Workers at the Naval Surface Warfare Center in Indian Head, Maryland, the only facility where torpedo fuel is produced for the U.S. Navy
In the event of a late-2020s Chinese attempt to invade Taiwan, leading to a US-China military conflict, what would be the most urgent bottlenecks in military equipment?
Where is there the greatest need to scale up defense manufacturing?
I am not an expert in military matters, so take all this with a grain of salt. But I looked into this question and it seems to have a single clear-cut answer: energetics. (That is, explosives and propellants for munitions like missiles and torpedoes).
The US has extremely small manufacturing capacity for energetics, and in the event of a Pacific war, demand would rapidly exceed supply.
What Does a US-China War Look Like?
Military experts think there’s a substantial, but not overwhelming, chance that China will invade Taiwan before 2030.
Metaculus predicts an invasion at 13% by 2028, and 25% by 2030.
The “Davidson window”, named for retired Admiral Philip S. Davidson, refers to the view that China will invade Taiwan by 2027, which “gained widespread attention” following former CIA director William J. Burns’ 2023 announcement that U.S. intelligence had found that Xi Jinping had told the People’s Liberation Army to be ready for an invasion by 2027.
The “Davidson window” model is not universally shared within the US defense & foreign policy establishment. In a Defense Priorities survey of 51 experts, 85% rated a Chinese invasion of Taiwan as “somewhat unlikely” or “very unlikely.”
On the other hand, the 2025 Pentagon’s Annual China Military Report to Congress straightforwardly declares that “China expects to be able to fight and win a war on Taiwan by the end of 2027.”
The bottom line is that there is substantial uncertainty and disagreement about the chance of a near-term Chinese invasion of Taiwan — but that still puts it as enough of a possibility to be worth planning for.
What will the US do in case of an invasion of Taiwan?
Here, we have an actual official answer, from the US’s 2026 National Defense Strategy. The strategy is known as “deterrence by denial”: that is, deterring China from invading Taiwan via a credible threat to sink the Chinese fleet before it crosses the strait.
This involves setting up defenses along the First Island Chain, which runs from Japan through Taiwan and the Philippines to Indonesia. Accordingly, this is where US forces in the Pacific are currently stationed: mostly in Japan, Hawaii, South Korea, and Guam.
Recent CSIS wargames represent the best non-classified information about how the US expects a conflict to play out. They present the US deploying submarines in and around the Taiwan Strait and Philippine Sea, as well as bombers launched from Pacific bases like Guam and Kadena. Bombers and submarines would launch missiles against Chinese ships and ports, with the goal of preventing an invasion of Taiwan.
The invasion always starts the same way: an opening bombardment destroys most of Taiwan’s navy and air force in the first hours of hostilities. Augmented by a powerful rocket force, the Chinese navy encircles Taiwan and interdicts any attempts to get ships and aircraft to the besieged island. Tens of thousands of Chinese soldiers cross the strait in a mix of military amphibious craft and civilian rollon, roll-off ships, while air assault and airborne troops land behind the beachheads.
However, in the most likely “base scenario,” the Chinese invasion quickly founders. Despite massive Chinese bombardment, Taiwanese ground forces stream to the beachhead, where the invaders struggle to build up supplies and move inland. Meanwhile U.S. submarines, bombers, and fighter/attack aircraft, often reinforced by Japan Self-Defense Forces, rapidly cripple the Chinese amphibious fleet. China’s strikes on Japanese bases and U.S. surface ships cannot change the result: Taiwan remains autonomous.
Munitions Shortages
Fundamentally, this is a situation where the US (and its allies like Japan) has a roughly fixed supply of ships and aircraft, which are continuously delivering munitions (precision-guided missiles and torpedoes) to the Chinese fleet. The munitions are “consumables” that get used up; the ships and aircraft are (hopefully!) more durable. So a priori you’d expect the limiting factor on supply to be munitions.
In fact, that’s exactly what national security experts have been observing for years. (See e.g. this 2025 piece from the Foreign Policy Research Institute.) The US has shortages of artillery shells, THAAD anti-missile defenses (which are themselves missiles), and precision-guided missiles.
The key munitions likely to be used in a US-China war over Taiwan include:
AGM-158C LRASM, a bomber-launched anti-ship missile.
AGM-158B JASSM-ER, also a precision-guided long-range anti-ship missile, and also suffering from low production and stockpiles.
