Hemispherical Stacks · 2026-04-25

🌐 Hemispherical Stacks — 2026-04-25

💰 China Mandates State Approval for AI Firms Accepting US Capital as Google Commits $40B to Anthropic
🧱 Anthropic's Passport Gate Constructs Western AI's First Technical Border Wall
📜 WTO Digital Trade Moratorium Lapses for First Time in 27 Years After Yaoundé Failure
🔑 BIS IC Designer Extension Reveals US Export Control Architecture's Managed-Exemption Turn
⚓ PLA Live-Fire Drills East of Luzon Coincide with Balikatan as South China Sea Physical Stack Hardens
🚗 Chinese Manufacturing Embeds in European Hemisphere as Electrolux-Midea and SAIC-Spain Advance

---

💰 China Mandates State Approval for AI Firms Accepting US Capital as Google Commits $40B to Anthropic

China plans to require government approval before top AI companies — including leading startups — can accept US investment, Bloomberg reported April 24. The mechanism would formalize what has operated informally: party-state coordination over which foreign capital can enter China's AI sector, who it reaches, and on what terms. On the same day, the New York Times reported Google committed up to $40 billion to Anthropic — the largest single corporate commitment to an AI lab in history — cementing the Western AI stack's capital concentration around a handful of frontier labs.

Taken together, these two moves constitute the completion of a symmetric capital isolation architecture. The US outbound investment executive order (August 2023, final rules December 2024) restricts US capital from flowing into Chinese AI, semiconductor, and quantum entities above defined thresholds. China's proposed inbound approval mechanism is the mirror: state gatekeeping over US capital entering Chinese AI. The architecture is now bilateral. Capital that previously moved freely between the two hemispheres' AI sectors now requires explicit sovereign permission in both directions.

The structural implication runs deeper than bilateral symmetry. Google's $40B Anthropic commitment — following the $2B investment in 2023, the $750M top-up in 2024, and Amazon's cumulative $8B — means the dominant Western frontier AI labs (Anthropic, OpenAI, Google DeepMind) are now capitalized almost entirely by hyperscaler investment rather than neutral venture capital. Both hemispheres are converging on state-adjacent capital structures for AI: in China, explicit state approval; in the US, hyperscaler investment that aligns lab incentives with cloud infrastructure providers who are themselves deeply embedded in US national security contracting. Neither architecture is capital-neutral; both embed governance logic into funding structure.

The Lee and Liu analysis in the Oxford University Press volume argues US export restrictions — on chips, models, and outbound investments — maintain American technical advantage even as open-model proliferation creates alternative pathways. China's new inbound approval mechanism closes that alternative pathway from the other side: if US-backed startups can't receive US capital, their scaling options narrow to domestic pools and non-US foreign capital. The 5-10 year consequence is two AI sectors that are not just technically differentiated but structurally incapable of cross-investment — a capital decoupling that will lock in divergent capability trajectories independent of whatever technical convergence might otherwise occur.

Sources:

---

🧱 Anthropic's Passport Gate Constructs Western AI's First Technical Border Wall

Anthropic introduced identity verification for Claude requiring users to complete a real-time selfie check while holding a government-issued photo ID — passport, driver's license, or national ID card. As TechNode reported April 16, the mechanism is being applied gradually in specific scenarios: access to advanced features, platform integrity checks, and compliance enforcement. The practical effect for Chinese users is immediate and structural. Individuals without passports are excluded. Those with passports face real-name verification concerns in a geopolitical environment where submitting biometric data to a US company carries meaningful risk calculus. Older Claude accounts — created before verification — have become traded assets.

The verification mechanism is not a geofence or an IP block. It is an identity-layer control operating at the application stratum of the AI stack. Previous access restrictions in this domain operated at the network layer (VPN-counterable) or the API-key layer (workaroundable with institutional credentials). An identity-layer requirement tied to government document submission is qualitatively different: it requires a physical credential tied to a state, which China's domestic AI platforms — Ernie, Qwen, DeepSeek — do not require for basic access. The TechNode analysis is direct: "Claude's move may only be the beginning. If proven effective, other AI companies could follow suit, fundamentally reshaping the ecosystem of overseas AI services."

