🌐 Hemispherical Stacks Daily — 2026-03-23

Table of Contents

🛡️ Pentagon Locks Maven AI as Permanent Core System — Palantir Wins $1.3B Permanent Budget Embedding ⛏️ China's Maoniuping Mine Triples Rare Earth Reserves — 9.7M Tons REO Plus 27M Tons Fluorite Discovered 🔬 US Rare Earth Processing Breaks China Monopoly — Saskatchewan AI Plant Delivers 80% Labor Reduction ⚙️ ASML EUV Monopoly Strengthens as High-NA Shipments Begin — Zero Substitutes Lock Global Chip Production 📡 Export License Framework Reshapes Chip Sales — NVIDIA H200 Moves to Case-by-Case Approval 🇦🇺 AUKUS Pillar 2 Advances Despite UK Navy Weakness — Technology Sharing Outpaces Submarine Delays

---

🛡️ Pentagon Locks Maven AI as Permanent Core System — Palantir Wins $1.3B Permanent Budget Embedding

Deputy Secretary of Defense Steve Feinberg's March 9 memo designates Palantir's Maven Smart System as an official "program of record," effective September 30, 2026. The designation removes contract-win uncertainty by embedding Maven into the permanent defense budget cycle, converting what was a time-limited $1.3 billion contract (through 2029) into structurally guaranteed appropriations. Oversight transfers from the National Geospatial Intelligence Agency to the Pentagon's Chief Digital AI Office within 30 days. Future contracts will be managed by the U.S. military directly, signaling Maven's evolution from pilot project to operational backbone.

The program of record status makes Maven substantially harder for future administrations or rival bidders to displace. It establishes Maven as the standard AI operating system for weapons targeting, intelligence fusion, and multi-domain awareness across all branches—Army, Navy, Air Force, Marines, Space Force. Feinberg framed the move as providing warfighters "the latest tools necessary to detect, deter, and dominate our adversaries in all domains." Wedbush analysts noted the designation converts pilot funding into permanent institutional infrastructure, similar to how F-35 or nuclear submarine programs lock in decades of budget allocation.

Palantir stock surged 5% on the announcement. The decision follows 18 months of operational testing in Ukraine, where Maven-powered drone targeting reduced engagement cycles from hours to minutes. Pentagon officials expect full integration across all commands by fiscal year-end. The $13.4 billion Pentagon AI budget for 2026 positions Maven as the central nervous system linking autonomous drones, satellite intelligence, and real-time battlefield coordination.

The designation also signals a broader shift: AI is no longer experimental technology under evaluation but operational infrastructure with budget permanence. For competitors like Google (which exited Project Maven in 2018) after employee protests, Microsoft, and Anthropic, the window to challenge Palantir's position has effectively closed. The Pentagon has chosen its AI vendor, and the decision carries multi-decade implications for who controls algorithmic warfare decision-making.

---

⛏️ China's Maoniuping Mine Triples Rare Earth Reserves — 9.7M Tons REO Plus 27M Tons Fluorite Discovered

Geological surveys at the Maoniuping mine in Sichuan's Mianning county identified 9.7 million tons of rare earth oxides (REO), bringing total proven reserves to 10.4 million tons—making it the world's second-largest producing light REE mine after Bayan Obo. The discovery includes 27.1 million tons of fluorite and 37.2 million tons of baryte at the same site, creating a concentrated hub for three separate strategic mineral supply chains. Wang Denghong, director of the Institute of Mineral Resources at the Chinese Academy of Geological Sciences, called the fluorite and baryte deposits "the truly stunning finds," noting their critical roles in semiconductor etching, lithium-ion batteries, and oil/gas drilling.

Fluorite supplies fluorine-based compounds for semiconductor wafer cleaning and battery electrolyte formulations. Baryte's high density makes it the only viable weighting agent for deep drilling fluids—without it, Wang noted, modern hydrocarbon exploration "would effectively come to a standstill." The Sichuan discovery positions China to set pricing and allocation terms across three markets simultaneously. While baryte deposits exist in the US and India, the scale of this find (37.2M tons) gives Chinese suppliers structural market power. Combined with China's April 2025 halt of rare earth magnet exports and December 2025 introduction of longer-term export licenses, the Maoniuping expansion demonstrates Beijing's strategy: control the processing chokepoint, not just the mining.

