We spent $15,000 on billboards targeting one person: the guy controlling all the chemical spend at a saltwater disposal company in Texas. We mapped his commute and bought every billboard between his house and the oil field. When we finally called, he said "I see your billboards everywhere." That landed us our first oil field contract. At the time our entire operation was a $10,000 reactor built from PVC pipes from Home Depot, turning corn sugar into industrial chemicals. People keep trying to throw it away. It still works. That leaking reactor started a multibillion-dollar company. @ycombinator visited our plant in Houston. The original PVC reactor is still on the floor next to the Bioforge.

He’s right that it’s a stunt, but wrong about fill time. Tesla’s engineering VP says filling all 600 liters takes 6 to 7 hours today, from an external nitrogen supply. A compressor in the car would need about 90 kWh: about 18 hours at 5 kW, the power of a large portable dive-air compressor. Only a compressor of about 0.5 kW would take a week. The Thrust version’s 200-liter tank gives about 5 to 6 seconds of thrust. If you want the car to fly, my first fix as a chemist would be a denser propellant. The 1961 Bell Rocket Belt flew on 90% hydrogen peroxide, broken down on a catalyst into steam and oxygen at about 740 °C. We started Solugen making peroxide. At 90%, it’s a strong oxidizer that can set clothing and oil on fire on contact. You wouldn’t put that in a car. The best use case for compressed gas and a few seconds of thrust is braking. Tesla says the thrusters push 4 to 5 times harder than the tires grip. Fired forward, that would cut the Roadster’s stop from 60 mph from about 95 feet to under 20, in under half a second. Tesla says the tank holds up to 10,000 psi. Cars already carry gas that compressed for years: some airbags inflate from sealed bottles of argon and helium filled to as much as 10,000 psi, built to work for more than 10 years. For a brake used once a year, the 6-to-7-hour refill doesn’t matter.

On November 10, China’s pause on its broadest rare earth export controls runs out, unless Beijing extends it. America is spending billions to rebuild its minerals supply chain. But the gap with China sits in the chemistry after the mine. Many critical metals are byproducts, including the heavy rare earths dysprosium and terbium. 38 of 62 metals come mostly out of another metal’s ore, per a 2015 study. Tellurium comes out of copper, germanium out of zinc and gallium out of aluminum, at the plant that refines the main metal, if that plant recovers them. Copper: smelters now process concentrate for a fee of $0, so more of their income comes from byproducts such as sulfuric acid and the sludge under the copper anodes. That sludge holds tellurium, and about 70% of the world’s tellurium goes into solar cells. Rio Tinto’s Kennecott refinery in Utah recovers it and ships it to Montreal, where 5N Plus refines it, mostly for First Solar, the largest US solar maker. The US did not refine any tellurium last year. Rio Tinto built that circuit for $2.9M, and it also cut the waste Kennecott has to treat and dump. Germanium: China banned germanium exports to the US in December 2024, then paused the ban until November 27. Alaska ships some of its germanium to Canada inside zinc concentrate, where it is recovered. A 2025 Science study found that recovering under 1% of the germanium in US metal mines would end US imports. Korea Zinc plans a $7.4B smelter project in Tennessee, targeted for 2029, that will recover germanium. I manufacture chemicals. The chemistry that recovers these metals must be judged by its running cost against three things: the critical metal it recovers, the waste the plant no longer has to treat, and any extra copper or zinc it lets the plant produce. A process can cost more and recover less of the critical metal, and still be worth running because of the other two. The plants that gain the most from that chemistry are the ones with a waste or throughput problem. Say a more selective process costs $2M a year to run and recovers $0.8M of metal. If it also saves $1M in waste treatment and frees $3M of copper output, it’s still worth running even if the metal’s price falls to zero. A process built to recover as much metal as possible might recover $2M of metal, save $0.2M in waste and cost $1.5M to run, a net of $0.7M. If enough copper and zinc plants run the more selective process, they can sell these metals at prices that producers who mine only that metal cannot match. A supplier can charge more for that chemistry and still lower the plant’s cost.

Two PhDs pouring peroxide into strangers' hot tubs. Our parents were so disappointed. In Dallas, @TungstenSeanide and I loaded peroxide from our reactor into his Subaru and drove it to the city's float spas. On weekends we poured the chemicals into people's hot tubs ourselves. It was full service. Then the reactor started making about $10,000 a month. For two guys with day jobs, that was like an infinite amount of money. We knew we had something real, so we looked up @ycombinator. At the time, YC didn't have much of a track record with hard tech.

