force path to the ground in case of unexpectedly high voltage during a solar storm where the capacitive blocking path is engaged. This would presumably allow SolidGround to protect a transformer during a Carrington Event just fine. For 500K a pop! Installation takes hours, and doesn't affect the grid at all, only the grounding cable right below the transformer.

Three such devices were installed in federal utilities (like the TVA), paid for as a result of an Obama executive order and a later Trump 1.0 EO. They worked exactly as intended during the 2024 Gannon storm (the same storm NZ successfully handled by shifting their grid around). Validated by national labs, and two administrations... and then nothing. 

China reportedly stole the IP behind SolidGround and has been installing their own copies to protect against GICs.^[6] But Americans haven't done much more. All of SolidGround's customers are in foreign nations (clients not disclosed).

The yearly cost from mundane solar storms alone—whenever an aurora is sighted anywhere, that's a solar storm—is estimated at 10 billion dollars. That's without pricing in tail risks like a Carrington Event, which would easily cost the economy trillions and kill tens of thousands of people, which is billions in expected value every year. Meanwhile, protecting the grid's "6,000 most critical transformers" with SolidGrounds would cost 3-4 billion dollars to the US. That's 0.3% of the 2021 infrastructure bill, recuperable within the year! 

Note especially that economies of scale apply to SolidGround devices, vastly more so than Fabergé eggs like large power transformers. In fact I don't buy the report's estimates of 3-4 billion for protecting 6,000 transformers; economies of scale could plausibly reduce costs down to about 200-300K per device (these things really aren't complicated, they're just big and require sturdy material quality), which would reduce the total order to about 2 billion dollars, and make it far cheaper for other countries to protect their own grid.

Why hasn't this happened? Probably a combination of: a) transformers last a while, and making one last 50 years instead 42 years might in theory save the company 2 million dollars down the road, but raise operational costs in the meantime ("an extra 500K to all new projects?? So that we can last to 2060?") and b) tail risks like a Carrington Event are not insurable because they're massively correlated risks and the "insurance" backstop here is the government itself. There's not much incentive to hike transformer installation costs when the government is responsible in case it fails due to a Carrington Event.

Government preventative intervention is probably necessary to get anywhere near a state where we'd be safe from a large solar storm. And that's in the US—the rest of the world doesn't even necessarily have the budget to buy these things en-masse. The kind of country that can't afford these also correlates well with the kind of country that would incur incredible human costs if plunged in a blackout for months. 

Buying a few dozen SolidGrounds to install in a few dozen countries as a demonstration (e.g. New Zealand would be amenable to this, given their seriousness so far wrt solar storms) might be a neglected philanthropy cause. 

Conclusion

A Carrington Event level solar storm—a Coefficient Giving report estimates the likelihood at about 0.70% per decade—would wreak a lot of damage on Earth. But beyond knowing that, it's hard to tell exactly how much, and where, the havoc would be. 

Most utilities keep their exact specifications secret. How many transformers are already shielded? How many are safe inside grids that are sensitive and isolated enough like Quebec's to shut themselves down thanks to break trippers? How many are located in a node of the grid that would shut itself off first, dooming every transformer around it BUT leaving it intact in a grid-blackout-domino scenario? 

As for the effects of plunging 50% or 75% or [aaaaah no dataaa^[7]] of a country into darkness, it's hard to tell how bad they'd be. There are only so many examples of prolonged blackouts, such that drastic claims like "society would collapse and tens of millions would die" (as a congressional testimony by an ex-CIA director posited) are difficult to quantize. It's also difficult to tell how effective rerouting the surviving transmission lines to priority infrastructure like water treatment and sewerage would be.^[8]

But a Carrington event would not knock out electronics on Earth (though it may damage satellites), as I've seen some people on Twitter claim. It would also probably not damage power sources and their generators, given those tend to already be protected against any funny business from the grid. Distributed solar, e.g. your roof and battery pack, would be fine. Diesel generators powering hospitals would be fine, until they run out of diesel. A large solar storm's effects are essentially entirely confined to a specific failure mode in large power transformers, which happens to be entirely preventable by existing, relatively cheap technology. We just do need to, y'know, actually install it.