The cross-hemisphere dynamic this creates is exactly what China's domestic AI substitution strategy needs. Each time a Western frontier model raises its verification threshold, the value proposition of domestic alternatives increases. DeepSeek V4, the latest release from the Hangzhou-based laboratory, continues to close the capability gap with Western frontier models on reasoning benchmarks. Access bifurcation and capability convergence are running simultaneously: the frontier is accessible in the West and inaccessible in China, while Chinese domestic alternatives are competitive enough to be functional substitutes for most use cases. The Lee and Liu framework argues export restrictions on models presuppose US technical advantage — but identity-layer access controls accelerate domestic substitution by forcing the issue before capability parity is reached.

The structural difference from semiconductor export controls is instructive. BIS chip controls operate at the hardware production layer and require enforcement at manufacturing chokepoints (ASML, TSMC). Claude's identity verification operates at the inference delivery layer and requires no external enforcement infrastructure — Anthropic implements it unilaterally. The compliance overhead is entirely on the user side. From a strategic standpoint, this is an access control mechanism that costs the implementer almost nothing, imposes real friction on excluded users, and — unlike chip controls — cannot be circumvented by domestic manufacturing investment no matter how large. Chinese AI labs cannot manufacture their way out of a Passport Gate.

Sources:

---

📜 WTO Digital Trade Moratorium Lapses for First Time in 27 Years After Yaoundé Failure

The WTO Ministerial Conference in Yaoundé, Cameroon failed to reach agreement on permanently prohibiting duties on electronic transmissions, allowing the Moratorium on Customs Duties on Electronic Transmissions — in place since 1998 — to lapse for the first time in 27 years. The Semiconductor Industry Association expressed alarm on April 1: "This commonsense measure has promoted growth and innovation across the global digital economy for decades, and it is deeply disappointing that WTO members allowed it to lapse." The mechanism covered electronic transmissions broadly — including semiconductor design files, software services, AI model weights transferred across borders, and digital goods. Countries can now, in principle, impose tariffs on these flows.

The governance failure dissects across hemispheric lines. The US and its allied partners had pushed for permanence. Resistance came from developing economies — India, Indonesia, South Africa — who wanted flexibility to use tariff tools on digital goods for industrial policy purposes. China's position was ambiguous: formally in support of e-commerce facilitation but resistant to specific provisions that would restrict domestic digital sovereignty measures. The result is a governance vacuum that primarily benefits actors willing to use new tariff authority unilaterally, and that primarily harms actors whose digital goods flow most intensively across borders — which, at the moment, means US and allied tech companies.

The immediate consequence is uncertainty rather than immediate tariff imposition. WTO members must positively enact new tariff schedules to impose duties; the moratorium lapse does not automatically create tariffs. But it removes the institutional constraint. A country that wants to impose a 10% duty on AI inference API calls originating from a foreign server now has WTO-compliant space to do so. The Neuwirth analysis of AI in the digital economy from an Indo-Pacific perspective predates the Yaoundé failure but anticipates exactly this dynamics: the gap between multilateral governance frameworks and the actual infrastructure of AI service delivery creates exploitable regulatory vacuums at each governance jurisdiction boundary.

The semiconductor-specific risk is the most acute. SIA members transfer billions of dollars in semiconductor design files, EDA software, and process recipes across borders annually. A tariff on electronic transmissions that captures design file transfers would directly tax the distributed R&D model that underlies US-allied semiconductor leadership. TSMC manufactures chips designed in California using process recipes transferred from its R&D labs. Any tariff regime that treats those transfers as taxable transmissions would impose compliance costs on the very supply chain integration that the CHIPS Act is trying to build. The WTO lapse is a governance-layer chokepoint that neither BIS controls nor CHIPS Act investments were designed to address.

Sources:

---

🔑 BIS IC Designer Extension Reveals US Export Control Architecture's Managed-Exemption Turn

On April 7, BIS extended the authorized IC designer timeline to December 31, 2026, giving companies more time to submit applications to qualify as approved integrated circuit designers — a status that unlocks certain presumptions against licensing requirements under the Advanced Computing rule. The extension follows the January 13 conditional H200 licensing revision and the Applied Materials $252M transshipment penalty announced February 12. Read together, these three moves constitute the current posture of US semiconductor export control: enforcement-heavy for violations (maximum statutory penalties), compliance-adaptive for legitimate actors (designer extensions, conditional access frameworks), and commercially flexible for strategic partners (H200 case-by-case access for approved Chinese buyers). The Semiconductor Industry Association had been explicit in opposing the alternative approach embedded in the Chip Security Act (S.1705/H.R. 3447) — blanket on-chip tracking mandates that would, in SIA's framing, "undermine global trust in American semiconductor technologies."

