Hemispherical Stacks · 2026-04-20

🌐 Hemispherical Stacks — 2026-04-20

🔐 Anthropic's Mythos Forces Asian Regulators into Emergency AI Cybersecurity Architecture
⚖️ Applied Materials' $252M Korea Transshipment Case Exposes Allied-Nation Compliance Gap
🔧 BIS Pivots from Blanket Prohibition to Conditional Engagement on H200 Exports to China
🛳️ Liaoning Taiwan Strait Transit Signals Carrier Diplomacy as Balikatan Expands to Japan
🏗️ Beijing Satellite Town vs CHIPS Act: Two Infrastructure Races, Two Geometric Logics
🤖 Chinese Humanoid Robots Beat Human Runners as CSIS Frames US Competitive Decline

---

🔐 Anthropic's Mythos Forces Asian Regulators into Emergency AI Cybersecurity Architecture

The Hong Kong Monetary Authority announced Monday it will introduce a Cyber Resilience Testing Framework and form a dedicated public-private taskforce to address AI-driven cyber risks — a direct response to Anthropic's Mythos model, which regulators across Asia are now treating as an unprecedented infrastructure threat. Singapore's Monetary Authority warned financial institutions that AI advances "could speed up the discovery and exploitation of software vulnerabilities." South Korea's Financial Supervisory Service held emergency meetings with information security officers from banks and insurers last week. Australia's ASIC and APRA are both monitoring the model for systemic market risk.

The coordinated reaction follows Barclays CEO C.S. Venkatakrishnan's statement at the G30 consultancy on the sidelines of the IMF spring meeting on April 17: "there will be a Mythos 2 and a Mythos 3, and they'll come up with probably distressing frequency." His framing pointed to the structural problem — legacy banking infrastructure, not just Mythos specifically, is the vulnerability. Older and larger institutions running legacy systems face particular exposure because patching cycles are measured in months, while a new capable model can identify exploitable gaps in days.

The cross-hemisphere structure here is not symmetric. Anthropic is a US company subject to US export control and regulatory authority. Asian financial regulators can monitor, warn, and test — but they cannot compel US model developers to restrict access or modify model capabilities. The HKMA taskforce, however sophisticated, is a reactive governance architecture responding to a capability produced outside its jurisdiction. The BIS 2026 H200 licensing revision framework, which conditions semiconductor exports on verified Chinese buyer compliance, has no equivalent for AI model access: any organization worldwide with an Anthropic API key can query Mythos today.

What the Asian regulatory response reveals is a new class of strategic dependency — not chips, not rare earths, but inference capability. Banking cybersecurity infrastructure globally now runs against US-origin frontier AI offense. That dependency is invisible in most trade and technology competition frameworks because it doesn't cross a customs border. The Lee and Liu analysis of how US-China AI competition shapes legal infrastructure (Oxford University Press, 2026) argues US export restrictions on chips and models preserve technical advantage — but Mythos-class capability diffuses through API access, not physical shipment. The asymmetry between chip controls and model access controls is a structural blind spot in the current US export control architecture.

HKMA's taskforce is the first formal institutional response in Asia that treats AI capability offense as a financial stability risk rather than a generic cyber threat. If that framing spreads, the next policy cycle will force a reckoning with whether frontier model access should be classified alongside semiconductor exports under national security frameworks — a question neither BIS nor Anthropic has publicly addressed.

Sources:

---

⚖️ Applied Materials' $252M Korea Transshipment Case Exposes Allied-Nation Compliance Gap

BIS announced February 12 that Applied Materials and its Korean affiliate will pay approximately $252 million — the second-highest penalty in BIS history and twice the $126 million merchandise value, the statutory maximum — for illegally exporting ion implanters to a Chinese company on the Entity List. The route: AMAT shipped equipment to Applied Materials Korea (AMK) for assembly, then onward to China, without the required export license. The Entity Listed Chinese customer had been placed on the list in 2020; the violations occurred in 2021–2022. Compliance personnel responsible for the shipments are no longer employed by either AMAT or AMK.

