The Refinery

In the last chapter we left a railcar of white sand rolling out of the mountains of North Carolina. That sand — quartz, silicon dioxide, refined to ninety-nine point nine-nine percent purity — was clean enough to make the crucible. It was nowhere near clean enough to make the cell.

This is the part that trips people up, so it is worth saying slowly. The quartz from Spruce Pine does not become the silicon in your solar panel. It becomes the bucket that holds the silicon in your solar panel while that silicon is being grown. Two different materials, two different purities, two different links in the chain. The bucket needs to be four-nines pure. The thing growing inside the bucket needs to be a thousand times purer than that.

So where does that silicon come from? Not the bucket — the actual silicon, the stuff that catches the sunlight?

It comes from a refinery. And the refinery is, in many ways, the most extreme single step in the entire journey from rock to rooftop — a place where a cheap, abundant, filthy industrial commodity is transformed, at staggering cost and temperature, into one of the purest substances human beings know how to make in bulk. If Spruce Pine is a story about a lucky accident of geology, the refinery is a story about brute-force chemistry, oceans of electricity, and — increasingly — geopolitics.

Let us walk down into it.

From beach sand to "metallurgical" silicon

The starting point is not even the good quartz. It is ordinary quartz gravel — high-grade, but nothing exotic — and it goes into something called a submerged-arc furnace, along with carbon, in the form of coal, coke, and wood chips.

The job here is chemically simple to state and physically violent to perform: rip the oxygen off the silicon. Quartz is SiO₂ — one silicon atom bonded to two oxygen atoms — and those bonds are ferociously strong. To break them, the furnace runs at around 1,800 degrees Celsius, with electrodes pumping enough current through the mix that the carbon grabs the oxygen (floating off as carbon monoxide and carbon dioxide) and leaves behind molten elemental silicon.

What pours out is called metallurgical-grade silicon. It is about 98 to 99 percent pure. In almost any other industry, 99 percent pure would be a triumph. Here, it is dirt. That remaining one-to-two percent — iron, aluminum, calcium, carbon, boron, phosphorus — is a catastrophe for a solar cell, where, as we established in the last chapter, a single atom in the wrong place is a pothole in the road the electrons have to travel. Metallurgical silicon is fine for making aluminum alloys or silicones. It is useless for catching sunlight.

The world makes a lot of it — on the order of nine million metric tons of silicon metal a year. It is cheap. It is everywhere. And it is the raw feedstock for the step that actually matters, the step that separates the people who dabble in silicon from the handful of companies on Earth who can make the real thing.

That step is the refinery proper. And here the numbers stop sounding like chemistry and start sounding like aerospace.

The Siemens process: a $1.2 billion machine that cannot be turned off

The dominant way to turn dirty metallurgical silicon into ultra-pure polysilicon is a method developed in the 1950s and named, like so many things in this industry, after a German company: the Siemens process. As of 2025 it still accounts for roughly two-thirds of all polysilicon made on the planet — about 66 percent of the market — and for the very highest purities, the eleven-nines material the semiconductor industry demands, it is essentially the only game in town.

Here is how it works, and the thing to hold in your mind is that every stage exists to chase out impurities that the stage before couldn't reach.

First, you take the metallurgical silicon and react it with hydrogen chloride to produce a liquid called trichlorosilane. This is the clever move: by turning the silicon into a liquid compound, you can purify it the way chemists purify everything — by distillation, boiling it and condensing it over and over, each pass leaving the heavier impurities behind. Distill trichlorosilane enough times and you arrive at a liquid of almost unimaginable purity.

Then you reverse the trick. Inside a sealed reactor, you run that ultra-pure trichlorosilane gas over thin silicon rods — "seed" filaments — heated to around 1,100 degrees Celsius. The silicon, and only the silicon, deposits out of the gas and onto the rods, atom by atom, growing them slowly into thick columns of polysilicon over the course of days. The impurities, having been left behind in the distillation, never make it onto the rod.

What you end up with is polysilicon: purity in the range of 99.9999999 percent — "nine nines" for solar, and up to eleven nines for semiconductors. To put nine-nines purity in human terms: it is the equivalent of one impure atom for every billion silicon atoms. The Spruce Pine quartz that so impressed us in the last chapter, at four nines, is by this standard a grubby rock.