The timing aligns with intensifying US-China tech competition. China now holds 9.67 million metric tons of light REE at one site, while the US scrambles to build domestic processing capacity. MP Materials' Mountain Pass mine in California produces concentrates but ships them to China for separation. The Pentagon's January 1, 2027 ban on Chinese-sourced rare earths in defense supply chains creates immediate procurement urgency—but no Western facility can match the scale and integration of Maoniuping.

China's export licensing framework (introduced April 2025, extended validity periods December 2025) reveals asymmetric leverage. European buyers who adapted to licensing requirements saw shipments resume; US exports remain subdued. The Gansu province discovery of 51,455 tons of antimony (a 50% increase in provincial reserves, used in flame retardants for circuit boards and military equipment) further consolidates China's position. The message: China will supply allies on its terms, while adversaries face structural scarcity.

---

🔬 US Rare Earth Processing Breaks China Monopoly — Saskatchewan AI Plant Delivers 80% Labor Reduction

The Saskatchewan Research Council's (SRC) new rare earth separation facility reduces Chinese-style 200-worker operations to approximately 40 workers via AI-powered automation receiving thousands of real-time data points per second for valve and chemical tank adjustments. Built at 25-30% of full-scale Chinese commercial capacity as a technology demonstrator, the plant already produces higher purity metals at greater output than equivalent Chinese facilities. REalloys (NASDAQ: ALOY) holds exclusive offtake rights to approximately 460 tons of defense-grade rare earth metals per year once the plant reaches full production in early 2027, with plans to scale Ohio metallization capacity to 18,000 tons/year of heavy rare earth permanent magnets.

The breakthrough matters because processing, not mining, is the strategic chokepoint. As President Trump noted, "rare processing" is critical to national security—turning concentrates into defense-qualified alloys. Chinese rare earth plants employ 200+ workers managing chemical separation tanks manually; SRC's AI automates the most labor-intensive step (separating 17 chemically similar elements). The plant was designed domestically after China's 2020 export control law restricted access to separation technology. What began as a forced workaround produced a system with better output and zero Chinese supply chain dependencies—no Chinese chemicals, furnaces, anodes, or consumables.

REalloys' Ohio facility converts SRC metals into finished alloys and magnets for the defense industrial base, with confirmed Pentagon contracts. The supply chain runs entirely through allied nations across four continents, with zero Chinese inputs. Graphite anodes (which must be replaced several times weekly in furnaces) come from non-Chinese suppliers—a critical detail, since most Western rare earth projects still depend on Chinese consumables. "1% reliance on China is 100% reliance on China," REalloys executives note.

The Pentagon's January 1, 2027 ban on Chinese-sourced rare earths creates immediate demand. Every F-35 carries 435 kilos of REE; a destroyer needs 4.5 tons; a nuclear submarine requires 1.5 tons. Ukraine produced 1.2 million combat drones in 2024—every magnet manufactured in China. When Trump threatened 100% tariffs, China responded by halting processed rare earth exports. Trump backed off. Ford plants shut down almost immediately during last year's brief export restriction. The Saskatchewan-Ohio chain addresses what Japan learned decades ago: without 2-3 years of strategic stockpiles, just-in-time supply from an adversary is a kill switch on military production.

---

⚙️ ASML EUV Monopoly Strengthens as High-NA Shipments Begin — Zero Substitutes Lock Global Chip Production

ASML Holding's stock recovered to €1,186.40 on March 13, 2026, following confirmation that volume shipments of High-NA (high numerical aperture) EUV tools will begin mid-2026 after successful R&D installations at TSMC and Intel. The Netherlands-based company maintains a complete monopoly on extreme ultraviolet lithography systems—the only technology capable of manufacturing chips below 7nm. Japanese rivals Nikon and Canon abandoned EUV development years ago, leaving ASML as the sole provider for the entire global semiconductor ecosystem. Every advanced AI chip, smartphone processor, and data center accelerator depends on ASML tools with >1 year lead times and zero substitutes.