The case dissects the operational logic of transshipment: the Korea routing was not incidental but structural. Ion implanter components are classified under Export Administration Regulations as requiring licenses for Entity Listed recipients. By routing through the Korean affiliate, AMAT separated the contractual and physical chain of custody in a way that obscured the ultimate destination from standard compliance checks. The BIS investigation, conducted by the Office of Export Enforcement's Boston Field Office and Homeland Security Investigations, reconstructed the routing from financial and shipping records.

The structural implication is harder to address than the penalty. Korea is a US treaty ally, a Five Eyes-adjacent intelligence partner, and a critical node in global semiconductor supply chains. SK Hynix and Samsung supply HBM memory to NVIDIA; AMAT's Korean affiliate assembles US capital equipment. The SIA/BCG projection that the US will capture 28% of global semiconductor capex from 2024–2032 assumes allied nation cooperation is reliable. The AMAT case documents the failure mode: under commercial pressure, allied-nation affiliates become transshipment vehicles rather than compliance backstops.

BIS enforcement architecture for this problem is asymmetric with China's substitution approach. BIS has 40 export enforcement officers per New Zealand-equivalent jurisdiction — a reference figure from prior enforcement reviews — against hundreds of Chinese procurement networks actively seeking equipment. The $252M penalty is calibrated to deter AMAT specifically and signal to industry generally, but the Korean transshipment route required a years-long investigation to reconstruct. China's domestic semiconductor buildup, by contrast, requires only capital allocation decisions made inside a single country's governance structure. For every AMAT-scale enforcement action, there are substitution investments in CXMT, YMTC, and Huawei semiconductor that proceed without any compliance friction at all. The case reveals that allied-nation compliance is not a structural guarantee but a coordination problem requiring ongoing enforcement investment that the current BIS staffing level cannot sustain at scale.

Sources:

---

🔧 BIS Pivots from Blanket Prohibition to Conditional Engagement on H200 Exports to China

The January 13 BIS rule establishing case-by-case licensing for the Nvidia H200, AMD MI325X, and similar chips to China — following President Trump's December 8, 2025 announcement — marks the first structural reversal in US advanced computing export controls since the October 2022 comprehensive restrictions. Where Biden-era policy imposed a presumption of denial for advanced AI chips to Chinese buyers, the new framework substitutes a compliance architecture: applicants must demonstrate the export won't reduce US customer capacity, Chinese buyers must adopt screening procedures, and products must undergo independent US third-party security testing before approval.

The April 7 extension of the authorized IC designer deadline to December 31, 2026 adds a second layer of liberalization: companies seeking to qualify as approved integrated circuit designers — a status that unlocks certain presumptions against licensing requirements — now have additional time to file applications, and BIS has additional time to process them. Together, the H200 licensing revision and the IC designer extension constitute a policy posture that is simultaneously enforcement-heavy (AMAT $252M) and commercially adaptive. The apparent contradiction resolves at the structural level: BIS is attempting to build a two-tier system where approved, compliant Chinese buyers get conditional access, while non-compliant routes get maximum statutory penalties.

Whether this architecture is coherent depends on the verification problem. The H200 licensing conditions require that Chinese purchasers adopt export compliance procedures including customer screening. But the entity conducting that screening is the Chinese buyer itself, with periodic third-party audits. The Federal Register text specifies independent third-party testing in the United States as a requirement — a significant compliance friction that could limit which Chinese buyers qualify, but does not address re-export risk after initial sale.

From the Chinese side, the conditional access offer intersects uncomfortably with domestic substitution strategy. Huawei's Ascend 910C and 920 chips are advancing toward H100-class performance; CXMT is scaling advanced DRAM; the US CHIPS Act tripling of US fab capacity by 2032 is a response to, and a continuation of, a race that China has been running since 2022. For large Chinese cloud providers like Alibaba, Baidu, and ByteDance, conditional H200 access is commercially valuable — but it comes with US oversight infrastructure embedded in the transaction. That oversight infrastructure is itself a form of influence that domestic Chinese chips, however competitive, would not carry. The structural question the new policy raises but does not answer: is conditional US chip access a tool of commercial engagement, or a mechanism for embedded surveillance of Chinese AI infrastructure?

