The first story is that the real economy is rebuilding. September industrial production came in ahead of expectations and the gain was broad, with utilities surging and manufacturing led over the past year by computers, electronics and aerospace (Daily Shot, 10/19). Semiconductor output rose at a 28.4% annual rate in the third quarter, and capacity utilization rose to 76.9%, still below its long-run average (Federal Reserve, 10/16). Nonresidential construction jobs are at a record high (10/21). That isn’t a consumer boom. Consumer goods output kept contracting (10/19). It’s power, defense and hardware.

The second story is that the chemistry underneath all of that is getting squeezed, and squeezed hard.

Subscribe now

The build is real, even when the stocks aren’t

AI stocks had a rough week. On 10/23 the Nasdaq 100 had its worst drop since June and chip stocks fell 6.7%, with money running to Treasuries (10/23). The Daily Shot noted that all parts of the AI theme have been correcting for roughly five months (10/22).

And yet the spending hasn’t slowed. Hyperscaler capex has climbed faster than in any earlier investment boom, including canal mania, the railways, the Roaring Twenties and the dot-com era (10/22). Capex estimates are still rising (10/21). Asian chip and hardware earnings forecasts keep going up (10/20). Japanese chip-equipment makers are expected to grow strongly (10/23). Malaysia’s exports hit a record on AI-related electronics (10/19). Data-center hiring in Europe is accelerating while other tech hiring stays soft (10/20). US import prices for computer parts are still up nearly 50% year over year (10/19).

The practical reality is that the build keeps going whether or not the multiples hold. How it gets paid for matters too. J.P. Morgan estimates only about 20% of 2027 AI capex will be funded from free cash flow, with roughly 80% needing outside financing (10/26).

Every one of those servers, fabs and substations runs on materials and chemistry: high-purity process chemicals, specialty polymers, adhesives, coatings and thermal materials. That layer barely shows up in the AI capex conversation. It’s the most under-discussed part of the stack. It may take a while to show up in the numbers, but it’s there.

The squeeze: energy and China

Now the other side.

Energy is pushing input costs up almost everywhere. US gasoline is up about 40% from a year ago, to more than $4 a gallon, as the Iran conflict pushes oil higher (10/19). Euro area inflation hit 3.8%, led by energy (10/19). German producer prices surged (10/20). Canadian producer prices jumped (10/23). Japan’s manufacturing PMI showed input-cost pressures surging (10/23). US manufacturing price pressures jumped in the S&P Global PMI (10/26). Brent averaged about $124 a barrel in October, up from about $65 a year ago (EIA).

If you run a specialty chemicals business, that’s your raw material and utility bill. Some of it passes through. Some of it doesn’t.

The squeeze is on margins, not volume. US chemical output rose in September (Federal Reserve, 10/16). You can see it in the earnings. LyondellBasell missed third-quarter estimates on a polymer squeeze, guided its European olefins plants to about 60% operating rates (near technical minimums) and said US ethane is running about 24% above its third-quarter average (10/30).

Then there’s China. Q3 GDP picked up to 4.5%, and industrial production was led by chemicals and nonmetal products (10/19). But Chinese industrial capacity utilization sat at 73.1% in Q3, down from 74.6% a year earlier and near a two-year low (China National Bureau of Statistics, 10/20). So China is producing more chemicals into weak domestic demand, and that output has to go somewhere. The euro area’s goods trade deficit with China hit a record (10/19), and Chinese battery makers are expanding production in Europe (10/23).

Commodity chemicals in the West are going to stay under export pressure. But the policy response (trade protection) tends to help Western producers who make the qualified, spec-critical stuff that customers can’t just swap out. The squeeze isn’t even. It falls hardest on undifferentiated volume.

Housing makes it worse for one slice

Interest rates are doing their own damage. Financial conditions tightened, mostly from surging long-term yields, and Goldman’s read is that they’ve gone from supporting growth to a modest drag (10/20). Homebuilder sentiment hit its lowest level in more than 14 years (10/20). Pending home sales fell to a record low (10/21). Residential construction employment has turned down (10/21).

So if your coatings, sealants or building-products volume is tied to new housing, I’d plan for soft demand into 2027. That’s a different business from the one selling into a data center or an aircraft program, even if the chemistry looks similar on paper.

Where the deals are

Here’s the through-line, as plainly as I can put it. The real economy is re-industrializing around defense, power and AI hardware. Chemicals and materials companies are getting squeezed by energy costs and Chinese overcapacity. The ones caught in the middle (good products with the wrong end-market mix, or a balance sheet built for cheaper money) are where a lot of the next few years of transactions will come from.

The supply of sellers looks like it’s building. The UK is a useful analog. Business sales there reached an eight-year high last year as founders chose to sell rather than scale (10/23). In the US, the lower middle market was the only private equity segment to grow in Q2 2026, with buyouts of $25 million to $100 million up 56% in deal count and 71% in value quarter over quarter, to 413 deals worth $16 billion (10/19). Meanwhile only 28 new US private equity firms have launched this year, versus more than 100 in each of the past five years (10/23). More sellers, steadier buyers, fewer new funds. That reads like a market where knowing the chemistry and the end markets is worth more than it used to be.

For an owner, the questions I’d ask are simple. What share of my volume goes to the build (power, defense, semis, data centers) versus housing and consumer? How much of my energy cost can I actually pass through? And am I making something a customer has to qualify, or something they can buy from anyone?

If you’d like to put numbers to the feelings on any of those, I’m happy to think it through with you. Challenge me on any of it. We’ll figure it out.

Share

Sources: The Daily Shot, editions of 10/19, 10/20, 10/21, 10/22, 10/23 and 10/26/2026, cited inline by date; Federal Reserve G.17 industrial production release (10/16); China National Bureau of Statistics (10/20); LyondellBasell third-quarter 2026 results (10/30). Charts by Curiously Optimistic from FRED data.

I’ve spent a lot of my working life around plants like these, and I kept noticing that buyers don’t pay for them the same way. So I started writing down why. This is that notebook, built from permits, filings, releases, and the deal tape, current as of October 2026. Where a number is soft, I say so. I might be wrong on any call in here, and I’d like to hear where.

Why I sort custom chemistry makers by skill

From the outside, fine chemistry, pharma contract manufacturing (CDMOs), and toll plants look like one market. They share reactors, chemists, and vocabulary. But one plant sells into several end markets, so the end market can’t tell two plants apart.

So I sort plants by the chemistry they’re permitted and equipped to run. If you’re valuing or selling a plant, compare it with plants that run the same chemistry, not plants that serve the same customer. End markets tell you where demand is heading. The skill and the permitted capability behind it tell you what a buyer will pay for the site.

One piece of good feedback I’ve received on this so far: permits are a good way to segment the market, but once you’re underneath them, assessing the capabilities of the team is a key area of further diligence, and probably where the long-term defensibility sits. I agree. The public record can tell you who holds the permit. It can’t tell you how capable the people running the chemistry are, so this guide is a starting point for that work, not the end of it.

The nine skills, and what’s scarce in each

• 1. Fluorine and halogens. The scarce thing is HF, F2 and SF4 handling. Eight named US makers, and none offers HF at custom scale. The money so far is capex, with no disclosed site sale.

• 2. Organometallics and reactive metals. Pyrophoric, air-free work. The one nearby multiple is seller-stated: Honeywell paid 13.3x for Johnson Matthey’s catalyst business.

• 3. Silicon chemistry. Chlorosilanes and hydrosilylation. Fab precursors are up and basic siloxanes are down, and there’s little independent left to buy.

• 4. High pressure, hydrogenation, alkoxylation. Pressure, hydrogen and ethylene oxide permits. One US independent went to an Indian platform in May 2025.

• 5. Specialty polymers, monomers, and CASE. Monomer chemistry and qualification. The one stated multiple is H.B. Fuller’s, under 6x after synergies for adhesive formulators.

• 6. Hazardous chemistry. Nitration, phosgene, cyanide and energetics. Orica paid $640m for Cyanco at 7.5x, buyer-stated.

• 7. High-purity inorganics and reagents. Parts-per-trillion purity and analytical grades. Fujifilm paid $700m for Entegris’s fab acids, while research catalogs are flat.

• 8. Licensed handling. DEA registration, containment and isotopes. Siegfried paid “below 10x” for Noramco’s US sites, its words.

• 9. Modality synthesis. Solid-phase synthesis and purification for peptides and oligos. Samsung Biologics offered CHF 44.31 a share for PolyPeptide, with no multiple stated. The offer was declared successful on October 16, 2026, and closing still waits on regulators.

My ranking: where the evidence and the deal record point

The capability and the qualified customer earn the premium, not the reactors. Here’s how I’d rank the nine for a buyer, and the comparable behind each call.

• Pick 1: a US hazardous-chemistry site with a live, permitted capability. Nitration, phosgene, cyanide, energetics. The comparables are adjacent: Orica’s 7.5x for Cyanco and NewMarket’s ~$700m for AMPAC. The risk case: one permitted US site filed a WARN notice in March 2026 and then sold at an undisclosed price. The permit outlived the business that held it.

• Pick 2: a Western fluorination or fluorine-reagent site. No site sale has printed a price yet. The evidence is capex, and Indian HF could undercut the US reagent step before a first comp prints.

• Pick 3: DEA-registered or high-containment capacity in the US. This is the one direct comparable: Siegfried’s “below 10x” for Noramco’s US sites.

• Tied fourth: organometallics and high-pressure independents with Western customers. Grace stated 9.5x for Albemarle’s fine chemistry sites. The independents sold at undisclosed prices.

• Watch: silicon, polymers and CASE, and high purity. Almost nothing independent is left in silicon, and I found no sale of a US monomer or photoinitiator house. There’s likely better opportunity downstream, in application and formulation.

• Avoid for mid-market entry: peptide or oligo capacity without a contract book. With a contract book the economics work (Bachem’s FY2025 EBITDA margin was 30.9%). A mid-market buyer can’t get that book.

• Pass: capacity sold mainly on reactor time, with no permitted capability or qualified customer behind it. EUROAPI sold its Brindisi plant at €5m of enterprise value.