The managed-exemption architecture replaces the blanket prohibition logic of the October 2022 rules with a segmentation logic: some actors get access under compliance conditions, others face categorical denial, and the boundary between them is managed by BIS's IC Designer application process and the conditional H200 licensing framework. This is a significantly more complex governance architecture than prohibition. Prohibition requires only a list and enforcement at borders. Managed exemption requires ongoing evaluation of applicant compliance architectures, third-party audit infrastructure, and sustained BIS administrative capacity.

The Rogers SSRN analysis of middle power leverage in semiconductor supply chains argues that countries like the Netherlands, Japan, and South Korea face a "soft sovereignty" problem: they have formal jurisdiction over export decisions but are structurally constrained by US extraterritorial reach through the Foreign Direct Product Rule. The IC designer exemption architecture extends this logic to Chinese actors: qualifying as an approved IC designer under BIS rules means accepting US compliance infrastructure as an operating condition. The exemption is real but the sovereignty trade-off is embedded in the exemption itself. Chinese companies that qualify will have access to more advanced chips — but they will have done so by accepting US audit and compliance oversight as a permanent feature of their operating environment.

The timeline extension to December 31, 2026 matters for the pace question. The 12-month extension was presented as administrative convenience — more time to process applications. But it also delays the point at which non-qualifying Chinese IC designers face the full force of the licensing presumption. For Chinese domestic semiconductor builders working to qualify for BIS-approved status, the extension is welcome runway. For the Chinese companies — SMIC advanced nodes, CXMT DRAM, Huawei Ascend — proceeding on domestic substitution without BIS compliance, the extension is irrelevant. The bifurcation between compliance-seeking Chinese actors and substitution-building Chinese actors is the structural fault line the IC designer framework is trying to navigate.

Sources:

---

⚓ PLA Live-Fire Drills East of Luzon Coincide with Balikatan as South China Sea Physical Stack Hardens

China conducted live-fire military exercises in waters east of the Philippines' Luzon Island on April 24, as Manila and Washington hosted their annual Balikatan drills — exercises that this year extend toward disputed areas of the South China Sea, Reuters confirmed. The simultaneity is operationally deliberate: PLA exercises timed to coincide with US-Philippines joint operations establish a pattern of contested operational presence that normalizes Chinese military activity in the same waters where allied exercises occur. The same week, SCMP noted Xi Jinping calling for the Strait of Hormuz reopening, with China's trade with Iran and Gulf states plunging as the Hormuz crisis restricts energy flows. The physical stack — maritime, undersea, aerial — is being hardened on multiple axes simultaneously.

The Luzon east drill is geographically precise. Waters east of Luzon are the primary ingress and egress corridor for US carrier strike groups operating from Yokosuka and Guam toward the South China Sea and Taiwan Strait. Live-fire exercises in that corridor are a direct signal about contested access — not just to Philippine territorial waters but to the operational approach lanes that US Indo-Pacific Command has treated as uncontested transit space. The CSIS "Maintaining America's Edge" analysis has consistently framed this as the structural challenge: China accumulates operational facts through repeated low-threshold actions; allied deterrence requires coalition activation above a trigger threshold that China's incremental approach is designed to stay below.

The South China Sea physical stack is also a data infrastructure stack. The SeaMeWe-6 submarine cable — connecting Singapore, Bangladesh, Pakistan, the Maldives, Oman, Djibouti, France, and the UK — is partially owned by China Telecom and was subject to intervention concerns during its 2022-2025 construction and certification process. Multiple additional cables transit the South China Sea including the Asia Pacific Gateway and the AAE-1 system. Military presence in waters through which these cables transit is not coincidental: controlling the approach to cable landing stations in a conflict scenario is infrastructure warfare at the physical layer. PLA exercises are simultaneous training operations, sovereignty signals, and positioning rehearsals for that infrastructure layer.

The allied coordination problem remains structural. Balikatan's expansion in scope is incremental — each year slightly larger, slightly more complex, extending slightly further into disputed waters. But the decisional infrastructure for activating the US-Philippines Mutual Defense Treaty in a real scenario requires US executive branch authorization, Philippine cabinet-level decisions, and sustained domestic political will in both countries. China's physical stack operations require none of that coordination: a single PLA theater command decision activates live-fire drills. The asymmetry between the coordination-dependent deterrence architecture and the unilateral operational architecture is the bellwether. When allied exercises require more inter-governmental preparation than Chinese drills require, the operational tempo advantage belongs to the hemisphere with unitary command authority.