Tomahawk cruise missiles, launched from submarines, used against ships
Mk 48 ADCAP torpedoes, also launched from submarines and used against ships
The Heritage Foundation reports that the US only has 250 LRASMs, against a requirement exceeding 1000. We are estimated to have only enough precision-guided munitions for a few weeks of combat, with some specific types, like LRASMs, only lasting for one week. Torpedoes are estimated to be exhausted in months.
It’s not easy to make more of these, either. Two-year lead times are typical. “Annual production rates—115 LRASMs and 79 to 120 MK 48 torpedoes—are orders of magnitude below projected weekly or monthly wartime consumption.”
One thing to keep in mind is that defense contractors are oligopolistic: the same few suppliers make just about everything. LRASMs, JASSMs, and ADCAP torpedoes are manufactured by Lockheed Martin; Tomahawks are manufactured by Raytheon. LRASMs and JASSMs share the same turbofan engine, manufactured exclusively by Williams International.
But what is the fundamental bottleneck in manufacturing munitions? Why can’t Lockheed and Raytheon simply make way more of them?
There are several difficult-to-manufacture components, including radiation-hardened electronics and carbon-fiber motor casings, but the most extreme bottleneck is energetics.
The Energetics Supply Chain is Tiny
Say what you will about American manufacturing, there are still lots of things we’re great at making at scale.
US domestic production of materials like plastics, oil and gas, concrete, and fertilizer is excellent. So is production of high-tech, mass-produced finished goods like pharmaceuticals, automotives, and airplanes.
Even where we’re weaker, at intermediate products like fabricated metal parts, there are still machine shops across the country that are capable of doing the work, they’re just fragmented and inefficient.1
Energetics are not like that.
Explosives and propellants, and their precursors/inputs, are often made exclusively in one facility in the US — or not at all.
One facility in Louisiana, GOEX Industries, is the only domestic producer of black powder — and it only recently reopened in 2022 after a shutdown due to an industrial accident.
Only one US company, AMPAC, is certified to provide ammonium perchlorate, an oxidizer used in rocket propellant. (Again, AMPAC scaled down in the wake of a catastrophic industrial accident — the largest non-nuclear explosion in US history, in 1988).
Radford Army Ammunition Plant in Virginia is the only manufacturer of military-grade nitrocellulose, “the key ingredient in the manufacture of all propellants.”
Holston Army Ammunition Plant in Tennessee is the only manufacturer of RDX (cyclonite), the active ingredient in the polymer-bonded explosives used in JASSMs, LRASMs, and similar weapons. Mixing the polymer with the explosive filler, and loading the warheads with the mixture, are similarly only done at a handful of GOCO (government-owned, contractor-operated) facilities.
A typical chemical manufacturing facility isn’t good enough for explosives manufacturing; for instance, you need special concrete-reinforced bays to stop potential explosions. There’s substantial regulatory restrictions, due to the unusual safety and environmental hazards of making energetics. So there’s extra expense and expertise involved in making a new explosives plant, compared to other kinds of manufacturing.
Modernizing or constructing new explosive plants seems to cost high tens to hundreds of millions of dollars — $435M to construct a new TNT plant, $93M to restart M6 propellant manufacturing at Radford, $600M to triple production at Holston, etc. (These are all military budget allocations; the actual minimum costs are likely smaller).
Fundamentally, the reason energetics manufacturing has dwindled is that the US military is a monopsony. There is essentially only one buyer; in peacetime, demand is limited; and after the Cold War ended, the US DOD deliberately slimmed down the defense industrial base and shut down “idle” facilities, cutting costs to produce the “peace dividend” budget surplus of the Clinton years. Until recently, manufacturing lots of explosives was simply not a priority for the only buyer of military-grade explosives.
Is There A (Tech) Business Here?
The energetics industry looks really gnarly for new entrants: dirty, dangerous, highly regulated, with volatile demand and an absolute requirement for rare forms of domain expertise.
On the other hand, if you ask “in principle, could new technology help us make energetics more efficiently?” the answer is obviously yes. We’re using a lot of the same processes — in the same facilities! — that we did in WWII. There’s a ton of room for modernization.
First of all, if you ever want to make much higher volumes, you’d want to do as the pharmaceutical and chemical industries do, and switch from batch processing to continuous flow processing.
Second of all, mixing polymer-bonded explosives without causing defects (or accidents!) is fundamentally the same kind of composite manufacturing that you see in the plastics and ceramics industries, it’s just that the “fill” is highly chemically reactive. Like a lot of these materials manufacturing processes, it’s exactly the sort of messy, sensitive process that often doesn’t behave the way the hard-coded physics models say it will — which means it’s heuristically a great candidate for an AI application.