The High-NA system represents the next generation of EUV, increasing resolution for 2nm and below processes. TSMC's most advanced processes will remain Taiwan-exclusive; what gets built in Arizona lags one generation behind by design. This maintains Taiwan's chokehold on bleeding-edge production while giving Washington just enough capacity to feel secure. ASML's order book stabilized after turbulent 2025 US-China export curbs. The company now projects 2026 P/E at 40.3x, reflecting growth premiums tied to AI chip demand.

China's semiconductor industry faces an ASML wall. SMIC and Hua Hong both achieved 7nm production without EUV by using multi-patterning with older DUV (deep ultraviolet) tools—a workaround that increases costs and reduces yields. Reports suggest SMIC copied TSMC's first-generation N7 process. China's particle beam lithography research (announced March 2026) attempts to bypass ASML entirely, but electron beam approaches face throughput limitations that make mass production uneconomical. For 5nm and below, EUV remains mandatory. US-Netherlands export agreements ban ASML EUV sales to China, creating a hard ceiling on Chinese fab capabilities.

The monopoly extends beyond the tool itself. ASML systems require consumables, maintenance contracts, and software licenses that create operational dependencies even after purchase. The UK's AUKUS submarines provide a parallel: domestic construction still requires US software keys for operation. TSMC's Arizona fabs reduce Taiwan concentration risk but create permanent subsidized dependence on US government funding, deliberate technology lag, and ASML equipment bottlenecks with no substitute supplier. Pentagon rare earth sourcing illustrates the same dynamic: eliminating China processing dependency creates new dependencies on Canadian separation, Ohio metallization, and non-Chinese consumables. The chokepoint moves; it doesn't disappear.

---

📡 Export License Framework Reshapes Chip Sales — NVIDIA H200 Moves to Case-by-Case Approval

The US Commerce Department's January 15, 2026 update moved NVIDIA H200 and AMD MI325X China exports from blanket restriction to case-by-case licensing. NVIDIA confirmed March 17 receipt of government licenses to resume H200 sales to Chinese customers, marking a policy shift from the Trump administration's initial 100% tariff threats. Licensing rules now change monthly, unsettling suppliers, cloud operators, and investors. The framework creates tiered access: allies receive presumed approval, neutral countries face scrutiny, adversaries get denied or conditional licenses. Super Micro Computer co-founder and two executives face DOJ charges for allegedly smuggling $2.5 billion in AI technology to China, demonstrating enforcement gaps in the system.

The oscillating policy creates strategic uncertainty worse than outright bans. Blanket restrictions allow adversaries to plan workarounds (SMIC 7nm development, Huawei chip stockpiling). Monthly licensing shifts force continuous compliance re-evaluation. Cloud operators in Singapore, UAE, and Malaysia don't know if next month's allocation will arrive. Chinese buyers hedge by purchasing through third countries or investing in domestic alternatives. The January 15 rule change came after China halted rare earth exports in response to tariff threats—demonstrating symmetrical supply chain leverage. When the US restricts chips, China restricts materials.

The licensing framework also fragments the global market into geopolitical zones. TSMC Arizona fabs can sell to Pentagon customers with no export license delays. TSMC Taiwan fabs face case-by-case review for certain customers. Samsung South Korea occupies a middle position—allied but geographically vulnerable. As Silicon Canals noted, "Your choice of chip supplier is increasingly a geopolitical alignment, not just a procurement decision." Buyers must evaluate not just performance and price, but which political relationships their vendor requires to maintain supply continuity.

Bernstein analysts predict increased volatility as the US and China refine strategies. The H200 licensing approval suggests a "controlled engagement" approach: allow some sales to maintain market share and gather intelligence on Chinese deployment patterns, while retaining the option to cut access. This mirrors China's rare earth licensing (extended validity for compliant buyers, restrictions for adversaries). Both sides are weaponizing supply chains while maintaining enough trade to avoid immediate economic rupture. The gap between policy rhetoric (decoupling, resilience, independence) and operational reality (case-by-case licenses, monthly rule changes, strategic stockpiling) grows wider.