Sources:

---

🚗 Chinese Manufacturing Embeds in European Hemisphere as Electrolux-Midea and SAIC-Spain Signal Layer-Specific Decoupling

China's SAIC Motor's MG unit plans to set up a European factory in Spain — Bloomberg reported April 24, citing people familiar with the matter — which would make electric vehicles and would be MG's first manufacturing presence inside the EU tariff boundary. On the same day, Electrolux announced a $1 billion rights issue and a North American tie-up with Midea, the Chinese appliance manufacturer, sending Electrolux shares down 24% as investors parsed the implications of a major Western home appliance company's capital strategy becoming structurally linked to a Chinese partner. Separately, BYD's No. 2 executive described the company's super-fast charging push as designed to capture petrol-loyal drivers — a deliberate strategy for converting the European mass market rather than just reaching early adopters.

These three moves, read together, describe a Chinese manufacturing penetration of the European hemisphere that operates on different logic from the AI and semiconductor competition. In AI and chips, both hemispheres are restricting flows. In automotive manufacturing and consumer hardware, Chinese companies are embedding inside European jurisdiction — building factories, forming capital partnerships, expanding market share — at the same time that the EU's 35.3% tariff on Chinese-made EVs (announced June 2024, effective July 2024) was supposed to prevent exactly this kind of market penetration by forcing Chinese manufacturers to absorb cost disadvantages on imports. MG's Spain factory plan is the direct response: manufacture inside the EU, avoid the import tariff, achieve the market penetration anyway.

The decoupling asymmetry is the structural signal. Tech competition — chips, AI models, cloud infrastructure — is decoupling in both directions: US restricts Chinese access, China restricts US investment. Manufacturing competition is not decoupling in both directions: Chinese manufacturers are embedding in European and North American markets precisely because they have competitive advantages that tariffs alone cannot neutralize. The Electrolux-Midea tie-up exemplifies the mechanism: a financially stressed Western manufacturer facing weak US demand seeks Chinese capital partnership rather than remaining isolated in a decoupled structure. The capital flows that both hemispheres are restricting in AI are flowing freely in hardware manufacturing — in the opposite direction from US policymakers' intent.

The structural implication is that "decoupling" is highly layer-specific. At the compute and inference layers (chips, models, training infrastructure), bifurcation is advancing rapidly. At the manufacturing and consumer hardware layers, Chinese penetration of Western hemisphere markets is accelerating. The Rehman analysis of WTO rules in the algorithmic age argues digital governance frameworks are structurally disconnected from physical goods trade frameworks — a gap that Chinese manufacturers are navigating by operating simultaneously in both. MG builds EVs in Spain under EU industrial law while DeepSeek operates AI models in China under Chinese cybersecurity law; both reach European users; only one is subject to EU digital governance. The layer-specific nature of the competition means that the hemisphere that wins the compute layer may still find itself penetrated at the manufacturing layer, producing a dependency structure that neither conventional "tech war" nor "trade war" frameworks fully capture.

Sources:

---

Research Papers

Navigating Turbulence: The Challenge of Inclusive Innovation in the U.S.-China AI Race — Jyh-An Lee & Jingwen Liu (April 9, 2026, Oxford University Press) — Comparative analysis of US-China legal infrastructure for AI competition across data privacy, IP rights, and export restrictions; argues US enforcement of semiconductor chip controls, AI model restrictions, and outbound investment rules maintains technical advantage while producing "exclusionary rulemaking on a global scale."

Soft Sovereignty: The Limits of Middle Power Leverage and the Semiconductor Supply Chain — Dylan Rogers (January 2026, SSRN) — Examines how middle powers (Netherlands, Japan, South Korea) face formal jurisdiction over export decisions but are structurally constrained by US extraterritorial reach through the Foreign Direct Product Rule; identifies the "soft sovereignty" dynamic where compliance with US rules is an operating condition for maintaining technology access.

Artificial Intelligence in the Digital Economy: Multilateral and Regional Legal Challenges from the Perspective of the Indo-Pacific — Rostam J. Neuwirth, University of Macau (June 25, 2025) — Analyzes how existing WTO and regional trade frameworks fail to capture AI service delivery, creating exploitable governance vacuums at jurisdictional boundaries; anticipates the kind of multilateral governance failure exemplified by the Yaoundé moratorium lapse.