Take a lot of process data to train on, and a big, nonparametric time series model (which is really what a Transformer is), and you can plausibly do a lot better than SOTA on estimating the quality of the final output and what you can do to improve it. Which, in this case, includes the chance it will go boom too soon or not at all.
When it comes to unstable chemistry and complex, non-uniform materials, we do not know with full mechanistic certainty what factors affect process failures — that’s why there are so many industrial accidents in this business. And that’s precisely why “learn from lots of data” and “detect anomalies, even subtle ones that don’t count as “failures” by the official rubric, and try to remediate them” is likely to be a good strategy here.
Admittedly, this is a little bit pie-in-the-sky compared to the challenge of building new manufacturing plants at all. It’s just a pointer in the direction of “yes, going high-tech here could be a substantial improvement in productivity and safety.”
So is anyone starting new companies in this space?
Deterrence raised $10.1M in a seed round in 2025 to make explosives with robots.
Supply Energetics, founded in 2025, is manufacturing nitrocellulose in Kansas.
Firehawk, founded 2020, is manufacturing 3D-printed propellant in Oklahoma for solid rocket motors.
X-Bow Systems, founded 2016, makes solid rocket motors, including propellant, in Texas
It’s not a ton, but it’s some evidence that new entrants into this field are viable at all.
It May Be Too Late
If there’s a war in the 2020s, with China or anyone else, that consumes large quantities of munitions, there’s a very serious time crunch for scaling up energetics production.
(And the conflict in Iran definitely isn’t helping the stockpile situation.)
If it takes over a year to get a new facility (or an expansion) built and ready to begin operations, and another year to get the process development optimized and production ramped up, and several more years to get qualified as milspec? Then the timeline doesn’t work for a war before the end of the decade.
The military would have to streamline and speed up its procurement and qualifications process. Or loosen its prohibitions on sourcing energetics from allied foreign nations. Or substitute inferior but more abundant materials. Or cannibalize munitions from elsewhere in the world to focus on an active conflict. None of these are comfortable, business-as-usual choices.
It’s an unfortunate situation. One could make a rah-rah American Dynamism defense-tech pitch here, but it would ring a little hollow; I’m honestly not sure if this problem is going to get solved at all.
And, Austin Vernon argues, American metal fabrication could be globally competitive if it modernized its “back office” operations like quoting and billing, with the magic of ~~software ~~
In the event of a late-2020s Chinese attempt to invade Taiwan, leading to a US-China military conflict, what would be the most urgent bottlenecks in military equipment?
Where is there the greatest need to scale up defense manufacturing?
I am not an expert in military matters, so take all this with a grain of salt. But I looked into this question and it seems to have a single clear-cut answer: energetics. (That is, explosives and propellants for munitions like missiles and torpedoes).
The US has extremely small manufacturing capacity for energetics, and in the event of a Pacific war, demand would rapidly exceed supply.
What Does a US-China War Look Like?
Military experts think there’s a substantial, but not overwhelming, chance that China will invade Taiwan before 2030.
Metaculus predicts an invasion at 13% by 2028, and 25% by 2030.
The “Davidson window”, named for retired Admiral Philip S. Davidson, refers to the view that China will invade Taiwan by 2027, which “gained widespread attention” following former CIA director William J. Burns’ 2023 announcement that U.S. intelligence had found that Xi Jinping had told the People’s Liberation Army to be ready for an invasion by 2027.
The “Davidson window” model is not universally shared within the US defense & foreign policy establishment. In a Defense Priorities survey of 51 experts, 85% rated a Chinese invasion of Taiwan as “somewhat unlikely” or “very unlikely.”
On the other hand, the 2025 Pentagon’s Annual China Military Report to Congress straightforwardly declares that “China expects to be able to fight and win a war on Taiwan by the end of 2027.”
The bottom line is that there is substantial uncertainty and disagreement about the chance of a near-term Chinese invasion of Taiwan — but that still puts it as enough of a possibility to be worth planning for.
What will the US do in case of an invasion of Taiwan?
Here, we have an actual official answer, from the US’s 2026 National Defense Strategy. The strategy is known as “deterrence by denial”: that is, deterring China from invading Taiwan via a credible threat to sink the Chinese fleet before it crosses the strait.
This involves setting up defenses along the First Island Chain, which runs from Japan through Taiwan and the Philippines to Indonesia. Accordingly, this is where US forces in the Pacific are currently stationed: mostly in Japan, Hawaii, South Korea, and Guam.