---

🇦🇺 AUKUS Pillar 2 Advances Despite UK Navy Weakness — Technology Sharing Outpaces Submarine Delays

The UK House of Commons published research on AUKUS Pillar 2 advanced capabilities (quantum, AI, hypersonics, electronic warfare) indicating technology sharing and supply chain integration are progressing faster than Pillar 1 submarine construction. Australia's first AUKUS submarine remains slated for the "early 2040s," with HMS Anson's recent Western Australia visit demonstrating UK-Australian defense industrial cooperation. The Guardian noted the UK's weakened naval capacity raises questions about whether Australia's submarine hopes depend on "the weakest link." Meanwhile, Western Australia partnerships with Babcock and Rolls Royce aim to build skills for long-term AUKUS support, positioning local businesses for decades of maintenance and logistics contracts.

Pillar 2's quiet progress reveals the operational asymmetry: physical platforms (submarines) face 15-20 year timelines, while software, AI, and electronic warfare systems deploy in 2-5 years. The Pentagon's Maven formalization, AUKUS quantum research sharing, and hypersonic development partnerships deliver immediate operational value. Submarines provide strategic presence and nuclear deterrence but require multi-decade capital commitments, institutional knowledge transfer, and supply chain depth (Rolls Royce reactors, Babcock maintenance, US software keys). Pillar 2 technology sharing bypasses those constraints by treating defense capabilities as software-defined systems that can be rapidly iterated and deployed.

The UK's diminished naval capacity doesn't undermine Pillar 2 because technology transfer doesn't require the UK to field the platforms—it requires the UK to share the IP. Australia gets access to British and American sensor fusion, AI targeting (Maven-derived systems), and electronic warfare techniques without waiting for UK shipyards to deliver hulls. The operational model shifts from "buy British submarines" to "integrate British software on Australian platforms." This mirrors the broader defense industry trend toward platform-agnostic capabilities. An F-35 isn't valuable because of the airframe; it's valuable because of the sensor fusion, stealth coatings, and network integration.

The submarine delays also expose the gap between announced programs and operational readiness—a recurring theme across both hemispheres. TSMC Arizona fabs reduce Taiwan risk but lag one generation behind Taiwan production. Pentagon rare earth bans (Jan 1, 2027) create procurement mandates but don't create processing capacity (Saskatchewan produces 460 tons, Pentagon needs thousands). AUKUS Pillar 1 promises submarines in the early 2040s, but Pillar 2 delivers AI and quantum capabilities by 2027-2028. The pattern holds: infrastructure programs announce autonomy, technology programs deliver capability.

---

Research Papers

Note: Academic publications with direct relevance to hemispheric geopolitical infrastructure dynamics typically appear on 2-4 week lag cycles due to peer review and arXiv submission patterns. The 24-36 hour research window for March 23, 2026 yielded limited domain-specific papers. Recent relevant work includes:

"US Microelectronics Packaging Ecosystem: Challenges and Opportunities" — Georgia Tech researchers (October 2023) — Examines CHIPS Act implementation for domestic advanced packaging capabilities. Key finding: bringing fabs onshore without equivalent packaging capacity creates new single-points-of-failure rather than resilience. Authors note that while fab construction receives political attention, packaging infrastructure remains the operational bottleneck that determines actual supply chain security.

"Emerging Ultra-Wide Band Gap Semiconductors for Future High-Frequency Electronics" — University of California materials science team (August 2025) — Addresses gallium supply chain vulnerabilities (DOE energy-critical material classification) and identifies alternative materials for compound semiconductor production. Relevant to rare earth strategic mineral discussions—shows pattern of critical material dependencies extending beyond rare earths to bauxite-derived by-products. Emphasizes that compound semiconductor supply chain resilience requires materials diversification not just manufacturing capacity.

"Human Vs. Machines: Who Wins In Semiconductor Market Forecasting?" — Carnegie Mellon economics group (May 2024) — Compares expert forecasts (WSTS industry polls) against machine learning models for semiconductor demand prediction. Finding: expert forecasts outperform ML in complex, dynamic systems with sparse history. Relevant to supply chain planning under geopolitical uncertainty—algorithmic approaches fail when upstream/downstream linkages shift due to policy rather than market forces.