Thanks to Sentinel for funding this post, Opus 4.6 for a lot the research, and roastmypost.org for last-minute fact-checking. 

I would like to continue doing research of this type. You can support me (with money) on Patreon! 

1. ^{^}

   Estimating this correctly is interesting because on the face, it should be easy: Google Earth allows you to see every transformer on the planet and you could train a model to individually count them. But transformers are so fucking bespoke that knowing the amount of transformer-shaped things there are doesn't help you estimate how bad a solar storm would be. 

2. ^{^}

   Manufacturers tend to be: Hyundai Electric (South Korea), Siemens (Germany), Hitachi Energy (Japan/Switzerland), HICO (South Korea). 
   
   There is very little domestic capacity for LPTs in the US (a few dozen LPTs a year). Meanwhile, China produces about 60% of the world's transformers (a few hundred LPTs a year), mostly for the domestic market. A reminder:

3. ^{^}

   Not going into detail here because, unlike other parts of this post, humanity can't directly toggle solar storms yet. 
   
   Importantly though, the original Carrington Event arrived in 17 hours' time rather than the typical 1-3 days for solar storms; by chance, previous solar storms had "cleared the way" of particles which would've otherwise slowed it down. A carrington-class storm that narrowly avoided our orbit by 9 days in 2012 traveled at similar speeds. A worst case scenario could even involve warning time of 15 hours only, given precedent! 
   
   It's worth noting speed and intensity are positively correlated at 0.66 for run-of-the-mill storms; stormy regions of the sun's surface fire multiple CMEs over days, which both increases the likelihood of a carrington-class solar storm AND the likelihood of a snowplow CME. 
   
   Just to give an idea of the scale here, the 2024 Gannon storms spawnpoint was visible with the naked eye and spanned 10-15 Earth diameters. CMEs expand rapidly when lobbed toward Earth, so we travel through a hyperactive region's firing line for 5-7 days.

   The reason Carrington events are rare is because you need a goldilocks frontier of an unusually large and energetic active region + peak flaring during the Earth-facing window (coin flip; peak flaring can happen any time during the 2-3 weeks of activity, and the sun rotates in ~27 days) + southward magnetic field orientation at Earth impact (coin flip) + snowplow preconditioning for maximum speed and intensity. 

   

4. ^{^}

   LPTs are heavy and intransportable enough as they are: providing more saturation margin than 0.2 teslas would be expensive. 

5. ^{^}

   Damage to the kraft paper and machine oil substance that a transformer tank is filled to the brim with can be measured by taking a sample and establishing how much dissolved gas it contains. This directly correlates with how much chemical decomposition was caused by a solar storm. 

   Finding elevated concentrations of hydrogen means "I'm experiencing partial discharge." Acetylene means "something is arcing inside me, this is bad." Rising carbon monoxide means "my paper insulation is baking."
   
   For example, here's what South African engineers found in one of their transformers (graph found in a previous solar storms risk breakdown by Opefficient Givanthropy):

   

6. ^{^}

   Costs from transformer damage in the US due to common low-level solar storms is estimated at 10 billion dollars a year. (Same for the EU.) 

   This is frustrating to me, because Emprimus is having trouble finding customers, and the reason their device costs half a million in the first place is because the only people who are scaling it paid them nothing for the 10 years and millions in R&D that went behind the design. 

7. ^{^}

   If utilities weren't so f***ing secretive about things, I could get Claude Opus 4.6 to run simulations of grid failures in every country in the world to estimate which transformers are the most vulnerable, how exactly the grid would fail, etc. This is what New Zealand did successfully! And it's well within current AI capabilities. 
   
   But as with a lot of claims about AI capabilities—cancer cure, etc.—it's bottlenecked by data. 

8. ^{^}

   The winter storm that hit Texas in 2021 knocked out half of energy generation capacity at its peak and caused a third of Austin's sewage lift stations to go offline, causing overflows of sewage in the streets and, far worse, into waterways later used by treatment water plants unable to do their job without power, thereby injecting sewage into tap water and putting 17 million people under boil-water advisories for weeks. (I don't even—what—designing sewerage systems that don't do that is a different post.) 
   
   Texas estimates 80-130 billion dollars in damages and at least 250 dead—from a state with 30 million people in it.