Now, the cost of doing this. A new Siemens-process polysilicon plant of meaningful scale — call it 100,000 metric tons a year — costs around $1.2 billion to build. And the running conditions are merciless. The reactors operate at over 1,100 degrees for days at a stretch; the distillation columns and recovery loops run continuously. The defining feature of a polysilicon plant, the thing that shapes the entire economics and even the geopolitics of this industry, is that it does not like to stop. Starting up and shutting down a plant is slow, expensive, and hard on the equipment. So these plants are built to run flat-out, around the clock, for years on end. They are less like factories and more like blast furnaces or oil refineries: enormous, continuous, capital-devouring machines that must be fed power and feedstock without interruption or they bleed money.

Which brings us to the single most important input of all. Not the sand. The electricity.

Why polysilicon is really congealed electricity

Making polysilicon by the Siemens process is one of the most energy-intensive industrial processes in widespread use. The furnaces, the reactors held at 1,100 degrees for days, the endless distillation — all of it runs on electricity, and the appetite is enormous. There is an old line in the industry that polysilicon is really just "solidified electricity," and it is barely an exaggeration. The cost of a kilogram of polysilicon is, to a first approximation, the cost of the power it took to make it.

This one fact explains almost everything about where the polysilicon industry has gone in the last fifteen years.

If your product is congealed electricity, then you will build your factory wherever electricity is cheapest — and you will not be too fussy about where that electricity comes from. For a long time, the cheapest electricity in the world, available at the scale a polysilicon industry needs, was Chinese coal power, much of it in the far-western region of Xinjiang. And so the polysilicon industry, which in the 1990s and 2000s had been led by Western and Japanese names — Hemlock in Michigan was the global leader from 1994 to 2011, alongside Wacker in Germany — migrated, almost wholesale, to China.

The migration is now essentially complete, and the figures are stark. As of 2024, China holds something like 93 percent of global polysilicon production capacity, and produces over 80 percent of actual output. Nine of the world's ten largest polysilicon manufacturers are Chinese. The top four alone — Tongwei, GCL Technology, Daqo New Energy, and Xinte Energy — command about 65 percent of global output between them. Tongwei, the largest, has a capacity north of 900,000 metric tons a year, all by itself. The market leader of 1994, Hemlock, had by 2024 fallen to around 14th place, overtaken by Chinese newcomers that did not exist a few years earlier.

There is exactly one non-Chinese company left in the global top ten: Wacker, the German chemicals group, refining away in Bavaria much as it has for decades. It is the last Western holdout near the top of a podium it once owned.

This is the second great irony of our journey, stacked right on top of the first. Recall from Chapter One that the rock — the irreplaceable high-purity quartz — is overwhelmingly American, dug out of North Carolina. And yet the refining of silicon, the step that turns abundant rock into the purest material in the supply chain, is overwhelmingly Chinese. America mines the bottom of the chain and China refines the next layer up. We will see this pattern repeat, layer after layer, all the way to the rooftop. The United States keeps owning the parts of the solar supply chain that come before the value is added, and importing the parts that come after.

For most of the last decade, the world mostly shrugged at this. Polysilicon was a commodity; it came from wherever it was cheapest; nobody at the top of the chain thought much about the refinery at the bottom. And then a single word changed everything.

The word that split the market: Xinjiang

A large share of the world's cheapest polysilicon was being made in Xinjiang, the region of northwestern China where coal power is abundant and cheap. It is also the region at the center of the most serious forced-labor allegations of the past decade. In 2021 the United States passed the Uyghur Forced Labor Prevention Act — the UFLPA — which created a rebuttable presumption that any goods with inputs from Xinjiang were made with forced labor and are therefore barred from entry into the United States. The burden flipped: an importer now has to prove a clean supply chain, rather than customs having to prove a dirty one.

For most consumer goods this was a headache. For solar, it was a structural earthquake — because so much of the world's polysilicon, the substance at the base of every panel, traced back to exactly that region. Overnight, the question of where your specific batch of polysilicon was refined went from a triviality nobody asked to a compliance question that could get an entire shipment of finished panels seized at a US port.