👀

A filter press is a row of square plates standing on edge, squeezed together by a hydraulic ram, with a hollow chamber between each pair. The biggest have 141 plates measuring 5 by 5 meters and weigh 400 tonnes. Pumps fill the chambers with mine tailings and force the water out through filter cloth. The solids stay behind in each chamber as a damp cake firm enough for bulldozers to drive on. Then, within one minute, all 141 plates open to drop the cakes. In 2019 Brazil banned upstream tailings dams, which are raised in steps on top of their own tailings. Brazil still has 42 of them. Tailings are the wet ground rock left after the iron is separated out. Most mines pump them into a pond behind a dam. Loose tailings that stay soaked are unstable. A small push can raise the water pressure between the grains until the grains stop bearing on each other. The tailings then liquefy and flow out as mud. That happened at Vale’s Dam I in Brumadinho, more than two years after the mine stopped sending tailings to it. The tailings had been slowly deforming under their own weight, and heavy rain at the end of 2018 weakened the top layers. Within 10 seconds the dam collapsed, and the mud buried 270 people. At Mineração Morro do Ipê, each press has 208 plates measuring 2.5 by 2.5 meters, in a machine 29.4 meters long. A cycle takes 23 to 25 minutes. The plates open, cakes with 14% moisture drop onto a conveyor, and dozers compact them in layers into a dry stack. Three presses handle 13,344 tonnes of dry tailings a day and return 9,700 cubic meters of water to the plant. Most mines still pump slurry behind a wall because filtering and stacking costs about $2.00 to $3.50 a tonne, against about $1.00 for a conventional dam. In a 2021 survey of 1,743 disclosed facilities, 74 were dry stacks.

About 1 in 8 American adults say they take a drug like Ozempic or Zepbound. The first GLP-1 drug, Byetta, was a synthetic copy of a peptide from Gila monster venom. Human GLP-1, the gut hormone that tells the pancreas to release insulin, lasts about 2 minutes in the body. An enzyme called DPP-4 cuts off its first two amino acids, right after the alanine in position 2, and the shortened hormone no longer triggers insulin release. The lizard's peptide has glycine there, and DPP-4 doesn't cut glycine. Its synthetic copy, exenatide, lasts 2.4 hours. John Eng, an endocrinologist at a veterans' hospital in the Bronx, read that Gila monster venom made the pancreas swell. In 1992 he pulled a 39-amino-acid peptide out of that venom. About half its amino acids matched human GLP-1. Eng's employer, the VA, declined to patent it because it wasn't specific to veterans, so he filed the patent himself in 1993. For 3 years, almost every big diabetes drugmaker ignored him or passed. In 1996 Amylin licensed it, and the FDA approved it as Byetta in 2005. Novo Nordisk built Ozempic from human GLP-1, so DPP-4 would cut it too. Ozempic's semaglutide swaps that second alanine for Aib, an amino acid that isn't in human proteins, to protect it from DPP-4. Eli Lilly partnered with Amylin on Byetta from 2002 to 2011. In 2025, Lilly's own GLP-1 drug, tirzepatide (Zepbound and Mounjaro), sold $36.5B and passed Merck's Keytruda as the world's best-selling drug. Lilly ended tirzepatide's chain with amino acids copied from the monster's peptide.

In 2021, a tree fell onto a bare PG&E power line in Northern California and started the Dixie Fire. It burned 963,309 acres and destroyed 1,329 buildings. PG&E and Southern California Edison alone spent $25.6B on wildfire prevention in 2020–2023 and budgeted $25B for 2026–2028. Last year, PG&E reported 879 fires involving its equipment. 840 of them started on its distribution lines, the lower-voltage wire that runs through neighborhoods. Most US overhead distribution wire is bare aluminum. A branch that hits it can carry current to the ground and catch fire. Wind can slap two wires together and throw molten aluminum. A broken wire can land live in dry grass. Wildfire costs made up about 27% of what PG&E charged customers in 2024. Utilities fix bare wire by replacing it. PG&E buries it at $3.25M a mile, has buried more than 1,400 miles since 2021, and this month filed a plan to bury about 5,000 more by 2037. Jacketed wire costs about $1M a mile. Kratos Materials, a startup founded by @UnchartedSherpa, makes a silicone-based coating for existing power lines. Silicone is a chain of silicon and oxygen, the same two elements as glass, with carbon groups attached. When an arc heats it, the carbon burns off and the silicon and oxygen fuse into a glassy layer. In early tests, the coating vitrified. Its target price is $85K a mile. Since 2021, State Grid utilities in China have used robots to spray insulating coatings onto live 10,000-volt lines. No US utility does this yet. Kratos still has to run its branch, wire-slap and broken-end tests. At its target price, coating PG&E's 25,000 miles of overhead fire-district line would cost about $2B, while burying them costs about $81B.