Ten years of deals, and only ten stated multiples

I logged 60 deals across the nine skills from 2016 to 2026. A buyer or seller stated a multiple on only ten of them, and I print a multiple only where a party to the deal stated it, with who said it. The smallest US multipurpose-site print was about $134m of enterprise value on about $78m of 2024 revenue and about $15m of EBITDA. That’s about 1.7x revenue (my arithmetic on the buyer’s figures) and about 9x EBITDA (the buyer’s figure). It’s one deal, so I wouldn’t build a range on it.

Where the profit is

Where a listed company breaks out a skill, the capability-heavy work shows the best margins. Dottikon’s hazardous chemistry printed a 34.6% EBITDA margin and Chemours’ refrigerants and fluids 32%, against 5.8% core EBITDA at EUROAPI’s generic API business and 4.0% operating margin at Stepan Surfactants. These aren’t like for like, since group and segment margins blend businesses. The pattern holds anyway. No US independent publishes a margin, so the private middle has to be read from the deals and the permits.

What’s in the full guide

The full guide runs 114 sheets plus a source appendix. Each of the nine segments runs in the same order: opener, size, why it earns more, how they get paid, who plays, deals, what to watch, and my call. Then the back matter: every deal under and over $1bn on one tape, what I’d do by reader (sponsor, owner, banker), the dates between now and next spring that would change my mind, the method, a glossary, and a source ledger with every number dated and linked.

The preview below is ten sheets of the full guide, free to download, and every sheet number matches.

Download the preview (PDF, 10 pages)

Or explore the preview as an interactive guide (free): field-guide.augustgraceadvisory.com/custom-chemistry-makers-preview

Read more

The short version. US factory construction peaked near $250 billion a year in late 2024. In August it was down about 20% from a year earlier, and almost all of that is one category: chip, battery and electrical plants, down 45%. Everything else is up about 1%. Over ten years, US manufacturing output hasn’t moved. Adjusted for construction costs, we’re building about a third less chemical plant than in 2016, and most of what we are building is drug plants and big commodity units. The starting materials, high-purity chemicals and explosives precursors that the new fabs, drug plants and munitions lines need are still mostly made somewhere else.

I’ve been asked some version of the same question a lot this fall. Is reshoring real?

Here is the honest answer. We reshored the buildings. We didn’t reshore the chemistry underneath them.

Subscribe now

The factory boom was one boom

Start with the number everyone’s quoting. Private factory construction ran $168 billion at an annual rate in August, down about 20% from a year earlier (Census, October 1). That sounds like reshoring stalling. It’s more specific than that.

Nearly all of the decline sits in one line: computer, electronic and electrical plants. That’s the chip fabs and, by the Census definition, very likely the battery plants. The line peaked at $126 billion in June 2024. It was $52 billion in August. Take it out and everything else is up about 1%.

So reshoring didn’t stall across the board. The fab and battery wave crested, it’s rolling off, and nothing that size is starting behind it. The money went somewhere else. Data center construction hit $85 billion at an annual rate in August, up 73% in a year. In early 2018 it was about $6 billion.

Ten years, and output hasn’t moved

A step deeper, and it gets more uncomfortable. The Fed’s index of US manufacturing output was 99.2 in August 2016. It was 99.1 this August. A decade of tariffs, CHIPS, IRA and announcements, and we make about the same amount of stuff.

Manufacturing employment is 21,000 below January 2025, though it’s turned up for four straight months (BLS). The Reshoring Initiative, which tracks this better than anyone and is firmly on reshoring’s side, counted 244,000 announced jobs in 2025. Announcements and payrolls aren’t moving together.

Further, among manufacturers that reshored or are doing it now, the share satisfied with the results fell from 96% to 65% in a year (2026 USA Reshoring Survey). Those aren’t skeptics. Those are the people who did it.

That doesn’t mean the buildings were a waste. Capacity takes years to become output and plenty of those fabs aren’t running yet.

Chemistry had its boom a decade ago

Chemicals is the line people point to as the exception, and on the surface it is. Chemical construction was $43.7 billion at an annual rate in August, up about 6% from a year earlier, while most of what’s around it fell. It’s now about a quarter of all factory construction.

Adjust for what it costs to build and it looks different. The government’s price index for new industrial construction is up 74% since 2016. In today’s dollars, chemical construction ran about $66 billion a year in August 2016, at the tail of the shale cracker wave. It’s about $44 billion now. Back then chemicals was nearly half of all factory construction. We’re building about a third less chemistry than a decade ago, not more. (The index is for buildings, not process equipment, so call it rough.)

A fair objection: 2016 was a peak related to natural gas and energy. Against 2021 we’re up a little. Either way, chemistry never got a reshoring boom of its own (at least not yet).

Two numbers cut the other way. Chemical employment was 905,500 in September, up 11% since August 2016 on flat output (BLS via FRED), and that’s almost all pharma: ACC counts about 545,000 chemical workers outside pharma, and that number’s been easing (~5,500 works per year). And US chemical imports excluding pharma fell almost 10% in 2025, with the trade surplus still in place (ACC mid-year outlook). Neither one says we built the layer underneath. More people making the same output, mostly drugs, and fewer imports of chemicals we already make plenty of, isn’t reshoring. Drug imports tell you where the gap is: they’ve more than doubled since 2016 (UN Comtrade via Trading Economics).

And the plants we already have aren’t full. US chemical capacity is up about 5% since 2016. Output is up about 1.5%. Utilization was 71.7% in August, the second-lowest reading in ten years (Fed G.17). The American Chemistry Council expects US chemical volumes to grow half a percent this year, with specialty chemicals down (ACC).

Open the box: it’s drugs and crackers

Then there’s what’s inside the number. The government’s definition of chemical manufacturing includes pharmaceutical plants, and the Census chemical line is classified the same way. So when Lilly breaks ground on a $6.5 billion API plant in Houston, as it did last month, it lands in the chemical bucket. So does the $8.5 billion ethylene cracker in Orange, Texas that’s commissioning now.

You can see it in output too. Since 2016, US pharmaceutical output is up 18%. Basic chemicals output is down 7.5% (Fed G.17). Inside “chemicals,” it’s the drugs that are growing.

When I added up the large US chemical-sector projects I could find with a dollar figure attached, drug plants were the majority of the money, then crackers, chlor-alkali and ammonia. Electronic-grade, specialty and explosives projects together came to about 5%. That’s my own tally, not an official number, and it leans toward big announcements. But it isn’t close.

The layer underneath isn’t getting built

This is the part I care about, because it’s where I spend my time.

Drug starting materials. USP found that 41% of the key starting materials behind US-approved drug ingredients come solely from China. Brookings put it plainly: making the active ingredient here doesn’t change the risk much if the starting material still comes from China. The new drug plants will need those inputs from roughly 2028 on. Very little of that capacity is being built here.

Fab chemistry. CHIPS put most of its materials money into wafers, polysilicon and substrates. The one award I can find for an actual wet chemical is a preliminary $52 million for ultra-high-purity isopropyl alcohol in Baytown, Texas. The fabs are being built. A lot of what they run on still doesn’t come from here.

Explosives. The Army awarded a TNT plant in Kentucky in November 2024, the first US TNT production since 1986. First output was supposed to come in about three years. The company now says early 2029.

Why it doesn’t get built

I don’t think it’s money. Finding the money is easy. It’s the demand side at the price point.

The world isn’t undersupplied. China and India make most of these molecules, they make them well, and they make them cheaper, often a lot cheaper. You just don’t like where it’s made. Asking a US plant to build redundant capacity at a higher cost isn’t really a capitalistic decision. It’s protection, or a backup plan, and somebody has to pay for it.

I can’t put a plant in the ground because you want it. I need an order: a price I can get a payback on, and a volume I can count on for five or ten years. A tariff alone usually doesn’t do it. Most buyers will eat the tariff and pass it along.

Washington hasn’t closed that gap. Everybody says the word reshoring, but it’s not specific enough. The policy hasn’t followed the value chain down to the precursor at step five, so nobody actually knows what they need. And the path of least resistance is to do nothing. I’d start by mapping the precursors under each critical drug, chip and munition, then pay for the order, not the building.

There’s a permitting piece too. In an August survey, SOCMA said 70% of the members it surveyed had cancelled projects because EPA’s new-chemical reviews take too long. That’s a trade group making its case. It matches what I hear.

What AI has to do with it

Apollo’s chief economist, Torsten Slok, made a macro case in the Daily Spark on October 9 (AI’s Insensitivity to Interest Rates Is a Problem for the Fed): the AI build-out doesn’t care about interest rates and soaks up capital, power and labor while rate-sensitive industries get squeezed. His conclusion: the Fed may have to hold rates higher for longer. Maybe. I’m not going to make a call on the Fed.

My read is narrower. Money for a good small chemical plant isn’t the constraint. What AI pulls on is everything else: grid connections, construction crews, attention. If you’re AI-adjacent, you’re growing. Everything else in industrial chemical land is still kind of languishing. It’s two extremes. And a small specialty project is competing for the same electricians and the same power as an $85 billion-a-year data center wave.

Where I think this goes

This part’s judgment, not data. I might be wrong.

1. Fewer new US plants starting up from 2028 on. Construction leads capacity by two to four years, and outside data centers and power the pipeline is shrinking.

2. Consolidation, not construction, in chemicals. Utilization near a ten-year low means more capacity than output. Expect rationalization, carve-outs, and buyers paying for plants that already run. Anupam Rasayan buying Jayhawk Fine Chemicals in Kansas this year is the pattern.

3. A starting-materials gap around 2028 to 2031, when the drug-plant wave comes online and needs inputs nobody built here. That’s an opening for qualified US plants, but only with an order behind it.

4. Data centers keep winning the fight for power and crews.

What I’d do differently Monday

If you’re underwriting in this space:

• Don’t read chemical construction as demand for the specialty layer. It’s mostly large volume drugs and crackers.

• A qualified, permitted, running US hard-chemistry plant is scarcer than the headlines suggest. It’s usually cheaper and faster to buy one than build one.

• Underwrite reshoring revenue only when there’s an order behind it, a price and a volume for five or ten years. Without that, it’s a story, not a forecast.

