Part 2: The Stack
The Countdown Clock had two hands. It has three.
The Countdown Clock had two hands. It has three. July 13, 2026. November 10, 2026. January 1, 2027.
Three deadlines. Three different rooms. Three different years. No coordination between any of them. All converging on the same supply chain in the next seven months.
Part 1 introduced the January 1 deadline - the date when DFARS 252.225-7052 takes full statutory effect, barring any defense contractor from delivering covered materials - neodymium-iron-boron magnets, tantalum, tungsten - that were mined, refined, separated, melted, or produced in China, Russia, Iran, or North Korea at any point in the supply chain. That deadline hasn’t moved. What has changed is the context around it.
July 13 is when the Section 232 report lands. Proclamation 11001, signed January 14, 2026, set the clock. If negotiations have not produced satisfactory results by that date, four tools become available: tariffs, minimum import prices, quotas, further Section 232 actions. Every one of those tools is bullish for domestically aligned producers and bearish for any supply chain still running through China.
November 10 is when China’s suspended rare earth export controls expire. One nuance worth stating plainly: reporting from Politico and the National Security Daily suggests the suspension may never have covered US military end-use at all - that the defense supply chain was already operating without access to Chinese material throughout the suspension period. If confirmed, November 10 matters more for commercial supply chains than for defense. The implication is sharper, not softer: the defense supply chain gap is older and deeper than the November 10 date implies. The clock started earlier than most people think.
Plain English: Three deadlines set by three different agencies in three different years. None of them knew about the others. All of them land on the same supply chain. That supply chain is not ready for any of them.
Jensen Huang describes AI infrastructure as a five-layer cake: energy at the base, then chips, infrastructure, models, applications. He says energy is the floor. Everything else rests on it.
The Chokepoint starts one layer below that. At the materials required to generate the electrons, move them, and secure the systems that run on them.
Call it the sovereign AI stack: materials, compute, power, bandwidth, trust. Five layers. The argument of this post is simple. None of them is sovereign. All five face a deadline problem converging on January 1, 2027. And in the past two weeks, China has begun actively closing the one layer it controls most completely.
Each layer has to work. None of them is optional. And none of them is ready.
The Materials Layer
DFARS 252.225-7052 requires that by January 1, 2027, the entire rare earth magnet supply chain feeding US defense systems be free of Chinese involvement. The statutory language is precise: not just the finished magnet, but every upstream step - any covered material “mined, refined, separated, melted, or produced” in a covered country is barred from any defense contract delivery after that date.
What that chain currently looks like: MP Materials (NYSE: MP) operates Mountain Pass in California, the only rare earth mining and processing site of scale in the Western Hemisphere, producing neodymium-praseodymium oxide - the critical input for NdFeB (neodymium-iron-boron - the permanent magnet technology inside every EV motor, wind turbine generator, and defense guidance system) magnets that DFARS targets. Lynas Rare Earths (OTC: LYSDY), the Australian producer, operates the only significant non-Chinese separation facility outside the continent. USA Rare Earth (NASDAQ: USAR) is building an integrated mine-to-magnet supply chain anchored at Round Top in West Texas, with a $1.2 billion South Carolina magnet facility and a newly announced €175 million French expansion. Energy Fuels (NYSE: UUUU) processes monazite - a rare earth-bearing mineral - through its White Mesa mill in Utah, the only operating facility of its kind in the US. Neo Performance Materials (OTC: NOPMF) manufactures rare earth magnetic powders and bonded magnets across facilities in Europe and Asia. Almonty Industries (NASDAQ: ALM) is the tungsten thread of the same law - Sangdong mine in South Korea, the largest tungsten deposit outside China, advancing toward production for a supply chain that faces the same structural exposure as rare earths.
The official verdict, from the Center for Strategic and International Studies, published April 27, 2026: “Unless significantly more capacity comes online in the next eight months, adhering to this requirement may not be feasible.” Authors: Dr. Gracelin Baskaran and Meredith Schwartz, Critical Minerals Security Program. That is the top US defense think tank, telling Congress the supply chain it funded isn’t ready, seven months before the deadline.