Meta signed a $6B deal with Corning in January for fiber-optic cable. Nvidia put $500M into Corning in May. In 1880 that same company blew Edison’s light bulbs by hand. Every electronic revolution since the light bulb depended on glass: television, the telephone network, the chip, the smartphone, AI. Edison’s lamp needed a sealed glass bulb that held a vacuum around the filament, or the filament burned up. Corning Glass Works blew the bulbs by hand. A team needed a whole day to turn out 165 bulbs. America’s first televisions went on sale in 1939, after RCA showed them at the New York World’s Fair. The picture tube inside was a glass bottle under vacuum, screen as its base, and made by Corning. Through the 1950s it supplied the black-and-white tube bulbs for American set makers. In 1966 Charles Kao showed that a glass fiber would beat copper if it lost less than 20 decibels of signal per kilometer. No glass that clear existed. In 1970 three Corning scientists made one at 17 decibels per kilometer. Phone companies began laying glass instead of copper. More than 1.6 billion kilometers of fiber are installed worldwide today. Since the 1950s most chips have started as a silicon crystal pulled from a bath of molten silicon above 1,400°C, and the bath sits in a quartz crucible. The quartz has to be above 99.999% pure, or metal from the crucible wall ends up in the chip. Sibelco and The Quartz Corp, the two companies that sell quartz that clean in volume, both mine it in Spruce Pine, North Carolina. In the 1960s Corning found a way to chemically strengthen glass and tried to sell it for windshields. It found no buyer at scale, and shelved it in 1971. Then, in early 2007, Steve Jobs decided the iPhone would ship in June with a glass screen, only five months out. Corning’s CEO told Jobs it could not be done in time. Corning converted a plant in Harrodsburg, Kentucky, from LCD glass, reworked the old chemistry into a thin sheet, and was shipping it before the launch. That glass has since been designed into more than 8 billion devices. In an AI data center the GPU racks are connected to each other with fiber-optic cable. An AI rack needs many times more fiber than a normal server rack: every GPU gets its own network link, and the GPUs have to talk to each other at full speed. Corning is building three new plants in North Carolina and Texas to make ten times as much fiber-optic cable in the US, and expanding its cable plant in Hickory, North Carolina, with Meta as the anchor customer.

Hold a cold spoon over a candle and black carbon collects on it from the vapor. Chipmakers build AI chips with a controlled version of the same process, called chemical vapor deposition (CVD). Three companies (Applied Materials, Lam Research, Tokyo Electron) sell most of the CVD chambers, and lead times for standard thin-film deposition tools have doubled to 12 months this year. The transistors on a chip sit under a thin layer of glass, silicon oxide, which a CVD chamber puts down. Etch tools cut tiny holes through that glass. A chamber like Lam's ALTUS then fills the holes with tungsten to wire each transistor to the metal above. Inside, the wafer sits on a pedestal heated to about 400°C. Two gases come down through a showerhead, tungsten hexafluoride (WF6) and hydrogen, and react on the hot surface. The tungsten builds up from the walls of each hole inward. If the metal seals the top of a hole before the bottom fills, it leaves a gap, and that connection can fail. Most tungsten chambers run on WF6, the densest gas known at room temperature, about 11 times heavier than air. A small group of plants in Japan, Korea and China make it. China put tungsten on its export control list in February 2025, and by July 2026 the tungsten price at Chinese ports had risen more than 600%. Chinese spot WF6 went from about $44,000 a tonne to about $220,000 this year.

Every AI data center needs power, and every power plant needs a step-up transformer to connect to the grid. If you ordered one in 2021, it arrived in under 500 days. Today it takes more than 1,120 days. A large transformer core is made of grain-oriented electrical steel that spends about five days in a 1,200°C annealing furnace. Only one US company makes that steel: Cleveland-Cliffs. Inside a bell annealing furnace, coils of iron with about 3% silicon stand on end in hydrogen while the furnace heats them about 15 to 20°C an hour. It takes two to three days to reach 1,200°C. As the coils heat up, tiny particles that pin the grains in place slowly dissolve, and about one grain in a million starts growing and swallows its neighbors. Then the coils sit at 1,200°C for 20 to 40 hours. Grains that started 0.01 mm across end up 10 mm wide, nearly all pointing the same way. That alignment cuts the power a transformer core loses as heat. Cleveland-Cliffs can supply about 12 to 20% of what the US needs, by DOE's estimate. China was in a similar spot before 2000: its only volume producer covered 30% of demand. The few foreign companies that could make the high-grade version refused to license their process. Baosteel spent ten years on research and made its first good coil in 2008. By 2013 it could make 300,000 tons a year, three times what all of China made before 2000. I build chemical plants. If you're in hard-tech, chances are you are now competing with AI data centers for the same grid equipment.