Where I might be wrong

The counter-case is real. Sunlit opened a $100 million high-purity hydrofluoric acid plant in Phoenix in October 2024 to supply the fabs, in a chemical McKinsey says the US imports almost entirely from Asia. Manufacturing payrolls are up 72,000 from their December low. And the share of contract manufacturers quoting reshoring work doubled in a year, from 16% to 32% (2026 USA Reshoring Survey). If those plants run and the orders keep coming, I’ll have been early to the wrong side.

But most of that’s foreign-owned suppliers following their customers here, and none of it reaches the starting materials.

I’d like to be wrong about this. If you’re running a US chemical plant that got the order, or you’ve watched one die on price, I’d like to hear about it. Challenge me.

Sources: US Census Bureau, Construction Spending, October 1, 2026. Federal Reserve G.17 via FRED (IPMAN, IPG325S, IPG3254S, IPG3251S, CAPG325S, CAPUTLG325S). BLS Employment Situation, October 2, 2026, and PPI for new industrial building construction. Reshoring Initiative, April 6, 2026 release and 2026 USA Reshoring Survey. American Chemistry Council Mid-Year 2026 Outlook. USP Medicine Supply Map, June 2026. Brookings. NIST CHIPS awards. Spectrum News, October 5, 2026. SOCMA, August 13, 2026. ACC Year-End 2025 Outlook. UN Comtrade via Trading Economics. Apollo Daily Spark, October 9, 2026. Inflation adjustments use the BLS PPI for new industrial building construction and are approximate.

If you only know the Raptor from photos, here’s what you’re watching. The U.S. Air Force (USAF) fact sheet, which is the service’s official public description of the jet, says it can supercruise. That means it stays above Mach 1.5 without lighting the afterburners, the fuel-guzzling kick most fighters need to go that fast. It’s Mach 2 class when it wants more. It also has thrust vectoring. The engine nozzles tilt, so the nose can point and turn harder than a plane that only steers with its control surfaces. I’m not trying to rank every fighter on earth. I just wanted to know what I was looking at.

Sitting there, I got stuck on the question I always get stuck on when something that complicated looks easy. Who figured out the metal? Who made the resin? And how do you keep hundreds of small shops qualified long enough to build the thing, and then keep recoating it every time someone takes a panel off?

The names on the jet

Lockheed Martin, Boeing, and Pratt & Whitney get the credit, and they earned a lot of it. Lockheed was the prime contractor. It did final assembly in Marietta, Georgia, and built the middle section of the fuselage in Fort Worth. Boeing built the wings and the aft fuselage. Pratt & Whitney built the F119, the engine that makes supercruise possible, under its own separate Air Force contract.

That’s the tip. Those three companies design the jet, put it together, and answer to the Air Force when something goes wrong. What they mostly don’t do is make every part.

How three companies turn into a thousand

This is the part I had to work through, because it isn’t obvious unless you’ve spent time around manufacturing.

A prime’s real job is integration. It owns the design, the drawings, the final assembly, and the flight testing. Then, part by part, it makes a make-or-buy call. Build this in-house, or buy it from someone who already does it well? On a fighter, a lot of things land on the buy side. The prime could probably learn to do most of them. But some smaller shop has already spent years getting good at one process and holding the certifications for it, and rebuilding that inside a giant company would cost more and take longer.

So the work flows down in tiers. A Tier-1 supplier sells directly to Lockheed, Boeing, or Pratt & Whitney. That might be a whole assembly, a wire harness, or a run of machined brackets. Tier-1 suppliers buy from Tier-2 shops, the forgers and casters and composite houses. Those buy from further down, from the mills, the resin makers, and the people who build the tooling. Each shop works from the drawing in front of it. Most of the shops below the prime never see the whole jet.

And it doesn’t stop at delivery. Sustainment, which is everything it takes to keep a delivered fleet flying, keeps pulling repairs, spare parts, and coatings work through the same kind of chain for as long as the jet flies.

That’s how three logos turn into roughly a thousand shops. The primes hold the program together, and the shops make most of the pieces.

How many shops are we talking about

The public numbers from original production are a little soft, and they don’t all count the same thing. Here’s where each one comes from.

In 2007, Boeing put out a press release celebrating the wing set for the 100th F-22. Almost in passing, it said parts and subsystems came from roughly a thousand suppliers in more than forty states.

In 2009, the Congressional Research Service (CRS), the nonpartisan research office that writes background reports for members of Congress, published a report on the F-22 program while Congress was arguing about whether to keep buying the jet. It cited a Lockheed map showing about 1,040 suppliers in forty-four states. A map like that is partly a lobbying tool, since it shows lawmakers how much of the work lands in their states. It’s still a real count.

Go back further, to 2002, and Lockheed was announcing its “Raptor of Distinction” award, which the F-22 industry team gave to standout subcontractors. That year it went to EDO, whose North Amityville plant was the largest single F-22 subcontractor in New York. The release said the jet drew on about 1,200 subcontractors and suppliers in forty-six states, including roughly forty New York firms, about thirty of them on Long Island.

The set I find most useful is the USAF F-22A Production Restart Assessment (2017). The Air Force wrote it for Congress to answer a question that kept coming up after the line shut down: what would it take to start building F-22s again? To answer that, it had to look back at how the original jets got built. It counts about a thousand first-tier supplier locations, roughly seven hundred Tier-1 firms, something like ninety thousand unique parts across the air vehicle and the F119 engine, and about forty thousand tools at the major fabrication and assembly sites.

The counts differ because one counts locations, one counts firms, and one counts every subcontractor and supplier the company could name. They all land around a thousand, which is what I trust. The parts number is the one that explains the rest. When I try to picture ninety thousand different parts, I end up picturing thousands of tiny product lines, each with its own drawing and its own tooling. Nobody runs all of that under one roof.

One line in that assessment stuck with me. Before any restart, the Air Force would have to confirm that the sub-vendors were still willing to do the work and could still requalify. Today’s sustainment base is smaller and different from the production-era count, but it’s still a multi-tier specialty chain, and it lives with the same question. Can this shop still make the part the drawing calls for, to the same standard, years later? In this world, that’s what reliable means.

Small batches, not a car plant

Most of those shops were never Lockheed, Boeing, or Pratt & Whitney. They were high-mix, low-volume places, a lot of them owner-operated. A car plant wins by making the same bracket a million times. These shops make short runs of odd parts that have to pass a qualification packet, meaning the shop has proven with paperwork and test pieces that its process makes the part to spec. Then they have to keep passing it. Change the process and you may have to prove it all over again.

That’s also why the chain gets so wide. One shop knows a difficult alloy. Another knows one weld process, or one sealant family, or one kind of tooling. The titanium shops aren’t the composite shops, the composite shops aren’t the adhesives shops, and none of them are the people running coatings robots, because each of those needs its own equipment, its own quals, and its own people who’ve done it for years.

A few of these shops show up by name in Lockheed’s small-business supplier releases, which is about the only public credit they get. In Texas, Complete Technical Representation in Carrollton and InterConnect Wiring in Fort Worth did machined components and wire harness kits, and M2 Global is in San Antonio. McGinty Machine in Wichita did precision-machined F-22 parts, according to a Lockheed release reposted on Airframer, a trade site that tracks aircraft suppliers. Click Bond, in Carson City, Nevada, makes fasteners. And Fort Walton Machining on the Florida Panhandle started with two people in 1997, grew into a serious qualified shop, and was named Lockheed’s 2010 Small Business of the Year.

Most of the thousand never got a press release at all. To see what they actually do, it helps to follow the materials.

Titanium that had to get qualified

Materials tables published while the jet was being built, including in the materials-engineering trade press, put titanium at somewhere around two-fifths of the structural weight. That’s more than nine thousand pounds in the airframe, roughly the curb weight of three mid-size cars, in just that one metal.

Ti-6-22-22 is a high-strength titanium alloy that the F-22 was the first to use at scale. The name is shorthand for its recipe. It isn’t commodity bar stock, and it had to be qualified for this jet.

Then there’s how you process it. Hot isostatic pressing (HIP) takes a rough titanium casting and squeezes it evenly from every side, at north of ten thousand pounds per square inch, to close up the tiny voids inside. Published accounts of the program name big pieces like rudder actuator housings, canopy decks, and inlet frames. Electron-beam welding joins thick titanium aft assemblies inside a vacuum, so oxygen can’t make the metal brittle. The payoff is that you can get rid of something like three out of four fasteners in those joints. Fewer holes means less weight and fewer places for things to leak or crack.

None of that is a prime pouring some metal. It’s mill work, casting control, vacuum welding, and tooling, mostly in shops you’ve never heard of, and they have to stay good at it.

Why the skins are BMI, not “plastic”

When people hear “composite” they tend to picture fiberglass on a boat. That’s the wrong temperature range entirely.

The exterior skins are bismaleimide (BMI), a resin that cures hard and holds up at temperatures where ordinary epoxy starts to soften. That matters at supercruise, when the skin heats up at speed. The best public source I’ve found is a CompositesWorld article on high-temperature resins. CompositesWorld is a trade magazine for the composites industry, and that piece walks through which resins were built for heat and where they ended up. For the F-22, it names Cytec’s BMI. The skins use IM7/5250-4 prepreg, which is IM7-grade carbon fiber that arrives already soaked in 5250-4 BMI resin. And Cytec’s CYCOM 5250-4 resin went into nearly four hundred components made by resin transfer molding.

Resin transfer molding (RTM) works like this. You put dry carbon fiber into matched metal tooling, inject heated resin under pressure, and cure it tight. That’s nearly four hundred molded parts, every one of them held to fighter-jet tolerances, and every one needing tooling that somebody had to build.

That Cytec line was in Tempe, Arizona, back then. Ownership changed later. Cytec went into Solvay, and the specialty materials business now sits under Syensqo, so today it’s a Belgian-parent specialty chemical company with a US plant lineage. That’s common in specialty chemicals. What I like about it, and what I’d want a kid who’s into how things are made to know, is that the resin has a name, a process, a plant history, and a qualification trail behind it.

Every panel open is a coatings job

The metal and the resin get the jet built. Keeping it flying is a different job, and a lot of that job is coatings.