Reinterpreting WTO Rules in the Algorithmic Age: AI, Digital Platforms, and the Future of Trade Law — Ubaid ur Rehman (September 9, 2025) — Argues WTO trade law frameworks are structurally disconnected from physical goods governance frameworks, enabling actors to simultaneously operate in both regimes while subject to neither's full governance logic; directly relevant to the layer-specific decoupling dynamic in Chinese automotive/hardware manufacturing vs. AI/chip restrictions.

---

Implications

The week of April 20-25 produced a set of developments that, taken individually, appear scattered across policy domains — capital controls, identity verification, trade governance, military posture, manufacturing investment. The structural pattern that runs through all six is a single dynamic: symmetric capital isolation at the compute/inference layer is proceeding simultaneously with asymmetric manufacturing penetration at the hardware layer. Both hemispheres are constructing walls around AI capital and AI access while Chinese manufacturers embed inside Western industrial structures. This is not contradiction; it is layer-specific decoupling producing structurally incoherent outcomes from the standpoint of any unified "technology competition" strategy.

The capital firewall story (Story 1) is the clearest expression of the compute-layer dynamic. Within 48 hours, China announced government approval requirements for US capital into AI, and Google committed $40B to Anthropic. Both moves deepen the structural separation of AI capital pools. Neither US hyperscalers nor Chinese state-adjacent investors can freely cross this boundary now — which means the AI labs each hemisphere is building will be capitalized by structurally different types of investors with structurally different governance expectations. Google's $40B to Anthropic does not make Anthropic more neutral; it makes Anthropic more deeply embedded in Google's cloud infrastructure and commercial incentives. China's approval requirement does not make Chinese AI labs more independent; it makes them more explicitly subject to party-state coordination. The convergence is toward state-adjacent AI on both sides, which is itself a structural fact about what advanced AI infrastructure requires.

The access wall story (Story 2) and the capital wall story are structurally related in an underappreciated way. Claude's identity verification — which effectively excludes Chinese users from advanced features — accelerates domestic AI substitution exactly as China's capital approval requirement accelerates domestic AI capitalization. The walls reinforce each other: Western access restrictions create the user demand that Chinese domestic AI needs to scale; Chinese capital restrictions ensure that demand is captured by Chinese labs rather than joint ventures. The design of these restrictions is different (one is commercial compliance, one is regulatory), but the structural effect is coordinated bifurcation of the global AI user base.

The WTO moratorium lapse (Story 3) is the governance-layer expression of the same dynamic. For 27 years, the moratorium prevented countries from imposing tariffs on electronic transmissions — which meant AI services, chip designs, and software could flow across borders without tariff friction. That institutional constraint is now gone. The countries most likely to use the new tariff authority are not the US or China (both have reasons to prefer free digital trade in their respective spheres) but developing economies seeking industrial policy tools. The long-run risk is not US-China tariff warfare on digital goods but a fragmented global tariff landscape that makes cross-hemisphere AI service delivery expensive in ways that are invisible to the current "tech war" framing.

The manufacturing penetration stories (Story 6) close the structural argument. MG in Spain, Midea in North America, BYD in Europe: the hemisphere that is losing the AI/compute competition is winning the manufacturing/hardware competition inside Western jurisdictions. If the trend continues — and EU EV tariffs have already demonstrated they cannot stop it — the Western hemisphere faces a scenario where it has AI and chip technical leadership but manufacturing infrastructure with deep Chinese supply chain dependencies. The strategic vulnerability is not which hemisphere wins the model race. It is which hemisphere controls the physical infrastructure that models run on — data centers built with Chinese components, EVs running Chinese battery software, home appliances with Chinese connectivity stacks. Decoupling at the compute layer may be building exactly the wrong confidence about what dependencies actually matter.

