Industrial Capabilities Report to Congress
This is a report that the US DoD makes every year on the defense industrial base, ie., the network of factories, supply chains, contractors, workers, and other stuff that the military runs on. It’s remarkable how fast the tone went from “yup, we have defense contractors and they make defense stuff for us” to “holy fucking shit, we have some major problems and we need to fundamentally rethink how we do absolutely everything”. Here’s the 2016 version , it’s extremely boring.
But the 2020 version starts off with: “Together, a U.S. business climate that has favored short-term shareholder earnings (versus long-term capital investment), deindustrialization, and an abstract, radical vision of ‘free trade,’ without fair trade enforcement, have severely damaged America’s ability to arm itself today and in the future. Our national responses — off-shoring and out-sourcing — have been inadequate and ultimately self-defeating, especially with respect to the defense industrial base.”
It goes from barely even mentioning software to: “Software acquisition remains one of the most expensive and most complex sectors in the DoD. For example, the F-35 Joint Strike Fighter has required more than eight million lines of code, almost all of which had to be written by its prime contractor and sub-contractors, virtually from scratch and, then again, after Chinese cyber-theft. All software ‘blocks’ — the systems designed to take the plane from testing to full production — experienced serious production and budgetary delays. These, in turn, contributed to expanding the Lightning II’s total price tag.
One could argue that today’s defense systems are no more or less than physical platforms for software, yet developing and buying that software had become a major bottleneck. Standard Pentagon programming was not designed to deal with software, so crucial to operating systems large and small, including networked warfare.
The Department of Defense has traditionally acquired IT and software-based systems in the way it bought aircraft carriers — as if they were physical items to be forged or welded or mass-produced. The standard acquisition cycle has been geared around multiyear milestones and intensive evaluation reviews that can take months or years. The modern software development cycle, by contrast, moves in weeks, days, and even hours and seconds — because software is a digital item, subject to real-time improvement and innovation, whose only limits are the human imagination and the speed of an electron. To take one example, given the unique iterative dynamic of software development, the Pentagon’s traditional serial approach to ‘the color of money’ — different budget accounts for development, production, and sustainment — was a major obstacle.”
And, hmmmm, that’s a bad sign: “The United States continues to lead the world in gross domestic spending on R&D in 2019, although China is rapidly and consistently closing the gap. Nonetheless, aerospace and defense companies are among the lowest R&D spenders compared to other critical sectors. America’s six biggest defense contractors have spent on average 2.5 percent of their sales on R&D each year. This compares to 10 percent of sales for ‘big tech’ firms like Facebook, Amazon, and Google.”
and, uh, yeah, that’s a problem: “…the number of cases, typically three to seven levels from the top of the supply chain, where there is just one — often fragile — supplier is staggering. This represents a significant deterioration from just a decade ago when three-to-five suppliers existed for the same component, let alone several decades ago, when the U.S. military generally enjoyed dozens of suppliers for each such item.”
And, well, you don’t say: “According to data released in late 2019 by the U.S. Census Bureau, approximately 291,000 manufacturing establishments operate in the United States. Nearly 99 percent of those establishments are small and medium-sized manufacturers (SMMs) with fewer than 500 employees. Multiple data sources indicate that most SMMs are unprepared to deal with a cyber-attack. This problem is acute within defense manufacturing supply chains, where SMMs — often lacking basic cyber controls — constitute the bulk of the critical lower supply chain tiers.”
And this part is just funny: “DoD has relied on a sole source for Service Dress Uniform fabrics for a number of years, as well as sources of fibers that protect against flame and ballistic threats, and many other essential components. As a result of [diminishing manufacturing sources and material shortages] from domestic suppliers, [the Defense Logistics Agency] has considered seeking a Domestic Non-Availability Determination for Service Dress Uniforms.”
And it goes on and on and on, biotechnology, hypersonics, battery supplies, machine learning, quantum sensors, laser/microwave weapons, missiles, etc. At least someone somewhere is trying. ( Full 2020 report .)
On the flip side, the problem of “everything gets outsourced to China” is somewhat self-limiting as it becomes riskier for American companies to do business there: “On October 20, 2019, in How to Avoid China Prisons: Know YOUR China Risks , we wrote how ‘every foreigner and foreign company is at risk in China; it’s just a question of how much’ and then we laid out how to reduce your risks.” ( Harris Bricken, China Law Blog ) Stupidity in defense contracting certainly isn’t new ( classic clip ). But you can get away with that if a) economic forces incentivize locating all of your production in the United States or a handful of close allies, b) your opponents are the Soviets who are doing something similar (or, better yet, Iraqi/Afghani insurgents with no advanced weapons), c) the rate of technological progress is low enough that a weapons system can last for decades (the B-52 was introduced in 1952 and is planned to last until the 2050s), and d) the base technology is well-understood enough that you can have lots of redundant suppliers at all levels. None of those things seem to be true anymore. In particular d) seems to break down whenever you’re pushing the tech frontier in a specialized area, as we see with mRNA vaccines or semiconductor fabs.
Hacker News on the difficulties of setting up fab supply chains: “Major suppliers include: Nikon, ASML, KLA-Tencor, Applied Materials (AMAT), Lam Research. The timeline they are suggesting isn’t literally impossible as a matter of physics but is, to me, laughably unrealistic and likely done to try and extract tax incentives. Unless they have already broken ground in Texas… and are just shopping for a better deal.”
“Add like 100 more companies to the list. A fab is not only a lithographic line, but a ton of other very unique, industry specific equipment. Air handling for fabs stuff is up to 24 months lead time, and has no use in any other industry. Ultra clean water plants — 9 to 18 months. Just two vendors on the planet. Gas plants — take few years to construct by themselves. Chem plants — again years long lead time equipment from only one vendor in the world. On site mask shops — equipment, and processes not much cheaper, or less scarce than the last gen chip making gear itself. Cleaning stations, degassing stations, industry specific SCADA hardware, AMHS… Post-backend stuff… The list will go on for a few screens.”
The Biden administration has ordered a review of supply chains, China dependency and domestic production possibilities for key sectors ; text of the executive order . The Pentagon’s industrial base office is also being promoted more generally .
Biden obviously wants domestic manufacturers in his political coalition, but in the case of TSMC/Samsung they aren’t even domestic companies, they’re foreign companies that the US essentially just bribed to put the plants there . The current political pressure likely follows from a bunch of manufacturing plants shutting down due to chip shortages , which seems to have partly been a result of not enough slack in the system / bullwhip effect and partly a result of US sanctions on SMIC, China’s largest supplier.