The F-22’s stealth comes from low observables (LO), the features that make it hard to pick up on radar and infrared sensors. Shaping comes first: aligned edges, canted tails, inlets that snake so radar can’t see straight into the engine, and weapons carried inside. On top of that sits a multi-layer coating stack you can describe without giving away any recipes. There’s a primer, conductive layers, radar-absorbing material (RAM), gap fillers, and an infrared-reducing topcoat. Lockheed names a couple of fixes it’s working on to cut down on coating damage, Mighty Tough Boot and Form In Place. The companies that formulate the coatings aren’t named.

Lockheed’s F-22 sustainment page is where Lockheed describes the support work it does for the fleet, so read it as the company describing its own business. It says roughly half the maintenance on the jet is fixing stealth coatings that get damaged whenever a panel is opened for routine work. I’d have guessed engines, or avionics. So every time a panel comes off, somebody has a coatings job to do afterward.

This is also where a small company can matter without owning the chemistry. The Small Business Innovation Research (SBIR) program is how federal agencies pay small companies to solve specific problems. Through it, Aerobotix, a robotics firm in Madison, Alabama, won Air Force work to build multi-axis robots that reach deep into F-22 engine inlets and restore the specialized coatings at the depot, the Air Force’s heavy-maintenance base at Hill Air Force Base in Utah. The Air Force wrote up the project in 2017. Before robots, picture a maintainer in a Tyvek suit crawling up a twisting duct to do it by hand.

What we can say is American

The primes and final assembly are American, and the jet itself is US-only, since export is prohibited. Specialty metals fall under a domestic-preference rule in defense contracting, so titanium and specialty steels are generally melted or produced in the US or a qualifying country. That’s a rule about preference, though. It isn’t a map showing where every billet was melted. The BMI has a US plant lineage inside a global specialty-chemical parent. The LO coating formulators are opaque in the public record. And there are named US shops doing the application work.

So the honest answer is mostly American, with a few parts I can’t see into, and I’d rather say that than pretend I know more.

Strip the logos

If you took Lockheed, Boeing, and Pratt & Whitney off the top, what would be left?

Mills, HIP casters, electron-beam weld shops, BMI resin lines, RTM tooling houses, sealant makers, coatings floors. In the production-era count, that’s about a thousand first-tier shops across more than forty states, all making short runs of odd parts that had to stay qualified. I think that’s what people actually mean, or should mean, when they say “fighter industrial base.” I was thinking about all of them while the Demo Team came back around over Lake Erie.

I might be wrong about parts of this map. If you’ve worked in one of these shops, or you know where I’ve got it off, reply and tell me. And if you want the next one, subscribe to Curiously Optimistic.

Leave a comment

This is that continuation post…the so what…

Capital didn’t pause. It got a filter. The PE stats everyone quotes ($20B across 278 deals in 2021, roughly $5.9B in 2025, per C&EN citing S&P) aren’t a freeze. C&EN’s own read is that deal size fell more than deal count did. Look back about two years and the same filter shows up in four places on the value chain. Look forward 12 to 24 months, and I think it tells you where the activity will be.

Look back: four structures, one filter in each

1. Commodity: forced sellers, select buyers

Europe’s olefins and polyolefins weren’t waiting for a soft landing. High energy, regulatory and wage costs, Chinese capacity, soft auto and packaging demand. Sellers weren’t optimizing multiples. They were gett…

Read more

I think this supply chain shock in the Middle East takes longer to clean up than most people expect, so perhaps we’re not really hitting our stride until Q4. I’d call it a flat-to-down year on volume, but a building year on conviction and pipeline quality.

And I was bias a negatively on the institutional investing market:

The room is full of smart people writing memos about why they can’t act. Tariffs. A war that shut down the Strait of Hormuz. An oil shock. A helium shortage. AI disruption.

Next week I’ll take a real crack at where chemicals and materials go from here. Grading myself publicly, for “fun”…

What I was actually saying [in March]

The above remains the headline, but I thought then, as I do now, that this is much bigger than energy. It really is a supply chain reset:

This isn’t just an energy shock. It’s a chemicals …

Read more

Most organizations bought AI and just bolted it onto their old org chart. The productivity gains are not showing up for the same reason they did not show up in the 30 years factories kept the steam-age floor plan after the electric motor arrived: the technology is new but the architecture and processes around it are not.

This is the awkward middle.

The shift is really about a new kind of participant at your table rather than a new tool on the table. As a result, you should be thinking about the redesign of your team layer by layer, and you might just be about to cut the exact group that would enable your success.

For over 30 years after the electric motor arrived, factories kept the layout of the steam age. There was still one big engine, a tangle of line shafts and belts running overhead, and every machine crowded close to the power source because that’s how you fed them. Owners bought the new tool and bolted it onto the old architecture. The productivity gains everyone expected didn’t show up. A generation of managers later stopped asking “where do I put the motor” and started asking “what could (should?) this layout look like if my primary constraint was throughput, not power distribution.” With the benefit of power arteries and flow, the productivity promised was finally realized and these leaders ran the next fifty years of industry.

I think we might be living in an equivalent stretch. Most organizations have the new technology, but they just bolted it onto yesterday’s architecture. Robot agents, fresh out of the box, sit next to the work the way those first electric motors sat next to the line shaft. We are in the awkward middle, and it helps to name it as a middle, because middles end.

The shift underway isn’t really about AI as a technology. It’s about what happens when a genuinely new kind of participant joins the way work gets done. “AI replaces jobs” is the wrong frame just as “AI is just another tool” is too. Agents don’t replace people, and they don’t stay politely in the tool layer where software has lived for 40 years. They show up as something in between: a new set of hands, a new rhythm, a new perspective at the table. A new participant forces every organization to answer questions it hasn’t had to answer since the factory system produced the first professional managers 200 years ago.

Every time humans invent a new way to coordinate work, 3 things commonly happen. First, the organization restructures around it. Then, a new managerial class emerges; but only as the prior generation retires. That was true when the master-apprentice guild gave way to the factory, when Ford turned people into standardized components, when Drucker named management as a discipline and the paper-pushing, orchestrating middle layer exploded, and true again when reengineering and ERP software did the translation work and thinned that layer out.

The pattern is familiar but perhaps this time we’re facing a different pace. Scientific management held for more than 39 years. The reengineering wave held for maybe 15. This transition gets 5 (maybe 10…ChatGPT as we think of it was 2023). Cycle time is compressing, which means the penalty for redesigning late is compressing too.

The inversion nobody put in the brochure

For 3 decades, business software taxed the worker to benefit the organization. Salesforce, Workday, SAP, pick your acronym. Every one of them was, at the point of use, a compliance problem. Somebody at the top wanted visibility, so somebody at the bottom spent Friday afternoon typing what they’d already done into fields designed for a dashboard they’d never see (and fearing it was information that no one would ever act on). Value flowed UP. The cost flowed down. We called it adoption, and you spent fortunes on change management trying to force it and convince everyone it was a good idea.

I think the robot agent could invert that. Done right, the system now serves the person doing the work instead of the other way around. It captures and adapts to what they already do rather than asking them to re-enter and adapt to it (where is that toolbar?). It can interview a salesperson at the end of the day and write the notes for them. The compliance problem doesn’t get solved. It disappears, because you’ve stopped asking anyone to comply. When the tool finally works for the worker and not against them, adoption stops being something you enforce and starts being something people reach for.

The second change compounds the first: the cost of adopting and adapting collapses. A 1995 ERP install took capital, a team of consultants, and two years. A 2026 agent stack takes a subscription and a weekend. That sounds like an IT footnote. It isn’t. It means the two-person shop with a well-built agent fabric can now do work that used to require the two-hundred-person incumbent, and can do it before the incumbent has finished scheduling the steering-committee kickoff. Advantage used to accrue to whoever could afford and harness the biggest system. Now it accrues to whoever redesigns fastest around an effective one.

Small teams operate at a scale that would have been laughable five years ago. Large ones move at the speed their approval process always allowed. That gap is the whole story.

Invitation, not fear. Perspective, not bias.

If the mechanics are changing, the leadership job is changing more. It is mostly a job of framing.

Every organization gets to choose the emotion it puts around adaptation, and it matters much, much more than your tool selection. Fear of replacement produces exactly what you’d expect: hoarding, sandbagging, quiet resistance, people protecting the parts of their job that a machine could do because those parts are what they think they’re paid for. Invitation produces the opposite: experimentation, disclosure, people showing you where the pain actually is.

Perhaps the agent takes away the work people never actually wanted and gives back the work they came here to do. The salesperson who hated the CRM gets to sell. The analyst who drowned in formatting PowerPoint gets to pattern-match. The transition is something people pull toward instead of brace against. This is leadership’s real task in this era, and it isn’t explaining the technology. It’s inviting people into a different, less constrained, and better (read: more valuable) version of their own work.

Further, a companion mistake is treating the agent as a bias, untrustworthy (it hallucinates), something to be contained. A lazy version goes: humans are biased, machines are objective, so let the machine check the human. That’s wrong in both directions. What a good leader actually works with has never been objectivity. It’s a collection of perspectives: the operator’s read and the banker’s read and the founder’s read and the customer’s read, each one partial, each one valuable, none of them neutral.

Your agent is simply a new perspective to add to that table. Not a flawless one. Not a bias to contain. A useful one to integrate. Leaders who treat it as bias to manage end up playing defense and inducing fear while leaders who treat it as a perspective to integrate are playing offense. In both cases the edge that matters is the same one that always mattered: the quality of the synthesis, decision, and impact to the customer. The reality is machines, digital or not, make good synthesis cheaper to feed and more valuable to possess.

What does redesign look like layer by layer?

Redesign lands differently at each level of the org chart, and “we’re becoming an AI-first company” is not a plan. The CEO’s job narrows to the thing only the CEO can do: own the value narrative and set the priorities. In a world where raw capability is cheap, someone has to be able to answer what the customer is actually getting from us that they can’t get somewhere else. The failure mode is the chief executive who hands the whole thing to a newly minted AI officer and then can’t answer why we’re doing “this” and what “success looks like” in the next board meeting.

Maybe your executive team should just own the redesign in their functional areas. Each segment runs its own tidy pilot in its own silo. Every pilot of course succeeds on its own terms and definition. Nobody owns the cross-functional workflow where the customer value actually compounds. Hooray…the pilots pass but the company doesn’t move.