Academic coverage of Maven formalization, Maoniuping mineral discoveries, and AUKUS Pillar 2 progress will likely appear in security studies and geopolitics journals on 3-6 month cycles. Real-time analysis currently concentrates in think tank publications (CSIS, Brookings, RAND) and defense trade press rather than peer-reviewed venues.

---

Implications

These six stories reveal a structural pattern: proclaimed independence initiatives systematically create new dependencies with different ownership but equivalent operational constraints. Pentagon rare earth bans eliminate Chinese processing dependency while creating permanent reliance on Saskatchewan separation, Ohio metallization, and non-Chinese consumables. TSMC Arizona fabs reduce Taiwan concentration risk while establishing subsidized dependence on US government funding, deliberate one-generation technology lag, and ASML equipment bottlenecks with zero substitutes. China's SMIC 7nm achievement without EUV demonstrates technical workarounds but accepts higher costs, lower yields, and a hard ceiling at 5nm without ASML access.

The Maven formalization exposes the difference between experimental pilots and program-of-record permanence. Palantir didn't win a contract; it won structural budget embedding that makes displacement prohibitively difficult for future administrations. This mirrors ASML's EUV monopoly—not a temporary market position but a chokepoint with 18-24 month lead times and zero alternative vendors. The question stops being "can we build this domestically?" and becomes "which external dependencies do we accept to maintain operation?" True hermetic supply chains don't exist at advanced technology nodes. Autonomy rhetoric masks dependency selection.

China's Maoniuping discovery demonstrates supply chain dominance through vertical integration at the mineral level: one site supplying REO for magnets, fluorite for semiconductors, and baryte for energy infrastructure. The Saskatchewan AI plant counters with automation efficiency (80% labor reduction, higher purity) but operates at 25-30% of Chinese commercial scale. The tonnage gap is vast; the strategic question is whether 460 tons breaks the 100% dependency. Japan answered this decades ago with 2-3 year strategic stockpiles. The US built none. Ford plants shut down during brief Chinese export restrictions. The Trump tariff threat ended when China halted rare earth exports. Supply chain leverage works symmetrically: the US restricts chips, China restricts materials.

AUKUS Pillar 2's technology-sharing progress while Pillar 1 submarines remain 15 years away illustrates the platform-versus-software divergence. Physical systems (submarines, fabs, separation plants) require multi-decade capital, knowledge transfer, and supply chain depth. Software-defined capabilities (Maven AI, sensor fusion, export license frameworks) deploy in 2-5 years. The operational consequence: countries compete on software-defined advantages while announcing multi-decade infrastructure programs that signal intent without delivering near-term capability. The gap between rhetoric (independence, resilience, decoupling) and operational reality (managed dependencies, licensing frameworks, strategic stockpiling) defines the actual geopolitical landscape.

The export license framework's monthly rule changes create worse uncertainty than blanket bans. Adversaries adapt to bans through workarounds (SMIC 7nm, Huawei stockpiling). Monthly shifts force continuous compliance re-evaluation, fragment global markets into geopolitical zones, and convert procurement decisions into political alignments. Both the US and China weaponize supply chains while maintaining enough trade to gather intelligence and avoid immediate economic rupture. Neither achieves autonomy. Both accept managed dependency with escalation options held in reserve. The chokepoints move between rare earths, EUV tools, software keys, and consumables—but they don't disappear. The infrastructure-level question is not whether dependencies exist but whose dependencies you're willing to maintain and which political relationships you need for operational access.