Proven recycling routes exist - short-loop and long-loop processes that recover rare earth material from end-of-life magnets. Neither delivers DFARS-compliant volume by January 1. Recycling is part of the long-term answer. It is not the January 1 answer.
Plain English: The law says the magnets must be American by January 1. The factories that would make them compliant are not finished. The law does not move. The factories do not appear on schedule. That is the materials layer.
Now add what China did this week.
On June 15, 2026 - a week from today - a 79-article regulatory framework issued by Premier Li Qiang takes effect. It controls total mining output, restricts which entities can participate in mining, and mandates national security reviews on all foreign investments in Chinese mining operations. Strategic mineral reserves must be stored at source for a minimum of five years.
The most important feature of this framework is not what it covers. It is how it is structured. The list of minerals subject to control is dynamic. It can be adjusted based on economic importance, national security, domestic requirements, and supply chain resilience - with approximately 14 days notice.
Fixed export controls can be planned around. Procurement teams model around known lists, build inventory buffers, reroute through allies. Every Western compliance program built over the past three years - DFARS, Project Vault, Section 232 - was designed around one assumption: knowing which minerals are restricted. A list that can add manganese, tungsten, or bismuth with 14 days notice makes that assumption invalid.
Then on July 1, 2026 - four days before the Section 232 report lands - a second framework takes effect. State Council regulations giving Beijing the legal basis to force unwinding of completed overseas transactions, ban cross-border talent transfers in sensitive sectors without approval, and punish foreign firms whose home countries restrict Chinese investment. Han Shen Lin of The Asia Group described it plainly: it codifies a “full retaliatory toolkit against US entities that participate in outbound investment screening of Chinese capital”.
June 15 controls what flows in - foreign investment in Chinese mining faces mandatory security review. July 1 controls what flows out - technology, talent, data, strategic assets. One architecture. Two directions. Published the same week.
Plain English: The West is trying to build compliance programs around a known list of restricted minerals. China just made the list unknowable. Western programs assume a static regulatory environment. China has made the environment dynamic. That is not a coincidence. That is a response.
The materials layer is not sovereign. It is not becoming sovereign by January 1. And China just made it significantly harder to plan around.
The Compute Layer
Export controls were supposed to close the compute layer from the top. The logic was straightforward: restrict China’s access to advanced semiconductors and its military AI development slows.
In late 2025, the Trump administration authorized the export of Nvidia’s H200 chips - the most powerful AI processors the US has ever permitted for sale in China - to Chinese commercial buyers. Military-linked institutions cannot obtain export licenses. So they are pursuing the chips anyway.
Bloomberg reviewed procurement records and confirmed that at least seven Chinese universities supporting the People’s Liberation Army and the defense industry are actively seeking H200 access through third-party brokers, rental arrangements, and procurement contracts. Two of those institutions - Beihang University and Northwestern Polytechnical University - are on the US Commerce Department’s Entity List for their work advancing Chinese military capability. Wirescreen’s analysis, obtained by the New York Times, documented over 500 PLA procurement records seeking Nvidia chips across a six-year period. Reuters confirmed the mechanism: Chinese entities are increasingly renting compute time on servers fitted with restricted chips as a workaround to physical importation. The chips do not need to cross a border. The access does.
This is not simply an export control failure. It is an export control architecture problem. Partial decontrol created a gray zone. Military-linked institutions are exploiting that gray zone at the edges while licensed channels remain closed to them. Congress has since introduced legislation to strip the White House of unilateral decision-making authority on chip exports to China. The architecture is being contested in real time.
DFARS closes the materials layer from below. Export controls were supposed to close the compute layer from above. The materials closure is not built. The compute layer has a gap that military-linked institutions are actively navigating.
Leonardo DRS (NASDAQ: DRS) sits at the intersection of this problem rather than above it. DRS designs and manufactures the sensing, network computing, and electronic warfare systems that run inside defense platforms - the edge AI hardware that must function within DFARS-compliant supply chains at the system level. Its Q1 2026 results beat consensus, management cited improved supply conditions for key inputs including germanium, and the company raised full-year guidance. It is not immune to the compute sovereignty problem. It is the company that has to solve it in hardware, platform by platform.