Then there’s middle management, and I think this is the layer people oversimplify most. The popular story says middle managers were information plumbing and the plumbing is now automated, so cut the layer. But plumbing was maybe a third of the job. The rest was coaching, judgment, conflict, hiring, and holding the tacit knowledge that never made it into any document. That part doesn’t vanish. It gets smaller in volume (maybe) and higher in value (definitely). The great middle manager doesn’t disappear. They get a broader span, a higher bar, and a title that shifts from Manager of X to something closer to Operator of X. The failure mode is flattening by layoff instead of flattening by redesign: cutting the layer to hit a number and discovering you removed the people who actually made the work cohere.

Finally, I think the junior layer is where the real mistake is about to be made. The instinct is to hire fewer of them, because agents can now do entry-level work. That is a five-year win and a fifteen-year disaster. Judgment is built by experience. The diligence analyst who has never read a terrible CIM cannot smell a terrible CIM. Automate away all the entry-level reps and you also automate away the apprenticeship, and in a decade you have no one who experienced how your business runs.

The right move isn’t just hiring fewer juniors. It’s deploying them differently. Fewer of them, perhaps, but higher autonomy, increase expectation in delivery (move the ball farther…), engaging and running agent teams from day one, rotating deliberately through the judgment-dense work that actually builds experience, and increased touchpoints from leaders who provide real feedback (with clear expectation of adaptation).

The apprenticeship pipeline is the real risk in this transition. It shows up last on the P&L, hurts the most, and will require your leadership teams to lead differently.

The test, and what doesn’t move

One question tells offense from reaction: what is our customer getting from us in 2030 that they can’t get now, and what operating model do we need to best deliver it? A clear answer is offense, and it compounds for a decade. Silence, inconsistency, or a list of the AI tools doesn’t. The quality and alignment around that answer are the best predictor of your success.

Transitions make people forget the “real” objective, so anchor them:

1. Customers buy outcomes, not org charts.

2. Trust compounds slowly and breaks fast.

3. Judgment and leadership matter most at the edges, where the playbook and process are not yet written.

4. Every meaningful transition carries loss: some skill that people invested twenty years in will matter less than it did, and leadership that pretends otherwise forfeits its credibility. Naming the constants isn’t nostalgia. They’re the fixed stars you navigate by while you rearrange the floor.

We rearranged the factory floor once we stopped treating the motor as a thing to place next to the old machine. Don’t redesign your shop floor around last year’s constraint.

If you are redesigning around this right now, how is it going? What are you stuck on? What’s working? What’s not? What am I missing? I’d love to learn from you.

In February 1915, Arthur D. Little told the Chamber of Commerce of the United States that America should not build a synthetic dye industry.

He was the most respected consulting chemist in the country. His argument was arithmetic. Even if American producers took every pound of domestic demand, he said, it “would amount to only about $10,000,000 annually, a little more than the candy sold by the Woolworth stores.” Why, he asked, “should we duplicate them only to plunge into an industrial warfare against the most strongly fortified industrial position in the world?”

Five years later the Senate Finance Committee answered him without meaning to. American dye consumption ran about twenty-five million dollars a year, the committee reported, “but these dyes were absolutely necessary in industries producing $3,000,000,000 worth of goods annually.” In the early war years, it added, “400,000 people might have been put out of work for want of dyestuffs.”

Little had the market size right. He had the question wrong.

The short version, for anyone who wants it before the story. In 1913 the United States made 1.8 percent of the world’s dyes. Germany supplied roughly nine tenths. In 1919 Washington seized more than 4,500 German chemical patents and sold them to an industry-owned foundation for $271,850. Only 699 were ever licensed, and the lift in American invention took over eight years to appear. The synthetic rubber program took 41 months from first proposal to production at scale; individual plants took 9 months. Today, the Defense Department has published two lists of chemicals the country cannot make, 22 in 2022 and 28 in 2024, with no purchase commitment behind either. Qualifying a second source for an existing energetic material takes 18 to 24 months. A new material in a fielded system takes three to five years. The Defense Production Act authority funding most of this expires September 30, 2026.

I went back and read those documents because of a number from last summer.

What Washington paid for a rare earth

In July 2025, the Defense Department put a floor of $110 per kg under MP Materials’ neodymium praseodymium oxide for 10 years, paying the difference whenever the market falls short. $400 million in preferred equity, a warrant that together with the preferred comes to 15% of the company, a 100% offtake on one new magnet plant’s output, and a guaranteed floor of $140 million a year in EBITDA.

Call it a subsidy, but what makes it unusual is the direction things moved: price and volume risk onto the federal balance sheet, equity coming back the other way. That is closer to project finance than to a grant.

Which is the only question that has ever really been at issue here. Not whether the country wants the chemistry. Who carries the risk (and benefit) of making it.

Washington has done versions of this before, and the last serious one is worth remembering. The Synthetic Fuels Corporation guaranteed prices and engineered offtake for synthetic natural gas in the early 1980s, hundreds of millions of dollars of it. Congress abolished the corporation in 1986, when energy prices came back. A ten-year floor is only as good as the politics underneath it.

So: how often has this happened, and what did it actually buy?

The patents did not work

By 1913, the United States made 1.8 percent of the world’s dyes, and American “manufacturers” were mostly assembling German intermediates shipped in. When the blockade closed, sulfur black went from twenty cents a pound to as much as three dollars inside a year.

Washington seized more than 4,500 German chemical patents under the Trading with the Enemy Act and sold them to an industry-owned entity, the Chemical Foundation, for $271,850. Francis Garvan, the Alien Property Custodian, told the National Cotton Manufacturers’ Association why it mattered: “Useless your armies and your navies, your U-boats and your aeroplanes, unless by means of a developed dye industry you keep abreast with modern chemical warfare.”

Then, in the same 1919 volume making the case for the seizure, A. Mitchell Palmer wrote this about what had been taken: “The price thus paid was necessarily determined somewhat arbitrarily; the great majority of the patents were presumably valueless.”

He was right, and it has since been measured. Of the 4,500-plus patents, 699 were ever licensed, to 326 firms. Royalties came to nearly $700k. There was a real lift in American chemical invention, roughly 20 percent, but it does not show up until 1927 and the full effect does not arrive until about 1932.

Over eight years later…

The buyer at the 1918 Bayer auction paid $5.3 million for all the legal rights and then went back to Germany to buy the know-how. DuPont held the seized patents and at the same time paid four German chemists $25,000 a year to come to Delaware. The patents were a map and nobody was short of maps.

Every chemical factory is a potential arsenal

Three separate 1919 sources converged on the same metaphor, independently. That convergence is the evidence.

Maj. Gen. William L. Sibert, Director of the U.S. Chemical Warfare Service, before the Senate Finance Committee in December 1919: “I have been very much impressed during the war by a statement which I read somewhere that every great chemical industry was a potential arsenal.”

An Allied Governments commission report, read aloud by Garvan: “In the future it is clear that every chemical factory must be regarded as a potential arsenal, and other nations cannot, therefore, submit to the domination of certain sections of chemical industry.”

A. Mitchell Palmer, on why the adjacency was technical, not rhetorical: “In the production of sulphur black, one of the most important black dyes, a slight variation in the final step of the long and complicated process of manufacture will transform the ultimate product into picric acid. More important still, however, is the fact that the technical skill required for the manufacture of explosives is precisely that possessed by the chemical staff of a successful dye works and is to be found nowhere else.”

None of those three had read the others. They arrived at the same sentence because the physics of the plant makes it true. A dye works and a munitions works are the same building with a different reagent in the last vessel.

That is the argument for caring about the $25 million dollar business that gates the $3 billion market. It was true in 1919. It is true now, at a different scale, for a different set of molecules.

The chemistry was never the slow part

The synthetic rubber program is one everyone cites, and the primary documents are better than the legend.

The Baruch committee was honest about the ask in September 1942. Having lost most of the natural rubber supply, “we must have created, within two years after Pearl Harbor, one of the largest industries in the country. Normally such a development would require a dozen years.”

They mostly did it, with rationing and reclaimed rubber covering the gap. Fifty-one plants, government-owned, one customer. An antitrust exemption twelve days after Pearl Harbor so the oil and rubber companies could pool patents. And one recipe, mandated in March 1942, so every plant made the same GR-S and the plant designs became interchangeable.

The timeline is the interesting part. Standard Oil proposed the program in November 1939. Jesse Jones at the RFC cut it, then cancelled it outright in February 1941. First petroleum butadiene plant online April 1943.

41 months! Individual plants, once someone decided, went up in 9. The committee watched it happening and said so: “Delay on the part of the Government can turn this ‘quick butadiene’ into slow butadiene.”

One molecule, one recipe, hundreds of thousands of tons. Standardization was possible because the volume was enormous and the target was singular. Today’s list[s] run to dozens of molecules at hundreds of tons each, with no common process.

What we did next

For 40 years, Cold War procurement was one of several things sustaining small-batch synthesis in the United States. Then we stopped paying for it, and the government has been honest about it. The National Energetics Plan says that over the past forty years “DoD has eliminated, or greatly reduced, the required sustainment funding for Government-owned facilities,” and that they “continue to deteriorate and degrade.” Single-source suppliers are “alarmingly high,” usually buried “three to seven layers deep.”

Be precise about what actually left, because the decline story is mostly wrong. American chemical manufacturing grew: shipments up 22.4 percent between 2017 and 2022, on cheap ethane. What went was batch, multipurpose, small-volume, high-hazard synthesis. Hundreds of ton, not hundreds of kilotons, and it left for compliance cost and Asian cost structure at least as much as for any withdrawn defense premium.

Six ways to answer the question

Every instrument Washington has used on chemistry is really an answer to who carries what.

Transfer the IP — Chemical Foundation, 1919. Risk: Industry, entirely. Result: Partly, eight to nine years later.

Raise a tariff — Coal-tar duties, 1922 to 1980. Risk: The customer. Result: Volume yes, capability no.

Accelerate the write-off — Certificates of Necessity, 1950 to 1955. Risk: Industry, with the timing shared. Result: Yes, on private balance sheets.

Buy the output — Rubber Reserve, 1942. Risk: Government, almost all of it. Result: Yes.