The market did what markets do: it split in two. Today there is, in effect, a two-tier price structure for polysilicon. Xinjiang-origin material trades at a discount — on the order of 3 to 5 percent — and carries seizure risk in the United States. Meanwhile, polysilicon refined outside Xinjiang — in Sichuan, in Yunnan, or by the Western producers — commands a premium of roughly the same size, precisely because it can clear US customs without a fight. Chinese producers have been physically shifting capacity to Sichuan and Yunnan to chase this clean-origin premium, even though hydropower there can cost up to a dollar a kilogram more than Xinjiang coal. They are paying more for electricity in order to make polysilicon that is legally cleaner, not chemically cleaner. The atoms are identical. The paperwork is worth a premium.

This is the moment the refinery stopped being a boring commodity step and became the beating heart of the entire American solar-trade story. Every argument you have ever heard about tariffs, about Domestic Content bonuses, about "where is this panel really from" — almost all of it, when you trace it down, bottoms out here, at the refinery, in the question of whose electricity congealed into your silicon and where.

The American attempt to come back

Here is where the story turns, slightly, and where it starts to matter directly for anyone selling solar in the United States in 2026.

If polysilicon is congealed cheap electricity, and if the US has decided — through the UFLPA, through tariffs, through the Inflation Reduction Act's Domestic Content bonuses — that it wants a solar supply chain that does not run through Xinjiang, then someone has to refine polysilicon outside China at competitive scale. And so, after years of retreat, there is the beginning of a Western comeback.

Hemlock Semiconductor, the fallen giant in Michigan, secured a $325 million federal grant in 2023 and is building new capacity, adding several thousand metric tons a year by 2026. REC Silicon spent some $200 million reviving its Moses Lake plant in Washington State, idle since 2019. Wacker keeps investing in Bavaria and in its US operations in Tennessee. None of this comes close to displacing China — the Western players remain a rounding error against Tongwei's 900,000 tons — but it is the first serious attempt in a generation to rebuild a non-Chinese polysilicon base, and it exists almost entirely because of policy, not economics. Left purely to the market, every gram of this material would still be made next to the cheapest coal plant on Earth.

For the buyer of solar modules in America, the practical upshot is this. There is now a meaningful, paperwork-backed distinction between a panel whose silicon was refined in Xinjiang, one refined elsewhere in China, and one refined in the West. Those three panels may be physically indistinguishable. They may even come off similar production lines. But they are three different products in the eyes of US customs and the US tax code — different in import risk, different in tariff exposure, and different in whether the project they end up on can claim the Domestic Content bonus. The refinery, four layers below the rooftop, reaches all the way up and changes the price and the legality of the thing you actually install.

What the refinery teaches us

Step back from the heat and the chlorine and the billion-dollar reactors, and the refinery teaches the same lesson the mine did, only louder.

The mine taught us that the chain is narrow: a huge industry resting on one Appalachian valley. The refinery teaches us that the chain is concentrated and political: a huge industry resting on a single country's cheap electricity, and now being slowly, expensively, deliberately pulled apart by trade law. Both lessons point the same way. The further down the chain you go, the fewer the players, the larger the machines, and the more a decision made at the bottom — a hurricane, a furnace fire, a forced-labor statute — ripples all the way up to the invoice on a rooftop in Atlanta.

And notice the through-line of irony now firmly established across two chapters. The rock is American. The refining is Chinese. The policy is American again, trying to claw the refining back. The whole drama of American solar in the 2020s is the story of a country that owns the very bottom of the supply chain and the very top — the quartz and the customer — and is fighting to win back the enormous, electricity-hungry middle it gave away.

That middle is where we are headed next. Because the polysilicon, once refined, does not yet look like anything you would recognize. It is chunks and rods and beads of grey, glassy, hyper-pure silicon. Someone has to melt it — in a Spruce Pine crucible, naturally — and grow it into a single perfect crystal, and slice that crystal into wafers thinner than a sheet of paper without shattering them.

That is the next layer. That is where the silicon finally becomes something that could, one day, sit on your roof.