---

HEURISTICS

`yaml heuristics: - id: independence-creates-equivalent-dependencies domain: [geopolitics, supply-chains, infrastructure] when: > Governments invest tens or hundreds of billions in supply chain "independence" through domestic fab construction, allied mineral sourcing, or proprietary technology development explicitly claiming to reduce adversary-nation reliance. Examples: TSMC Arizona subsidies claiming to reduce Taiwan risk, Pentagon rare earth sourcing eliminating China processing, AUKUS submarine programs asserting allied self-sufficiency. prefer: > Systematically map all residual dependencies introduced by independence initiatives before accepting autonomy claims. Track: new single-point failures in equipment bottlenecks (ASML EUV >18mo lead times, zero substitutes), atmospheric or elemental inputs (Qatar helium shipments for semiconductor fabs, days to disruption), software licensing creating operational dependency (UK AUKUS submarines need US keys despite domestic construction), consumable supply chains (graphite anodes requiring weekly replacement with no non-Chinese suppliers), and permanent subsidy requirements (TSMC Arizona loses money without ongoing government support). Measure whether "independence" reduced total dependency or transferred it to different vendors requiring different political relationships. over: > Assuming subsidized domestic manufacturing capacity eliminates geopolitical vulnerabilities or accepting government resilience narratives achieving stated independence goals without creating structurally equivalent or worse dependencies with different ownership requiring different political relationships to maintain operational access. The naive approach treats facility location as sovereignty proxy while ignoring equipment monopolies, consumable dependencies, knowledge transfer failures, and permanent subsidy requirements. because: > TSMC Arizona reduces Taiwan concentration risk but creates permanent subsidized dependence on US funding, deliberate one-generation technology lag (bleeding-edge stays Taiwan-exclusive by design), ASML EUV equipment bottleneck with 18-24mo lead times and zero alternative vendors, and Netherlands export license approval for tool shipments. Pentagon rare earth procurement eliminates China processing compromise but establishes Saskatchewan separation dependency (460 tons annually vs thousands needed), Ohio metallization chokepoint, non-Chinese graphite anode sourcing (weekly consumable replacement), and REalloys offtake exclusivity. UK AUKUS submarines achieve domestic hull construction but require US software keys for operational use, Rolls Royce reactor dependence, and Babcock maintenance contracts. China's SMIC 7nm without EUV demonstrates workaround capability but accepts higher costs, lower yields, and hard 5nm ceiling without ASML access. breaks_when: > True hermetic supply chains emerge at competitive cost without permanent subsidies, equipment monopolies dissolve through genuine substitutes (not just alternative vendors selling the same dependency), consumable inputs achieve multi-source redundancy with <30day switchover capability, or institutional knowledge transfers successfully via capital investment without multi-decade training pipeline requirements. Also fails when dependency transfer improves resilience through allied diversification despite not eliminating chokepoints (five aligned suppliers better than one adversarial monopoly even if none individually sufficient). confidence: high source: report: "Hemispherical Stacks — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1

- id: program-of-record-permanence-vs-pilot-uncertainty domain: [defense-procurement, budget-politics, institutional-inertia] when: > Military or infrastructure technology transitions from experimental pilot, limited-term contract, or provisional deployment to official "program of record" status with structural budget embedding, multi-year appropriations guarantee, and institutional ownership transfer to core military offices. Examples: Palantir Maven AI formalized March 2026 with Pentagon CDAO oversight and automatic appropriations through FY2029+, F-35 Joint Strike Fighter surviving cost overruns through program-of-record protection, nuclear submarine programs with decades-long budget commitments immune to administration changes. prefer: > Recognize program-of-record designation as creating institutional lock-in substantially harder to disrupt than contract value suggests. Track: budget authority transfer from project-specific to core institutional appropriations (Maven from NGA pilot to CDAO permanent system), oversight consolidation under mission-critical offices (not R&D or innovation labs), procurement timeline extension from 1-3yr contracts to 10-30yr programs, and displacement barriers for competitors (new vendors must replace embedded infrastructure not just win contract competitions). Understand that once formalized, reversing program-of-record status requires Congressional authorization battles, not just contract rebids. over: > Treating program-of-record designation as equivalent to large contract wins, assuming competitive procurement can displace embedded systems through better technology or lower cost, or believing administration changes can easily terminate programs once they achieve permanent status. The naive view sees dollars and timelines without recognizing institutional inertia, sunk cost political dynamics, and bureaucratic resistance to system replacement requiring workforce retraining and doctrine revision. because: > Palantir Maven formalization (effective Sept 2026) converts $1.3B limited-term contract into permanent budget appropriations with oversight transfer from NGA (geospatial intelligence project) to Pentagon CDAO (core AI office), making displacement require Congressional action not just contract competition. Feinberg memo positions Maven as operational backbone ("detect, deter, dominate in all domains") not experimental capability. Google exited Maven 2018 due employee protests; by 2026 window to challenge Palantir closed because Maven became infrastructure not vendor relationship. F-35 program survived $1.7T costs and persistent technical issues through program-of-record protection creating jobs in 45+ Congressional districts. Competitors (Microsoft, Anthropic) face institutional switching costs: retraining thousands of operators, revising doctrine, replacing embedded integrations across Army/Navy/Air Force/Marines/Space Force simultaneously. breaks_when: > Catastrophic system failures create political pressure exceeding bureaucratic inertia (F-35 total combat ineffectiveness forcing replacement), geopolitical shifts making vendor nationality unacceptable (Palantir acquisition by adversary forcing system change), or technological discontinuities rendering entire program category obsolete (hypersonics making aircraft carriers unviable forcing doctrine revision regardless of sunk costs). Also fails when program remains pilot-stage beyond 5-7yr window without formalization, creating vulnerability to budget cuts or competitor displacement during renewal. confidence: high source: report: "Hemispherical Stacks — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1