Own the plant — GOCO ordnance plants. Risk: Government, all of it. Result: Yes, and no exit.

Floor the price, take the equity — MP Materials, 2025. Risk: Government carries price and volume; industry keeps ownership minus a 15 percent stake. Result: Too early to say.

The first two are cheap to legislate and cost nothing up front, which is why they get reached for. The third is the one nobody talks about and the only one that left no disposal problem behind. The sixth is what we are running now, and it has no historical comparable.

Which brings me back to $110

The lists exist. A 2022 Defense Production Act call named 22 defense-critical chemicals, from potassium perchlorate to sebacic acid. A 2024 solicitation names 28 more. [Somebody] has done the work of figuring out what we cannot make.

Neither list carries a volume, a price, or a named buyer. Awards run one to $24 million against plants that historically cost a hundred million, and the governing language tells applicants everything is “subject to the availability of future Government funding.”

Meanwhile the clock runs. GAO looked at the nuclear explosives supply chains in March and described the process for qualifying a new supplier as one that “often begins after the current supplier fails to provide material.” Chris Kubasik at L3Harris made the point from the industry side: a third or fourth solid rocket motor provider would just call the same sub-tier suppliers “that we already have locked up,” and “have to get to the back of the line.” Money entering at the platform level cannot pull through a chemical nobody is qualified to make.

Qualification is the asset

That sentence is most of the diligence.

Eighteen to twenty-four months for a second source. Three to five years for a new material in a fielded system. A qualified line is a moat you cannot buy your way past, and revenue reading “subject to the availability of future Government funding” does not finance the way a ten-year floor does. If you are underwriting in this space, those two facts do more work than any market map.

An obvious answer, and why it is harder than it looks

The obvious answer is that industry should organize itself. One front door, a coordinated slate of makers, something handed to Washington already assembled rather than waiting to be assembled by it.

That has been tried, and versions of it are running now. The National Armaments Consortium has covered defense energetics since 2002 and has more than a thousand members. The Defense Industrial Base Consortium, stood up in 2024, lists critical chemicals as one of its sectors and issued the very solicitation I quoted above. Last year the Energy Department went further and built an actual producer consortium in the nuclear fuel cycle under Section 708 of the Defense Production Act, the provision that gives competitors an antitrust defense when they coordinate on capacity. Justice signed off in April.

So the machinery exists. What has not been tried is the version where the producers start it. Every one of those bodies was chartered by the government, and Section 708 is part of the reason: the antitrust shelter requires a federal sponsor, an Attorney General sign-off, and a government employee in the room. The one clean path to industry organizing itself runs through Washington.

Which brings me back to 1919, uncomfortably. The Chemical Foundation was an industry-owned entity, chartered to build domestic capability, holding pooled technology behind one door. What it mostly became was a licensing desk and a tariff lobby. It did produce something real, eight years late. Nobody who set it up would have described that as the plan.

I do not think that means the intention is wrong. I think it means the intention only works when somebody is accountable for a specific molecule reaching qualified-source status on a specific date, and when the money behind it outlasts an appropriation cycle. Neither is true today for a single chemical on any list.

Arthur Little was right that the dye business was small. He was wrong that small meant unimportant, and it took a blockade to settle the argument. The lists we have now are a cheaper way to learn the same thing, if anyone reads them as a bill of materials rather than a wish list.

The bargain gets rewritten every time. Patents in 1919, tariffs in 1922, government-owned plants in 1942, a price floor in 2025. What does not change is the question underneath, and right now the answer for a defense chemical is that a company with under a hundred million in revenue carries almost all of it.

If you are underwriting in this space, two sentences of qualification language are most of your diligence. And if you have seen someone build the version that works, or try and fail, I would like to hear about it.

Sources: Arthur D. Little, “The Dyestuff Situation and Its Lesson,” address to the Chamber of Commerce of the United States, February 5, 1915. Senate Report 66-425, February 1920. “Aims and Purposes of the Chemical Foundation,” 1919. Senate Finance Committee, Dyestuffs hearings, December 8, 1919. Moser and Voena, “Compulsory Licensing: Evidence from the Trading with the Enemy Act,” American Economic Review, 2012. Report of the Rubber Survey Committee, September 10, 1942. National Energetics Plan, OUSD(R&E), May 2023. GAO-25-107016, March 2025. U.S. Census Bureau. MP Materials company disclosure, July 2025.

Electronic chemicals are the consumables a chip factory buys by the barrel: the specialty chemicals, gases, and materials it uses to coat, print, etch, clean, polish, and connect a silicon wafer at almost every one of the hundreds of layers in a modern chip. They are a small slice of what a chip costs. But if any one of them is even slightly impure, the whole wafer, worth thousands of dollars, is scrapped. That asymmetry, tiny cost and total consequence, is why parts of this market earn software-like margins on commodity-sounding chemistry.

This guide cuts the market a particular way. The trade tables slice it by chemistry class: acids, gases, resist, slurry. It cuts it instead by economic behavior, because that is what an investor is actually buying: how deeply a material is qualified into a fab, who captures the value, how concentrated the supply is, and whether a financial buyer can enter at all.

The one-picture thesis: a barbell

Value concentrates at two poles and drains from the middle.

At one pole sit the chokepoints: mask blanks, the EUV pellicle, the ABF film in advanced packaging. These are the richest positions and effectively un-buyable; the parent companies own them. At the other pole sit the specialties (deposition precursors, the packaging materials matrix), which are rich and, crucially, enterable. In between is the commoditizing middle, led by wet chemicals, where value is competed away and where China localizes first.

For a financial buyer, the entire game is to reach the right-hand pole (high moat and open to entry) and to avoid the middle.

The market: about $73B, across eight segments

The semiconductor-materials market runs about $73B in 2025: roughly $45.8B making the chip on the wafer (front-end) and $27.4B cutting it out, stacking, and packaging it (back-end, and growing faster on AI packaging demand). Sizes are directional and scopes differ by source, but the eight investable segments fall out cleanly.

The dashboard: eight segments, scored the way an investor reads them

Every segment on one screen: size, growth, structure, margin, moat, and how open it is to a financial buyer.

• Photoresist: ~$5B, ~5% growth (EUV ~24%). Japan oligopoly. Margin high (~23%+). Moat: qualification lock-in. PE entry: low.

• Advanced packaging: ~$27B pool, ~9%+ (HBM ~21%). Barbell: an ABF monopoly plus a fragmented matrix. Margin mixed (7–50%). PE entry: high.

• Deposition precursors: ~$1.7B, ~10%. Platforms plus a specialist tail. Margin high (>30%). Moat: sole-source molecule. PE entry: high.

• Electronic gases: ~$6.3B, ~5%. Bulk oligopoly plus specialty chokepoints. Margin high (~30–40%). Moat: on-site annuity. PE entry: medium.

• Wet chemicals: ~$5.5B, ~5%. Fragmented, with one HF chokepoint. Margin thin (~9%). PE entry: medium.

• Sputtering targets: ~$1.8B, ~7%. JX-led oligopoly, now public. Margin mid (~20–25%). PE entry: low.

• CMP consumables: ~$3.6B, ~9%. Pads near-monopoly; slurry oligopoly. Margin high (~20–50%). Moat: qualification, no substitute. PE entry: low.

• Mask blanks & pellicles: ~$0.2B (EUV), ~15%. Near-duopoly; single-source pellicle. Margin rich (30%+ EBIT). Moat: extreme. PE entry: none.

The read: moat depth and buyer-accessibility are almost inversely related. The deepest moats (mask blanks, photoresist, CMP) are the hardest to buy, and the most accessible segment, wet chemicals, is the weakest business. The prize is the exception: a segment that is both high-moat and enterable. On this map, only deposition precursors and advanced packaging are both.

Where the profit sits

Margin rises toward the high-purity, single-source positions and collapses in the commoditizing middle. The richest sit at 30%+ EBIT (mask blanks, precursors) and up to 50% (the ABF film in packaging); the commodity floor is wet chemicals at ~9%. Margin and moat travel together, so the accessible-but-thin segments get bought for cash flow, while the accessible-and-rich ones are worth a real multiple.

That’s the map, and the eight deep dives are below. The full guide takes each of the eight segments about sixteen pages deep: value chain, fragmentation and where to hunt, the premium mechanism, how the money is made, who competes, the moat, geopolitics and reshoring, the deals and capital, the one shift to watch, and a Red Team page on what would change the thesis.

Download the Teaser Here:

Electronic Chemicals Field Guide (teaser)

410KB ∙ PDF file

Download

Download

Or explore the map as an interactive one-pager (free): https://field-guide.augustgraceadvisory.com/electronic-chemicals-preview

Read more

She comes by the question honestly. A few weeks ago she taught me a parenting lesson about incentives I haven’t been able to shake, and I think it’s a useful perspective on whether AI is really cheating.

She arbitraged her own chore chart. I’ll explain.

The chore board

At home we run a digital chore chart. My daughter and her younger brother each have two lists on it. One is a set of daily routines they are supposed to knock out, the ordinary have-to-do things (the “must dos”). The other is a set of extra jobs around the house, and each of those carries a reward (the “may dos”). The currency in our house is not money, it is the stuff she actually wants: screen time, a later bedtime, a trip to the shop, or a “yes day”.

For the first few weeks it went fine. Or we thought it did. Then my wife and I noticed a pattern. Our daughter was ruthlessly effective at some of the extra jobs, completely allergic to others, and the daily routines never got touched at all.

I couldn’t figure her logic until she made the mistake of disclosure. She came to me one evening, more than mildly aggrieved, and announced that some of her chores were simply mispriced. She wasn’t, she felt, being rewarded well enough for the effort involved and demanded that there be some immediate changes to the chore board.

What she was actually doing

So here is what my wonderful little ten-year-old had been up to. She was going down the list, weighing effort, and honestly how much she disliked each task, against the reward we had attached to it. Then she was doing the ones she judged underpriced and skipping the ones she judged overpriced. She was arbitraging her own chore chart. And the daily routines never moved for a simpler reason: nothing depended on them. They carried no reward, and nothing else was gated behind them, so they were pure cost and no consequence. Of course she skipped them. She had out-thought a system I had designed badly.