Recent CSIS wargames represent the best non-classified information about how the US expects a conflict to play out. They present the US deploying submarines in and around the Taiwan Strait and Philippine Sea, as well as bombers launched from Pacific bases like Guam and Kadena. Bombers and submarines would launch missiles against Chinese ships and ports, with the goal of preventing an invasion of Taiwan.
Munitions Shortages
Fundamentally, this is a situation where the US (and its allies like Japan) has a roughly fixed supply of ships and aircraft, which are continuously delivering munitions (precision-guided missiles and torpedoes) to the Chinese fleet. The munitions are “consumables” that get used up; the ships and aircraft are (hopefully!) more durable. So a priori you’d expect the limiting factor on supply to be munitions.
In fact, that’s exactly what national security experts have been observing for years. (See e.g. this 2025 piece from the Foreign Policy Research Institute.) The US has shortages of artillery shells, THAAD anti-missile defenses (which are themselves missiles), and precision-guided missiles.
The key munitions likely to be used in a US-China war over Taiwan include:
AGM-158C LRASM, a bomber-launched anti-ship missile.
AGM-158B JASSM-ER, also a precision-guided long-range anti-ship missile, and also suffering from low production and stockpiles.
Tomahawk cruise missiles, launched from submarines, used against ships
Mk 48 ADCAP torpedoes, also launched from submarines and used against ships
The Heritage Foundation reports that the US only has 250 LRASMs, against a requirement exceeding 1000. We are estimated to have only enough precision-guided munitions for a few weeks of combat, with some specific types, like LRASMs, only lasting for one week. Torpedoes are estimated to be exhausted in months.
It’s not easy to make more of these, either. Two-year lead times are typical. “Annual production rates—115 LRASMs and 79 to 120 MK 48 torpedoes—are orders of magnitude below projected weekly or monthly wartime consumption.”
One thing to keep in mind is that defense contractors are oligopolistic: the same few suppliers make just about everything. LRASMs, JASSMs, and ADCAP torpedoes are manufactured by Lockheed Martin; Tomahawks are manufactured by Raytheon. LRASMs and JASSMs share the same turbofan engine, manufactured exclusively by Williams International.
But what is the fundamental bottleneck in manufacturing munitions? Why can’t Lockheed and Raytheon simply make way more of them?
There are several difficult-to-manufacture components, including radiation-hardened electronics and carbon-fiber motor casings, but the most extreme bottleneck is energetics.
The Energetics Supply Chain is Tiny
Say what you will about American manufacturing, there are still lots of things we’re great at making at scale.
US domestic production of materials like plastics, oil and gas, concrete, and fertilizer is excellent. So is production of high-tech, mass-produced finished goods like pharmaceuticals, automotives, and airplanes.
Even where we’re weaker, at intermediate products like fabricated metal parts, there are still machine shops across the country that are capable of doing the work, they’re just fragmented and inefficient.1
Energetics are not like that.
Explosives and propellants, and their precursors/inputs, are often made exclusively in one facility in the US — or not at all.
There are zero TNT manufacturers in the US.
One facility in Louisiana, GOEX Industries, is the only domestic producer of black powder — and it only recently reopened in 2022 after a shutdown due to an industrial accident.
Only one US company, AMPAC, is certified to provide ammonium perchlorate, an oxidizer used in rocket propellant. (Again, AMPAC scaled down in the wake of a catastrophic industrial accident — the largest non-nuclear explosion in US history, in 1988).
Radford Army Ammunition Plant in Virginia is the only manufacturer of military-grade nitrocellulose, “the key ingredient in the manufacture of all propellants.”
Holston Army Ammunition Plant in Tennessee is the only manufacturer of RDX (cyclonite), the active ingredient in the polymer-bonded explosives used in JASSMs, LRASMs, and similar weapons. Mixing the polymer with the explosive filler, and loading the warheads with the mixture, are similarly only done at a handful of GOCO (government-owned, contractor-operated) facilities.
The Naval Surface Warfare Division facility in Indian Head, Maryland is the only manufacturer of Otto II fuel, which is used in torpedoes.
Why so sparse?
Well, many of these compounds do not have civilian applications, so there is little demand in peacetime.
Sometimes the recipes themselves are classified and require employees to have security clearances.
It can take 2-5 years to qualify a new facility’s output as “milspec” (up to the military’s specifications.)
Also, explosives manufacturing is really dangerous. Industrial accidents happen every year or so, most dramatically the 2025 Accurate Energetic Systems explosion that killed 16 employees. The Holston Army Ammunition Plant is a Superfund site, as are many other army ammunition plants; RDX has contaminated soil and drinking water around the US.