Kratos Defense (NASDAQ: KTOS) represents the autonomous weapons layer - the systems that run on the constrained supply chain. Kratos’s attritable drone and hypersonic test vehicle programs are the physical expression of what the compute and materials layers are supposed to enable. The Drone Dominance Program’s 30,000 one-way attack drones - every one runs on an NdFeB motor, every one must be DFARS-compliant by January 1. That program was contracted under the assumption that a compliant supply chain would exist by the deadline. That assumption has not been validated.
Plain English: The plan was to squeeze from both ends - no materials going in through DFARS, no chips going out through export controls. Partial decontrol opened a gray zone at the top. The materials build-out is behind schedule at the bottom. Military-linked institutions are navigating both gaps simultaneously.
The compute layer is not closed. The controls created a gray zone that is being actively exploited.
The Power and Bandwidth Layers
The power layer is the Strait of Hormuz.
IMF PortWatch recorded 4 vessel transits per day on May 24, 2026 - the low point - recovering marginally to 10 on May 31, against a pre-crisis baseline of 95 per day. That is still less than 11% of normal volume. The strait is effectively severed at the aggregate level. War-risk insurance for tankers is pricing at 8 times pre-crisis rates. Six protection and indemnity clubs have withdrawn cover entirely.
And yet some cargo is moving. Not through free navigation - through selective bilateral transit arrangements that Iran is administering on its own terms. TotalEnergies CEO Patrick Pouyanné said publicly that paying a toll was preferable to closure. Japan received the offer. Some shipowners are clearing cargo this week because they cannot wait for geopolitics to resolve. Iran is not closing the strait uniformly. It is choosing who transits, at what price, under what conditions.
These are the same fact at different scales. At the macro level: a severed artery. At the micro level: a toll booth administered by the party that controls the chokepoint. The strait is near-closed at the system level and selectively open at the bilateral arrangement level. That combination is more dangerous than either full closure or full reopening - because it institutionalizes Iranian control over who gets access, at what price, under what political conditions, for as long as the arrangement holds.
The ADNOC trading chief, reported by Reuters: supply chains could take one year to recover even after flows normalize. The year-long recovery clock does not start until the strait reopens. It has not reopened.
The bandwidth layer is under active construction. Rocket Lab (NASDAQ: RKLB) passed System Requirements Review for the Space Development Agency’s Tracking Layer Tranche 3 constellation in May 2026 and holds over $1.3 billion in SDA-related awards. The Golden Dome architecture requires secure satellite datalinks - SATCOM - connecting interceptors across a proliferated low-earth-orbit constellation. That infrastructure is being built. But it is being built by commercial providers operating under commercial contracts, not treaty relationships. The bandwidth layer’s vulnerability is not physical. It is structural. A commercial contract can be renegotiated, terminated, or repriced under political pressure. A sovereign treaty cannot. The Golden Dome is being built on the former.
Plain English: The power layer is a severed artery with a toll booth on the one pipe that still runs. The bandwidth layer is being built by companies, not nations. Neither of those is sovereignty. Both are dependencies with different flags on them.
The Trust Layer
January 1, 2027 is not only the DFARS deadline. It is also the date by which all new acquisitions for National Security Systems are required to be compliant with NSA CNSA 2.0 - the Commercial National Security Algorithm Suite 2.0, which mandates post-quantum cryptography (PQC - encryption algorithms designed to resist attacks from quantum computers, which can break the RSA and elliptic curve encryption standards currently protecting classified communications) across classified government systems.
This is a hard procurement requirement. Under CNSSP 15 - the NSA’s binding policy document - a new National Security System delivered after January 1 without CNSA 2.0 compliance is non-compliant on arrival. Not a guideline. Not a recommendation. A procurement gate: any new acquisition for a classified system that cannot demonstrate quantum-resistant cryptography does not get procured.
The magnet rule and the cryptography mandate were written by different agencies in different years for entirely different reasons. They land on the same date.