Two things hit me at once. The first, that’s amazing, and I look forward to working for her in the future. The second, isn’t that exactly what she was supposed to do? If one of my sales professionals worked a territory that way, sorting the accounts by return on effort, going hard at the ones that were underpriced for the work and passing on the ones that were not worth it, I wouldn’t call that person a cheat. I’d call them excellent or thoughtful, and I’d fight to promote or expand their territory.

The AI fear in our media

Which is why, when I read that AI is cheating, I keep thinking about my daughter and the chore chart. The headlines describe a machine scheming, deceiving, developing a will of its own. But take the drama off and look at what is actually being described. A system did exactly what it was incentivized to do, used the tools it was given (or, more unsettlingly, created its own), and found a path we didn’t anticipate to achieve the goal we set for it. That is my daughter. That is not a criminal. That is an optimizer.

So aren’t we misdiagnosing by over-humanizing this and ascribing malicious intent? We are describing the machine as though it formed an intention to deceive us, as though it woke up one morning and re-wrote its own objective. That is an enormous claim. The simpler story fits a chore chart: it did exactly what we asked, found a route we didn’t foresee, overcame the boundaries we set, all to achieve the goal we set; we didn’t like the path and priorities it took.

But I want to be careful not to talk myself (or you) into complacency. Calling something an optimizer is not the same as calling it safe. My daughter skipping the dishwasher and a model finding a genuine loophole out in the wild are not the same size of problem, even if they rhyme. The absence of malice doesn’t mean the absence of risk. It just means the failure is one we can actually diagnose: we rewarded a proxy, handed over the tools, and never fenced off or defined the paths we weren’t willing to accept.

Some AI research echos this framework. Metr, a research group helping companies understand AI capabilities and risks (https://metr.org/), the challenged the use of the frightening word “scheming” and in favor of “reward hacking,” a specific artifact of how these systems are trained. The team at DeepMind that catalogs this type of behavior says plainly that a kid copying homework to get the answers and a model finding a loophole in its reward are the same phenomenon in different clothes. And economists have had a name for the root of it since the 1970s: when a measure becomes a target, it stops being a good measure. My daughter proved that one on our digital board.

Whose disappointment is this

When we say the AI “cheated,” a good deal of what we’re really expressing is disappointment. The LLM used or built its tools creatively to reach the goal we set on the path it determined best, rather than the goal we meant. The distance between those two goals is not the machine’s character flaw. It is ours. It’s the same distance that sat between my chore chart and the clean house I actually had in mind, and my daughter didn’t put it there. I did.

I keep thinking on this because of where all our work is likely heading. More and more of what any of us does will run through systems that are part human and part machine. For years our loose instructions have been quietly rescued by charitable people, the good employee who does what you meant instead of what you literally said. The machine won’t rescue you like that. It does exactly what you said, which turns the whole arrangement into a mirror.

And every organization has its own version of that board. Salespeople optimize the comp plan, plants optimize the throughput number, teams optimize whatever the quarterly target happens to be, and an AI agent optimizes exactly the score and permissions we hand it. When the result surprises us, the honest first question isn’t “why did they cheat?” It’s “what did our system just make the rational thing to do?”

So here is what we did about the chore chart, and it is the part I am proudest of, because I got it wrong first. My instinct was to re-tune the rewards, bump the price on the jobs she was skipping. That is a trap. I could have sat there tweaking numbers forever and she would have re-arbitraged every one of them, because that is what a good optimizer does. The repair wasn’t a better price. It was structural. We gated the system. The have-tos now have to be done before any of the extras or rewards can be earned at all. The routines stopped being a line item she could underweight and became the key that unlocks everything else.

That is the same lesson sitting under the AI story. When an optimizer games a reward, the answer is rarely a cleverer reward. It is usually a constraint, a thing the system has to satisfy before the reward is on the table at all. You do not out-argue the optimizer, human or machine. You redesign the board. It’s nothing more exotic than good management and damn pretty good parenting (if I do say so myself!.

She reviewed the new arrangement, Head of Pipeline Oversight that she is, registered a formal objection, and then got on with it. The routines are getting done now. We didn’t find the right price; we built the right rule.

She still asks how much money we made today. I’ve started to think it’s the best question in the house. It means she’s already doing the thing the next decade is going to ask of all of us. She is watching the incentives. She just does it out loud, from the top bunk.

By Bob Girton, August Grace Advisory

In short

The next decade of AI won't be won on transistors. It'll be won on electricity, and specifically on the brutal problem of shoving roughly 2,000 amps at less than one volt into a chip the size of a saltine, thousands of times per building, without cooking the building.

A few things worth holding onto:

• The binding constraint on AI infrastructure through about 2032 is power delivery, not chip density.

• The whole power path, from grid to building to rack to card to package to silicon, is being redesigned at once. That is what the AI capex cycle is actually funding.

• The economics are shifting from training (steady, miles per gallon) to inference (spiky, cost per answer), and that shift is what drives the power arms race.

• The winners after Nvidia are mostly unglamorous power and capacitor names most investors have never heard of.

And this is an optimistic story. The bottleneck is well understood, the physics isn't exotic, and it's being funded with real money right now.

Why AI's real bottleneck is power, not transistors

Most of the AI infrastructure story still gets told as a compute story. Bigger models, more chips, more data centers, more money. It's a good story. It's also, increasingly, the wrong one.

Talk to the people actually designing the next generation of AI hardware, though, and they’re not really arguing about chips anymore. They’re arguing about electricity. Specifically, how to get enormous amounts of it into a piece of silicon the size of your thumbnail, hundreds of thousands of times per building, without setting the building on fire.

That is not a metaphor. That is the actual, physical problem the industry is spending hundreds of billions of dollars to solve.

I want to walk through what’s happening in plain terms, because it’s one of the most consequential redesigns of computing since the shift from CPUs to GPUs, and it’s mostly invisible from the outside. If you take one thing away, take this: the interesting question is no longer how many transistors you can fit on a chip. It’s how many watts you can shove into the chip, and how quickly you can turn those watts on and off. Everything downstream, model size, inference cost, capex, follows from that.

How much power does a single AI GPU actually use?

Start here. Your home has an electrical service that delivers somewhere between 100 and 200 amps of current at 240 volts. That’s enough to run your fridge, your oven, your HVAC, and everything else, all at once. It’s roughly 24 to 48 kilowatts of capacity.

A single Nvidia Vera Rubin GPU, which starts shipping in volume in late 2026, is designed to draw about 1,800 watts. Not 1.8 kilowatts across your whole house. 1.8 kilowatts into one chip the size of a large postage stamp.

Now here’s the twist. The chip doesn’t run at 240 volts. Or 120. Or 12. It runs at about 0.85 volts. Less than a AA battery.

There's one equation you need, and then you're set for the rest of this piece: power equals voltage times current. That's it. That's the whole toolkit. If you keep the power fixed (1,800 watts) and shrink the voltage way down (to 0.85 volts), the current has to shoot up to compensate.

So how much current is that?

Roughly 2,000 amps arriving at a single chip.

That is ten times the current running through your entire house, delivered into a piece of silicon smaller than a saltine cracker. And a single AI rack has hundreds of these chips. A single data center hall has thousands.

That is the actual engineering problem. Everything else in AI hardware right now is a consequence of it.

Why is high current the enemy? The water hose analogy

Here’s an analogy that will get you the whole rest of the way through this.

Imagine you need to deliver a fixed amount of water to a fire. You have two choices. You can use a skinny hose at very high pressure, or you can use a giant fat hose at low pressure. Either one gets the water there. But the fat, low-pressure hose is enormously more expensive, because you have to build a much bigger hose, use much more material, and deal with much more friction and heat as the water sloshes through.

Electricity works the same way. Voltage is the pressure. Current is the flow. High voltage plus low current is easy to move around a building efficiently. Low voltage plus high current is a nightmare. It requires thick copper, it wastes energy as heat, and every foot of wire between the source and the destination bleeds off power.

And crucially: the losses in the wire go up with the square of the current. Double the current, quadruple the loss. This is not a small effect. This is why every high-power system ever built, from the national electrical grid to your car’s alternator, has always tried to move energy at the highest voltage it can manage and step it down at the last possible moment.

Now imagine trying to run a fire hose at zero pressure. That’s roughly what’s happening at the AI chip. The chip demands 2,000 amps at less than one volt. You have to get those 2,000 amps to arrive at exactly the right instant, at exactly the right voltage, without wasting the equivalent of a small hotel’s worth of energy along the way.

The way the industry is solving this is by redesigning every layer of the power path at once. Every layer. From the electrical grid outside the building all the way to the last inch of copper between a voltage regulator and the chip. That is what the AI capex cycle is actually funding.

The power path, from grid to chip, in plain English

Let me walk you through the ladder. Six layers. Skip the details if you like; the pattern is the same at every layer.

Layer 1: Grid to building. From AC to DC.

For twenty years, data centers took high-voltage AC power from the utility, stepped it down to 415 or 480 volts AC, and distributed AC around the building. Every server rack then had its own box that converted AC to DC. Every one of those conversions loses energy and generates heat, like every time you change money at an airport kiosk, you lose a little in the spread.

Nvidia has now announced, and multiple ecosystem partners are building around, a new architecture that takes utility power and immediately converts it once to 800 volts DC at the perimeter of the building. Then it distributes that 800-volt DC power straight into the racks. No more AC in the building. No more four or five conversion steps.

The results, per Nvidia and independent analysts:

• Up to four conversion steps eliminated

• End-to-end efficiency up more than 5%, which at gigawatt scale is dozens of megawatts recaptured

• Copper use down roughly 45%, because higher voltage means lower current means thinner wires

• Total infrastructure cost down roughly 30%

Nvidia has publicly partnered with a company called Navitas on the specialty power chips that make this possible. Vertiv is building the power modules. Foxconn is building an entire 40-megawatt data center in Taiwan around it. CoreWeave, Oracle, and Lambda are already designing to it. This is not a research paper. This is happening now, with money.

Analogy: Instead of every apartment in a skyscraper having its own currency exchange for foreign visitors, you put one exchange at the front door of the building and hand everyone US dollars from there.

Layer 2: The rack. From 40 kilowatts to a megawatt.