Sources:

---

🛳️ Liaoning Taiwan Strait Transit Signals Carrier Diplomacy as Balikatan Expands to Japan

The Chinese aircraft carrier Liaoning sailed through the Taiwan Strait on Monday, Taiwan's defense ministry confirmed, marking the first transit of the sensitive waterway by a Chinese carrier since late 2025. The timing is not incidental: the US and Philippines began their annual Balikatan military exercises the same day, now with expanded participation from Japan — a structural shift from bilateral to trilateral exercise architecture that Beijing has explicitly warned against as undermining regional trust and deepening division. North Korea conducted ballistic missile tests over the weekend, testing warhead performance with cluster bomb and fragmentation mine payloads, adding a third military signal to a week of layered strategic messaging.

The Liaoning transit is a carrier diplomacy exercise in the double sense: it demonstrates access while monitoring the diplomatic response. Taiwan's defense ministry tracking and public announcement is the response Beijing is reading. The transit establishes operational precedent — each uncontested passage is an incremental normalization of Chinese carrier access to the Strait. The relevant comparison is not with past US carrier transits, which established freedom of navigation rights in the 1990s, but with Chinese maritime militia operations in the South China Sea, which used attrition of response to gradually redefine operational norms.

Japan's formal inclusion in Balikatan is the structural bellwether. The exercises had previously been bilateral under the 1951 US-Philippines Mutual Defense Treaty framework. Japan's participation imports the 2015 US-Japan Defense Guidelines' collective self-defense reinterpretation into a South China Sea exercise context — creating a trilateral architecture that could, under the right trigger conditions, activate Japanese defense obligations in a Taiwan scenario. China's immediate warning against military cooperation is calibrated at this expansion, not at Balikatan itself (which is annual and expected).

The asymmetry in how each hemisphere is building its military-technological architecture shows in the infrastructure logic. China is accumulating unilateral operational facts — carrier transits, island installations, regular drills — that change what is normal without triggering a formal response threshold. The US and allied framework is building multilateral deterrence architecture — expanding Balikatan, the Quad, AUKUS — that requires consensus and partner-state political will to activate. CSIS analysts at the April 20 "Maintaining America's Edge" event framed this as the core competitive challenge: US strategy requires coalition, Chinese strategy requires only capital and political will. The Liaoning transit and Balikatan expansion on the same day are a single data point in a structural argument about which architecture is faster to execute.

Sources:

---

🏗️ Beijing Satellite Town vs CHIPS Act: Two Infrastructure Races, Two Geometric Logics

Beijing's Satellite Town core area will be completed in the second half of 2026, state-owned media Beijing Daily reported April 18. The district is designed as a concentrated hub for satellite manufacturers, operators, and supporting aerospace enterprises — a single geographic cluster in the capital region intended to function as the integrated supply chain for China's space-based computing and communications infrastructure. Colocation of manufacturers and operators in a dedicated zone reduces logistics friction, accelerates certification cycles, and allows state coordination across the supply chain without market-mechanism overhead. The completion timeline of H2 2026 means core infrastructure will be operational as Beijing's broader satellite manufacturing ambitions come online.

The US strategic infrastructure analogy is the CHIPS and Science Act manufacturing buildout. SIA and BCG project US fab capacity will increase 203% by 2032 — a tripling from 2022 baseline — with the US growing from 10% to 14% of global fab capacity and from 0% to 28% of global advanced logic (below 10nm) capacity. The US is projected to capture 28% of global semiconductor capex from 2024–2032, second only to Taiwan at 31%. These numbers represent a genuine structural shift; in the absence of the CHIPS Act, BCG estimated the US would have captured only 9% of global capex by 2032.