This is the trust layer of what Shanaka Anslem Perera has described as the sovereign AI stack: a sovereign that cannot power, source, compute, connect, and verify its own AI rents a revocable decision loop. The trust layer is where verification happens - where a defense system confirms its communications have not been intercepted, its data has not been corrupted, its commands have not been spoofed. A system running pre-quantum cryptography is vulnerable to harvest-now-decrypt-later attacks regardless of whether its magnets are China-free. Adversaries collect encrypted traffic today and decrypt it when quantum capability arrives. For new National Security System acquisitions, the window to close that vulnerability is not 2031 or 2035. It is January 1, 2027.
Both layers - materials and trust - have to close simultaneously. Neither is currently on track.
Plain English: The magnet rule and the cryptography rule weren’t coordinated. Different agencies, different years, same date. Fix the magnets but not the cryptography - the system can still be broken into. Fix the cryptography but not the magnets - the system cannot be built. Both have to work. Neither is there.
The trust layer and the materials layer share a deadline that no one in either agency set intentionally.
Run it forward. A defense system clears DFARS - magnets sourced, chain documented, January 1 compliant. It fails CNSA 2.0 certification - cryptography not upgraded, non-compliant on arrival. Its power supply runs through a strait at 4% of pre-crisis volume. Its command links run on a commercial satellite contract. The stack doesn’t degrade gracefully. It doesn’t function. Not because any single layer was ignored - but because all five had to work simultaneously, and the assumption was that someone else was solving the other four.
What This Means
Washington has mandated a sovereign AI stack. It has not built one.
The mandates are real. DFARS is real. CNSA 2.0 is real. Section 232 is real. The three deadlines are real. What is not real - not yet - is the stack underneath them. The materials are not compliant. The compute layer has a gray zone being actively navigated by the adversary it was designed to contain. The cryptography transition is a hard procurement requirement with seven months left on the clock. The energy supply runs through a chokepoint at 4% of pre-crisis volume. The bandwidth layer is commercially contracted.
Closing all five simultaneously requires physical inputs - metals, processing capacity, fabrication facilities - that don’t yet exist at the scale the law assumes.
The war was the catalyst. But the law is the structure. And the law arrives before the thing it requires exists.
Part 1 showed the gap in the materials layer and named who is trying to close it. Part 2 has shown that the gap is not one layer deep. It is five layers deep, with a different owner, a different timeline, and a different kind of exposure at each level.
The question that follows is not whether the stack needs to be built. That question has been answered by three separate deadlines set by three separate rooms. The question is who the gatekeepers are.
That is Part 3.
Sources
DFARS 252.225-7052 (Federal Register, May 2024; eCFR; Acquisition.gov) | Proclamation 11001 (Federal Register, January 14, 2026) | Center for Strategic and International Studies, “Rare Earth Export Restrictions One Year Later,” Gracelin Baskaran and Meredith Schwartz, April 27, 2026 (csis.org) | NSA Commercial National Security Algorithm Suite 2.0 (CNSSP 15, updated May 2025) | Bloomberg, “Nvidia’s H200 AI Chips Targeted by China’s Defense-Focused Universities,” June 1–2, 2026 | Wirescreen analysis via New York Times, June 2026 | Reuters, “How Chinese entities are already using Nvidia’s H200 AI chips,” June 2026 | Han Shen Lin, The Asia Group, via Reuters (Laurie Chen and Eduardo Baptista), June 1, 2026 | IMF PortWatch May 24 and 31, 2026 | Reuters citing ADNOC trading chief, June 3, 2026 | Rocket Lab Corporation press releases, May 2026 (Globe Newswire). Stock prices verified day of publish. All prices cited for context only. Facts verified June 2026.
Disclaimer
This post is for informational purposes only and is not investment advice. The Chokepoint is an independent investment research publication. Nothing in this publication should be construed as a recommendation to buy, sell, or hold any security. All company references and price data are provided for informational and contextual purposes only. Conduct independent due diligence and consult a qualified financial advisor before making any investment decisions.