Every two days an ethane tanker sails from Texas to Chinese crackers that were designed around American ethane, carrying nearly a million barrels. Hormuz closed on February 28. Before that, more than half of the naphtha Asia imports, the oil-based feed most Asian crackers use, passed through it. Naphtha crackers across China cut their rates. To feed its ethane crackers, China imported 1.14 million tonnes of ethane in April, the most ever, more than 96% of it from the US. In early April, ethane crackers in Northeast Asia earned their best margins since 2018. A cracker heats its feed to about 850°C for a fraction of a second. About 80% of the ethane ends up as ethylene, the building block of polyethylene. Naphtha comes out 30 to 35% ethylene plus propylene, butadiene and gasoline-range liquids. An ethane plant has no units big enough to separate those. No other country ships ethane by sea in volume. In June the US exported 770,000 barrels a day, the most since EIA started tracking it in 2014. I manufacture chemicals in Houston. Hormuz is under attack again. And China's ethane crackers are still running. On Texas ethane.

USAntibiotics is the only company making amoxicillin pills in the US. They supply about 8% of US amoxicillin. India is the biggest foreign supplier of finished antibiotics to the US. For over a decade, India's amoxicillin started with Chinese Penicillin G. Chinese plants grow it from a mold fed with sugar in large tanks. Penicillin G is turned into 6-APA, the backbone of amoxicillin. China makes most of the 6-APA India imports too. Indian drugmaker Torrent stopped making Penicillin G in 2005 after Chinese imports undercut it. By 2012 the last Indian plants, at Alembic and SPIC, had shut too. Aurobindo restarted Penicillin G at Kakinada in 2024. In 2025 China cut its price about 40 to 50 percent. India answered with import price floors on Penicillin G, 6-APA and amoxicillin. By January 2026 the plant ran at about 9,000 to 10,000 tonnes a year, roughly what India uses. In October 2025 the US Pharmacopeia found that amoxicillin made for the US depends on four starting materials, each made almost entirely in China. The last American plant making penicillin ingredients stopped in 2004.

Drip molten glass into cold water and you get a Prince Rupert's drop. The water cools the outside first, and it hardens into a shell while the inside is still molten. When the inside cools, it shrinks and pulls on the shell. The surface ends up compressed at up to 700 MPa, about 10 times the minimum for tempered glass. The core is left under tension. A crack can only grow where glass is under tension, so cracks in the surface (every glass surface has microscopic flaws) cannot spread inward. But, if you break the tail, a crack can cross the compressed shell and reach the core, where the tension drives it forward at about 1,700 meters per second. Corning compresses the surface of its phone glass from the other side. Gorilla Glass contains sodium. Corning puts the finished sheet in molten potassium salt at about 400°C, and sodium ions leave the surface while potassium ions move in to take their place. A potassium ion is about 35% wider than a sodium ion. At 400°C the glass is too stiff to make room, so the bigger ions crowd the surface and leave it compressed at 650 MPa or more. Gorilla Glass has been designed into more than 8 billion devices. In Gorilla Glass 3, the compression layer is about 0.05 mm thick. Corning also has to keep the tension in the core low, or a cracked screen would break into many small pieces. Its scientists test this by breaking 50 mm squares of glass and counting the pieces. youtube.com/watch?v=bmUSaDGC…

20 of Germany's 50 richest people were born in 1965 or later. All 20 inherited their money (Forbes). In the US top 50, 10 of the 13 born in 1965 or later built their own companies. Twelve of the 20 Germans are Boehringer heirs. The family owns Boehringer Ingelheim, a drugmaker founded in 1885 that sold €27.8B of medicine last year. Most of the chemicals industry looks like this. BASF is 161 years old, Dow 129, and DuPont 224. They were founded when the basic reactions were discovered. In the 1930s, a job at DuPont was what a job at Google is today. Peter Thiel says chemical engineering hasn't been a field worth studying since the 1980s. In 2017, investors told us chemicals manufacturing was boring. YC partners said "do not expect to raise any meaningful money. No one understands what peroxide is." @TungstenSeanide and I built our first reactor from Home Depot parts. Our first sale was one gallon of peroxide in the back of Sean's Subaru. The companies founded in the 1800s were built on the first wave of chemistry. The second wave is here. youtu.be/B7yl7fEHeKM