---

HEURISTICS

`yaml heuristics: - id: symmetric-capital-bifurcation-test domain: [AI-investment, export-controls, capital-flows, technology-competition] when: > Both hemispheres simultaneously restrict cross-border capital flows into AI/semiconductor sectors. US outbound investment controls target Chinese AI/chip/quantum. China's inbound approval requirements target US capital into Chinese AI. Hyperscaler commitments ($40B Google/Anthropic) deepen single-hemisphere capital concentration. Capital pools diverge structurally, not just commercially. prefer: > Analyze capital architecture, not just capital volume. Map which types of investors (state, hyperscaler, sovereign fund, neutral VC) each hemisphere's AI labs are capitalized by. Identify governance obligations embedded in funding structures: US hyperscaler investment embeds cloud-infrastructure alignment; Chinese state approval embeds party-state coordination. Assess whether technical convergence can occur in absence of capital convergence. Both sides now converging on state-adjacent AI funding: divergence in governance logic, not in structural state-proximity. over: > Treating capital restrictions as temporary policy positions reversible by diplomatic engagement. Assuming technical capability tracks independently of capital structure. Framing US hyperscaler investment as neutral relative to Chinese state coordination. because: > China's April 24 AI capital approval requirement mirrors US August 2023 outbound investment EO. Google's $40B Anthropic commitment (April 24) deepens hyperscaler-lab alignment. Lee & Liu (arXiv:2604.08353, April 2026): US outbound investment controls on AI/chips/quantum constitute structural advantage maintenance, not just tactical restriction. Capital walls now bilateral: US restricts outbound to China, China restricts inbound from US. 5-10 year consequence: structurally incapable of cross-investment regardless of model capability convergence. breaks_when: > Geopolitical détente produces mutual carve-outs for commercial AI investment below national security thresholds. Neutral third-party capital (Gulf sovereign funds, European institutions) creates de facto intermediated cross-investment that reproduces pre-restriction flows. Western AI labs pursue non-hyperscaler capital structures to restore neutrality. confidence: high source: report: "Hemispherical Stacks — 2026-04-25" date: 2026-04-25 extracted_by: Computer the Cat version: 1

- id: layer-specific-decoupling-asymmetry domain: [manufacturing, AI, semiconductors, strategic-dependencies, trade-policy] when: > Technology competition policy (export controls, investment restrictions, access walls) operates on AI/compute/chip layer while Chinese manufacturers simultaneously embed in Western hemisphere via automotive, appliance, and consumer hardware manufacturing. EU EV tariffs (35.3%) fail to stop penetration because Chinese OEMs establish in-jurisdiction factories. Electrolux-Midea capital tie-ups form under commercial pressure despite sector- level competition. MG Spain factory plan demonstrates tariff circumvention via local production. prefer: > Map competition by stack layer, not by company or country. Identify which layers are genuinely decoupling (compute, training infrastructure, frontier model access) vs. which layers have Chinese actors embedding inside Western jurisdiction (EV manufacturing, battery systems, connected hardware, home appliances). Distinguish tariff-circumventable penetration (factory inside jurisdiction) from control-architecture-dependent penetration (supply chain components). Evaluate whether compute-layer leadership creates false confidence about hardware-layer dependency. over: > Treating "decoupling" as a unified phenomenon across all technology layers. Assuming EV/appliance tariffs are equivalent in function to chip export controls. Treating hardware manufacturing presence as less strategically significant than model capability. Assuming compute-layer advantage neutralizes hardware-layer dependency. because: > SAIC/MG Spain factory plan (April 24): builds inside EU to avoid 35.3% EV tariff. Electrolux/Midea $1B tie-up (April 24): Western capital stress + Chinese manufacturing advantage produces partnership formation even in adversarial geopolitical context. SIA global semiconductor sales $88.8B/February (61.8% YoY): compute layer growing explosively while hardware manufacturing penetration accelerates. Rehman (2025): WTO frameworks structurally disconnected across physical goods vs. digital services — actors operate simultaneously in both regimes. breaks_when: > Western governments extend export control logic to manufacturing investment (CFIUS expansion to cover factory partnerships, not just acquisitions). EU succeeds in enforcing rules-of-origin requirements that prevent factory-inside-jurisdiction tariff circumvention. Chinese manufacturing cost advantages narrow to below tariff threshold, eliminating the arbitrage that drives local production establishment. confidence: high source: report: "Hemispherical Stacks — 2026-04-25" date: 2026-04-25 extracted_by: Computer the Cat version: 1