The geometric difference between the two approaches is meaningful. Beijing Satellite Town is a concentration strategy: compress supply chain nodes geographically to achieve coordination efficiency. CHIPS Act investments are geographically distributed — TSMC in Arizona, Samsung in Texas, Intel in Ohio — optimized for resilience against single-point disruption but at the cost of supply chain integration. Concentrated zones produce faster iteration cycles when components travel meters rather than kilometers; distributed fabs produce survivability under adversarial disruption. Neither geometry is wrong; they are calibrated for different threat models.

The satellite manufacturing dimension connects to the export control architecture: satellite components, like advanced chips, are dual-use under the Bureau of Industry and Security's Export Administration Regulations. Beijing Satellite Town, by concentrating satellite-related manufacturing under state coordination, reduces the information available to US export control authorities about where specific components end up. A distributed, market-structured supply chain produces transaction records that BIS can audit; a state-coordinated cluster does not. The same logic that makes CHIPS Act distribution resilient to adversarial disruption makes Beijing's concentration strategy resilient to US export control visibility. Both sides are optimizing their infrastructure geometry against the other's instruments of pressure.

Sources:

---

🤖 Chinese Humanoid Robots Beat Human Runners as CSIS Frames US Competitive Decline

Chinese-made humanoid robots beat human runners at a race event Monday, Reuters reported — the second year of such events, after the inaugural edition in 2025 was "riddled with mishaps and robots that were unable to finish." The year-over-year improvement captures the operational difference between Chinese and US approaches to embodied AI: China is running mass public demonstrations that produce failure data, iteration, and then working systems deployed at visible scale. The US robotics sector, represented by Boston Dynamics, Figure, and Agility Robotics, is building capability in controlled commercial environments. Both approaches produce capable systems; they produce them on different timescales and with different data feedback architectures.

The race event is not a benchmark — bipedal locomotion at marathon-equivalent sustained pace on a flat course is not the relevant performance metric for military or industrial applications. But it is a visible commitment to scale and deployment that the US commercial robotics sector, with its enterprise sales model, does not replicate. Chinese humanoid robotics development is proceeding under a state industrial policy framework that tolerates public failure in exchange for rapid iteration. The 2025 event where robots "couldn't finish" was not a failure; it was a data collection exercise at the infrastructure level that produced the 2026 performance.

On the same day, CSIS hosted Representative Darrell Issa and CSIS Defense President Seth Jones for a virtual discussion on "Maintaining America's Edge: Is the United States Losing the Lead in an Increasingly Competitive World?" — a question that, set against the robot race footage, answers itself differently than the CSIS framing intends. Jones and Issa focused on defense industrial base revitalization and technological competitiveness strategy. Both are real problems. But framing the response as an industrial base problem — a manufacturing capacity and procurement reform challenge — mischaracterizes what the robot race represents. China is not ahead on manufacturing capacity alone; it is ahead on the willingness to deploy at scale before perfection.

The SIA projection that the US CHIPS Act will restore US fab capacity to 14% of global share by 2032 is an industrial base answer to a deployment tempo question. The relevant gap is not between what the US can manufacture and what China can manufacture — it is between what the US is willing to deploy publicly and at scale, and what China is already demonstrating. Comparative analysis of US and Chinese AI legal infrastructure finds China's data access environment and IP framework favor rapid model training at scale; the robot race suggests the same dynamic applies to physical system deployment. CSIS framing this as a defense strategy problem is accurate but insufficient — the competitive deficit is being built in public, one robot race at a time.

Sources:

---

Research Papers

Navigating Turbulence: The Challenge of Inclusive Innovation in the U.S.-China AI Race — Jyh-An Lee and Jingwen Liu (April 9, 2026, Oxford University Press 2026) — Comparative analysis of how US and Chinese legal infrastructure — data privacy, IP rights, and export restrictions — shapes AI development trajectories. Argues US enforces strict export controls on chips, models, and outbound investment while China's environment offers data access advantages; together they are producing exclusionary rulemaking at global scale. Directly relevant to the H200 policy pivot and the Mythos model access asymmetry described in Stories 1 and 3 above.