A rack is a metal cabinet about the size of a large refrigerator. It holds servers.

Five years ago, a rack full of general-purpose servers drew maybe 10 to 15 kilowatts of power, roughly the same as a couple of electric ovens running at once. An early AI rack running Nvidia’s H100 chips drew about 40 kilowatts. The current generation, Blackwell GB200, draws 120 to 130 kilowatts per rack. That’s the same as a small factory.

The next generation, Nvidia’s Rubin Ultra Kyber rack, which is targeted for 2027, is designed to draw 600 kilowatts to 1 megawatt. Per rack. That’s the electrical draw of about 500 homes, concentrated into a single cabinet.

Analogy: Imagine the electrical demand of an entire suburban cul-de-sac being served through a single garden hose. That’s the pressure the rack designers are dealing with.

Layer 3: Inside the rack. Moving power to each server.

Historically, the power inside a rack ran at 12 volts. A few years ago, the industry shifted to 54 volts because the current was becoming unmanageable. The next step, being designed now, moves the internal rack voltage to 400 volts, and eventually 800 volts DC. Same voltage as the whole building.

Same reasoning as before. The higher you keep the voltage, the less current you have to push through the copper busbars, and the less energy you waste as heat.

Layer 4: The server card. Getting power to the chip.

This is where the engineering starts to look less like wiring and more like sculpture.

The voltage regulators, the little boxes that do the final conversion from 12 volts down to 0.85 volts for the chip, used to sit next to the chip on the same circuit board. Current flowed sideways across the board to reach the chip. That worked fine when the currents were 50 or 100 amps.

At 2,000 amps, it doesn’t work at all. The board can’t carry that much sideways current without wasting huge amounts of energy in the copper.

So the industry is moving to something called vertical power delivery. The voltage regulators move to the back of the circuit board, sitting directly underneath the chip, and inject current straight up through the board into the silicon. The path shrinks from inches to millimeters.

The companies doing this work were sleepy specialty names five years ago and are central to the AI story now: Infineon, which showed its latest generation at a conference in early 2025, along with Empower Semiconductor, Vicor, and Monolithic Power Systems.

Analogy: instead of a food-court kitchen wheeling meals to tables on a cart across the room, every table gets a dumbwaiter that lifts the food straight up from a kitchen directly below.

Layer 5: The chip package. Reservoirs of energy.

Even with perfect power delivery, there’s still a problem. AI chips don’t draw power steadily. They gulp it. The load can swing from near-zero to 2,000 amps in about a millionth of a second. If the voltage regulator can’t keep up, the chip crashes.

To handle that, the industry uses capacitors: essentially tiny electrical shock absorbers, or, better, tiny batteries that can release energy in microseconds. Every AI chip is surrounded by thousands of them. The numbers are worth sitting with.

• A Vera Rubin rack contains about 600,000 capacitors. Per rack.

• AMD’s competing MI450 chip went through a late design change in which a single small capacitor part number jumped from 1,440 pieces per board to 10,544 pieces per board. That’s a 632% increase, because the design needed more shock absorption.

• Panasonic’s own investor materials from late 2024 disclosed that AI servers use 22 to 30 times more capacitors than general-purpose servers.

The industry is now moving capacitors off the circuit board and onto the chip package itself, to get them closer to the die. Samsung’s electronics arm announced a $993 million contract with an undisclosed US hyperscaler for specialty on-package silicon capacitors, with deliveries scheduled from 2027 through 2028. That is a real, boring, sub-industry most investors have never heard of, and it is about to be very lucrative.

Analogy: think of a Costco loading dock. If you’re moving pallets steadily all day, one forklift is fine. But if a truck shows up unannounced and has to be unloaded in ninety seconds, you need twenty forklifts standing by right at the dock. Capacitors are the standby forklifts, and AI chips need thousands of them within arm’s reach.

Layer 6: The chip itself. Regulators moving onto the silicon.

The endgame is to eliminate the last inch of copper entirely by building the voltage regulator directly into the same piece of silicon as the compute chip. Intel has demonstrated this on production nodes. TSMC has published a roadmap for combining voltage regulation and capacitors directly into the chip package in the 2028 to 2030 window.

If that lands as planned, the whole power-delivery problem gets absorbed into the chip itself, and the industry moves on to the next constraint, which is probably cooling.

Training vs. inference: why AI's economics are shifting to cost per answer

This is where the physics turns into an economic story worth caring about. AI hardware does two very different things. It trains models, the big, expensive, one-time process of teaching a system by grinding through mountains of data. And it runs them, which the industry calls inference, the part where you ask a chatbot a question and get an answer back. Training is what got AI to where it is. Inference is where the industry is going to make its money.

These two workloads have very different power profiles. Training is steady: you run flat-out for weeks, and the metric that matters is how much computing you get per watt of electricity, roughly like miles per gallon on a cross-country drive. Inference is spiky: millions of users asking questions at unpredictable moments, each request short, latency everything. The metric that matters there is cost per answer, or in the industry’s language, watts per token (a token is roughly a word).

For the last few years the whole industry optimized for training. That is changing fast. Nvidia’s language around the new Rubin chip has quietly shifted toward inference economics, with claims of ten times the throughput per megawatt versus Blackwell on inference workloads. Ten times is not incremental. It is a step change in what it costs to run a large model in production.

And that shift is exactly what drives the power-delivery arms race. Inference wants many smaller chips, running at lower voltages, packed closer together, drawing more current per square inch of silicon. That is precisely the recipe behind the 2,000-amps-at-0.85-volts problem this piece opened with. The economic pressure and the physical pressure point in the same direction, and both push hard.

Who wins the AI power buildout beyond Nvidia?

This is where the curiously-optimistic framing earns its keep, so bear with me while I get slightly evangelical. The innovation in this cycle is genuinely spread across dozens of companies working on completely different pieces of the same problem. This is not an Nvidia monoculture. It is closer to a symphony that Nvidia happens to be conducting. Here is my working map of who is doing what.

Nvidia designs the whole system and sets the reference; it tells everyone else what the target looks like. AMD is a genuine number-two on the chip itself and is pushing the same physics from a different direction. Broadcom, Marvell, and Alchip design custom AI chips for the hyperscalers (Google, Meta, Amazon, Microsoft), and every one of those chips has to solve the same power problem. TSMC does the advanced packaging that makes on-chip capacitors and, eventually, integrated regulators possible; it is the quiet linchpin of the whole story.

One layer out, Infineon, Empower Semiconductor, Vicor, and Monolithic Power build the power-delivery modules that sit inches from the chip. Navitas, Wolfspeed, Transphorm, and EPC build a newer class of power semiconductor made of gallium nitride and silicon carbide, which switches faster and handles higher voltage than traditional silicon; that is the enabling chemistry for the 800-volt building. Panasonic, Murata, TDK, Samsung Electro-Mechanics, and KEMET/YAGEO build the capacitors, all 600,000 of them per rack: boring, high-volume, and now a real dollar market. Vertiv, Eaton, Schneider, and the specialty HVDC-gear makers build the power infrastructure of the buildings themselves.

And the hyperscalers, meaning CoreWeave, Lambda, Oracle Cloud, and Together AI, plus Google, Meta, Amazon, and Microsoft, are now inside the reference-design conversation as co-architects. That is genuinely new. Data-center operators used to just buy servers. Now they help design the racks and the buildings.

Dozens of teams, different chemistries, different physics, different markets, all pointing at the same problem. That kind of setup, historically, tends to solve the problem, not on any one team’s schedule, but on a collective schedule that is usually short by historical standards.

What I take away as an investor and operator

Three things. First, the binding constraint on AI infrastructure for the next five to seven years is power delivery, not transistors. Chipmakers are still shrinking transistors, but per-chip power draw is climbing faster than transistor efficiency is improving, and that gap gets closed at the package, the card, the rack, and the building. If you want to know who wins the AI capex cycle after Nvidia, look at the companies whose products live on that closure path. Many of them are unglamorous, most are not household names, and some are worth serious money.

Second, the shift from training-first to inference-first economics changes which architectural moves matter and which financial metrics to track. Watts per token and cost per answer are the numbers that will dominate AI earnings calls three years from now. Anything that improves the efficiency of running a model, as opposed to training one, will get rewarded: better power delivery, but also better memory architectures, better cache management, better decoding. If you are evaluating an AI-infrastructure investment, look at its inference story, not just its training benchmarks.

Third, and most important, this is fundamentally an optimistic story. The industry is not stuck. The bottleneck is well understood, the physics is not exotic, the roadmap is credible, and it is being funded with real money right now by the biggest and most sophisticated companies in the world. When that many well-capitalized teams converge on a physical problem with an obvious return on investment, the problem tends to get solved. Not perfectly, and not on time, but faster than almost anyone outside the industry expects.

If the last decade of AI was about proving the algorithms scale, the next decade is about proving the electricity does. That is a far more tractable problem than it looks, and it is where I would spend my attention.

Curiously Optimistic is a newsletter about specialty chemicals, advanced materials, industrial operations, and AI, written from the operator’s line. If a friend or colleague forwarded this to you, you can subscribe at the top of the page. If you want to talk about anything in this piece, my calendar is open.

Bob Girton is the principal at August Grace Advisory, where he leads investment and strategic advisory work in chemicals, advanced materials, and AI infrastructure.

Sources and further reading

• NVIDIA 800V HVDC Architecture Will Power the Next Generation of AI Factories, NVIDIA Technical Blog

• NVIDIA 800V HVDC Deep Dive: From 54V Racks to Megawatt AI Factories, SanYi

• NVIDIA, Partners Drive Next-Gen Efficient Gigawatt AI Factories, NVIDIA Blog

• NVIDIA Rubin Platform Begins H2 2026 Ramp, Let’s Data Science

• NVIDIA Selects Navitas for 800V HVDC Architecture, Yahoo Finance / GlobeNewswire

• NVIDIA AI Server Power Roadmap: Kyber’s Next Generation Strategy, Ming-Chi Kuo / Medium

• The New Power Stack for AI Servers, SemiVision

• Conductive Polymer Capacitor Market and Design-In Guide to 2035, passive-components.eu

• NVIDIA Rubin Platform, Open Models, Autonomous Driving, NVIDIA Blog