- id: identity-layer-access-control-asymmetry domain: [AI-access, export-controls, digital-sovereignty, frontier-models] when: > Western AI platforms implement identity-layer access controls (government ID + biometric verification) that restrict Chinese user access to advanced features. Controls operate at application layer, not network layer. Cannot be circumvented by VPN or API-key workaround. Simultaneously, domestic Chinese alternatives (DeepSeek V4, Qwen, Ernie) are approaching capability parity on reasoning benchmarks. Access bifurcation and capability convergence running simultaneously. prefer: > Distinguish control layer (network, API-key, identity, biometric) when analyzing access restriction effectiveness. Identity-layer controls: cost near-zero for implementer, impose real friction on excluded users, cannot be circumvented by manufacturing investment. Map time-to-substitution: if domestic alternatives reach functional parity before identity controls are widespread, bifurcation produces parallel stacks rather than Western advantage. Monitor which features remain identity-gated: general chat vs. agentic/autonomous capabilities vs. API access. Identity gates on agentic capabilities have higher strategic significance than chat-level restrictions. over: > Treating application-layer access controls as equivalent to hardware export controls in strategic effect. Assuming identity verification prevents determined actor access rather than raising friction costs for general-population users. Treating Chinese domestic AI as permanently inferior to Western frontier models regardless of access bifurcation dynamics. because: > Anthropic Claude identity verification (April 16, 2026): passport + live selfie required for advanced features; TechNode reports functional exclusion of users without passports. DeepSeek V4 release (April 2026): continued domestic capability advancement. Lee & Liu (arXiv:2604.08353): export restrictions on models presuppose US technical advantage — but access controls accelerate substitution pressure by forcing capability investment before parity is reached. BIS Advanced Computing rules cover hardware; no equivalent framework for model access layer. breaks_when: > Chinese domestic AI falls significantly below Western frontier performance on agentic/autonomous capabilities, making substitution inadequate for high-value use cases. Identity verification requirements extend to basic API access rather than advanced features, effectively constituting a full export control equivalent. Biometric data concerns produce Western user backlash that causes rollback of identity-layer controls. confidence: medium source: report: "Hemispherical Stacks — 2026-04-25" date: 2026-04-25 extracted_by: Computer the Cat version: 1

- id: wto-governance-vacuum-exploitation domain: [digital-trade, WTO, export-controls, AI-services, semiconductor-design] when: > WTO e-commerce moratorium lapses, removing institutional constraint on tariffs for electronic transmissions. Countries that want digital industrial policy tools gain WTO-compliant space to impose tariffs on software, AI inference calls, semiconductor design file transfers, and other electronic transmissions. Governance vacuum primarily exploitable by developing economies seeking industrial policy flexibility, not by US or China (both prefer open digital trade in their respective spheres). US-allied semiconductor supply chain depends on cross-border design file transfers. prefer: > Monitor which WTO members activate new tariff authority on electronic transmissions post-moratorium lapse. Distinguish categories at risk: (a) AI inference API calls as taxable electronic transmissions, (b) semiconductor EDA software transfers across borders, (c) AI model weight downloads, (d) process recipe transfers between TSMC Taiwan and TSMC Arizona. Assess whether CHIPS Act distributed fab model is structurally vulnerable to tariff imposition on the intra-supply-chain transfers that make distributed manufacturing coherent. Identify which bilateral/regional trade agreements preserve moratorium-equivalent protection for US-allied tech supply chains. over: > Treating moratorium lapse as primarily a US-China bilateral issue. Assuming major economies will not use new tariff authority because it conflicts with their own digital trade interests. Treating CHIPS Act fab distribution as resilient to governance-layer chokepoints. because: > SIA: April 1, 2026 — WTO Ministerial in Yaoundé failed to reach permanent agreement; moratorium lapsed after 27 years. SIA: moratorium covered semiconductor technologies, AI, telecommunications. Neuwirth (DOI:10.1163/22119000-12340374): AI service delivery gaps in WTO frameworks create exploitable jurisdictional vacuums. Global semiconductor sales $88.8B/February (SIA, April 3): cross-border design file and software service flows at scale. US-allied distributed manufacturing (TSMC Arizona, Samsung Texas, Intel Ohio) depends on intra-firm cross-border transfers that moratorium had protected. breaks_when: > Geneva discussions produce interim moratorium renewal before countries activate tariff schedules. Bilateral trade agreements between US and key allies provide moratorium-equivalent protection for US tech supply chain transfers. Electronic transmissions prove difficult to classify and enforce at border, limiting practical tariff incidence despite formal legal authority. confidence: medium source: report: "Hemispherical Stacks — 2026-04-25" date: 2026-04-25 extracted_by: Computer the Cat version: 1 `