A typical chemical manufacturing facility isn’t good enough for explosives manufacturing; for instance, you need special concrete-reinforced bays to stop potential explosions. There’s substantial regulatory restrictions, due to the unusual safety and environmental hazards of making energetics. So there’s extra expense and expertise involved in making a new explosives plant, compared to other kinds of manufacturing.
Modernizing or constructing new explosive plants seems to cost high tens to hundreds of millions of dollars — $435M to construct a new TNT plant, $93M to restart M6 propellant manufacturing at Radford, $600M to triple production at Holston, etc. (These are all military budget allocations; the actual minimum costs are likely smaller).
Fundamentally, the reason energetics manufacturing has dwindled is that the US military is a monopsony. There is essentially only one buyer; in peacetime, demand is limited; and after the Cold War ended, the US DOD deliberately slimmed down the defense industrial base and shut down “idle” facilities, cutting costs to produce the “peace dividend” budget surplus of the Clinton years. Until recently, manufacturing lots of explosives was simply not a priority for the only buyer of military-grade explosives.
Is There A (Tech) Business Here?
The energetics industry looks really gnarly for new entrants: dirty, dangerous, highly regulated, with volatile demand and an absolute requirement for rare forms of domain expertise.
On the other hand, if you ask “in principle, could new technology help us make energetics more efficiently?” the answer is obviously yes. We’re using a lot of the same processes — in the same facilities! — that we did in WWII. There’s a ton of room for modernization.
First of all, if you ever want to make much higher volumes, you’d want to do as the pharmaceutical and chemical industries do, and switch from batch processing to continuous flow processing.
Second of all, mixing polymer-bonded explosives without causing defects (or accidents!) is fundamentally the same kind of composite manufacturing that you see in the plastics and ceramics industries, it’s just that the “fill” is highly chemically reactive. Like a lot of these materials manufacturing processes, it’s exactly the sort of messy, sensitive process that often doesn’t behave the way the hard-coded physics models say it will — which means it’s heuristically a great candidate for an AI application.
Take a lot of process data to train on, and a big, nonparametric time series model (which is really what a Transformer is), and you can plausibly do a lot better than SOTA on estimating the quality of the final output and what you can do to improve it. Which, in this case, includes the chance it will go boom too soon or not at all.
When it comes to unstable chemistry and complex, non-uniform materials, we do not know with full mechanistic certainty what factors affect process failures — that’s why there are so many industrial accidents in this business. And that’s precisely why “learn from lots of data” and “detect anomalies, even subtle ones that don’t count as “failures” by the official rubric, and try to remediate them” is likely to be a good strategy here.
Admittedly, this is a little bit pie-in-the-sky compared to the challenge of building new manufacturing plants at all. It’s just a pointer in the direction of “yes, going high-tech here could be a substantial improvement in productivity and safety.”
So is anyone starting new companies in this space?
Turns out, yes!
Critical Materials Group came out of stealth in 2026 to make C4 at a new facility in Texas.
Deterrence raised $10.1M in a seed round in 2025 to make explosives with robots.
Supply Energetics, founded in 2025, is manufacturing nitrocellulose in Kansas.
Firehawk, founded 2020, is manufacturing 3D-printed propellant in Oklahoma for solid rocket motors.
X-Bow Systems, founded 2016, makes solid rocket motors, including propellant, in Texas
It’s not a ton, but it’s some evidence that new entrants into this field are viable at all.
It May Be Too Late
If there’s a war in the 2020s, with China or anyone else, that consumes large quantities of munitions, there’s a very serious time crunch for scaling up energetics production.
(And the conflict in Iran definitely isn’t helping the stockpile situation.)
If it takes over a year to get a new facility (or an expansion) built and ready to begin operations, and another year to get the process development optimized and production ramped up, and several more years to get qualified as milspec? Then the timeline doesn’t work for a war before the end of the decade.
The military would have to streamline and speed up its procurement and qualifications process. Or loosen its prohibitions on sourcing energetics from allied foreign nations. Or substitute inferior but more abundant materials. Or cannibalize munitions from elsewhere in the world to focus on an active conflict. None of these are comfortable, business-as-usual choices.
It’s an unfortunate situation. One could make a rah-rah American Dynamism defense-tech pitch here, but it would ring a little hollow; I’m honestly not sure if this problem is going to get solved at all.
Let’s just hope for peace.
And, Austin Vernon argues, American metal fabrication could be globally competitive if it modernized its “back office” operations like quoting and billing, with the magic of ~~software ~~