- id: monthly-licensing-worse-than-blanket-bans domain: [export-controls, strategic-uncertainty, supply-chain-planning] when: > Export control regimes shift from blanket approvals or denials to case-by-case monthly licensing frameworks with unpredictable rule changes. Examples: US Commerce Dept moving NVIDIA H200/AMD MI325X China exports from blanket restriction to monthly case-by-case approval (Jan 15, 2026), China rare earth export licenses with extended validity for compliant buyers but monthly allocation uncertainty for others (Dec 2025), semiconductor equipment export rules changing quarterly based on geopolitical conditions. prefer: > Expect monthly licensing frameworks to create worse strategic uncertainty than blanket bans because adversaries can plan workarounds for permanent restrictions (SMIC 7nm development, Huawei chip stockpiling, domestic fab investment) but cannot optimize supply chains for oscillating monthly approvals. Track: buyer hedging through third-country purchases, accelerated domestic alternative investment despite lower performance, fragmentation of global markets into geopolitical zones requiring separate procurement strategies, and compliance overhead costs exceeding tariff/restriction costs for borderline cases. Recognize controlled engagement (some approvals to maintain market share and gather deployment intelligence) differs from strategic decoupling (complete cutoff forcing immediate workarounds). over: > Assuming licensing frameworks reduce tensions through flexibility or that case-by-case review provides better security than blanket bans. The naive view treats monthly rule changes as policy refinement rather than weaponized uncertainty. It assumes buyers accept compliance costs and wait for approvals rather than investing in alternatives, smuggling through third countries, or stockpiling during approval windows. It misses that adversaries prefer clarity (even if restrictive) over monthly uncertainty preventing multi-year planning. because: > US H200 licensing approval (March 17, 2026) followed Trump 100% tariff threats that ended when China halted rare earth exports, demonstrating symmetrical supply chain leverage and oscillating policy. Super Micro Computer $2.5B smuggling charges show enforcement gaps in licensing regimes incentivizing third-country workarounds. Monthly rule changes force cloud operators (Singapore, UAE, Malaysia) to continuously re-evaluate compliance and hedge through geographic diversification. China's rare earth licensing (extended validity for European buyers who adapted, continued restrictions for US buyers) fragments markets and rewards policy compliance. Bernstein analysts predict increased volatility as both sides "refine strategies"—meaning neither commits to full decoupling or unrestricted trade, maintaining managed dependency with escalation options. The gap between rhetoric (independence, resilience) and reality (case-by-case licenses, strategic stockpiling) widens. breaks_when: > One side commits to complete decoupling making licensing irrelevant (blanket bans with zero exceptions forcing immediate workarounds), licensing becomes fully predictable with multi-year stability reducing uncertainty to manageable compliance overhead, or third-party arbitrage (selling through neutral countries) becomes so prevalent that licensing loses effectiveness and reverts to blanket controls. Also fails when buyer power concentration (TSMC, ASML monopolies) makes denials too costly for restricting country, forcing de facto approvals despite licensing framework. confidence: medium source: report: "Hemispherical Stacks — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1 `