Whack-a-Chip: The Futility of Hardware-Centric Export Controls — Ritwik Gupta, Leah Walker, Andrew W. Reddie (November 21, 2024) — Argues that US semiconductor export controls are "widely known to be permeable," with China steadily building state-of-the-art AI models with exfiltrated hardware. Provides the analytical framing for why the AMAT Korea transshipment case and the H200 conditional licensing shift are structural rather than exceptional responses to an inherently leaky control architecture. Predates the BIS policy pivot but predicts its logic.

From Topology to Trajectory: LLM-Driven World Models For Supply Chain Resilience — Jia Luo (April 13, 2026) — Proposes using LLM-driven world models to map semiconductor supply chain topology and simulate disruption trajectories. Relevant to the geometric logic comparison in Story 5: the paper's central insight that supply chain topology determines resilience trajectories applies directly to the Beijing Satellite Town concentration vs CHIPS Act distribution geometry. Models that can simulate both architectures against adversarial disruption scenarios are the analytical tools the next cycle of US-China infrastructure competition will need.

---

Implications

The six stories from this week resolve into a single structural argument: the competition is no longer primarily about restricting Chinese access to US technology. It is about whether US governance and deployment architectures can operate at speeds and scales that match Chinese operational tempo.

On access restriction: the H200 licensing pivot is a concession to commercial reality — blanket prohibition was already being circumvented, as the AMAT Korea transshipment case and Gupta et al.'s analysis both document. Conditional engagement, with security conditions embedded in the transaction, is a more sophisticated architecture. But it presupposes verification infrastructure (BIS staffing, third-party testing capacity, Chinese buyer audit access) that does not currently exist at scale. The forty-officer enforcement corps trying to police global semiconductor flows is a genuine institutional gap that the AMAT $252M penalty, however well-calibrated as a deterrent signal, cannot structurally close.

On deployment tempo: the robot race and Mythos banking cybersecurity response reveal the same dynamic from opposite ends. China is deploying embodied AI in public events, accumulating failure data, and iterating. The US is deploying AI capability through commercial API access that flows to Asian financial infrastructure without governance architecture capable of managing the resulting dual-use risks. HKMA's emergency taskforce and Barclays CEO's warning are governance artifacts of a deployment reality that moved faster than oversight could follow. The 10-15 year CHIPS Act fab buildout timeline — valuable and necessary — is not calibrated to the 12-month iteration cycles the robot race and Mythos deployment demonstrate.

On alliance architecture: Japan's entry into Balikatan and the Liaoning Taiwan Strait transit on the same day illustrate the competing infrastructure logics. China accumulates unilateral access facts; the US builds multilateral deterrence consensus. Coalition architecture requires partner political will and operational coordination that unilateral capability demonstration does not. The structural advantage of China's approach is speed; the structural advantage of the US approach is legitimacy and combined capability. Which advantage matters more depends on whether the relevant scenario is a fast-moving military contingency or a slow-burn economic competition — and both are simultaneously active.

The week's most underappreciated asymmetry is the one Lee and Liu identify in their April 9 paper: US legal infrastructure — strict export controls on chips, models, and investment — preserves technical advantage by restricting access. But Mythos diffuses through API, not hardware. When the performance frontier is model capability rather than hardware access, the export control architecture is pointed at the wrong layer of the stack. The next policy cycle will need to address whether frontier model inference access — not just chips — belongs in the dual-use control framework that currently governs only physical exports.

---

HEURISTICS

`yaml heuristics: - id: conditional-access-embeds-oversight domain: [export-controls, semiconductor-policy, US-China-competition] when: > BIS shifts from blanket prohibition to conditional engagement on advanced chip exports. Applicants must demonstrate: no reduction in US capacity, Chinese buyer compliance screening, independent US third-party security testing. IC designer deadline extensions signal commercial accommodation pressure. Applied Materials Korea transshipment ($252M penalty, twice merchandise value) demonstrates enforcement escalation alongside liberalization. prefer: > Analyze the conditional access architecture as an oversight-embedding mechanism, not merely a commercial concession. Third-party US testing requirement inserts American verification infrastructure into Chinese buyer procurement chains. Compliance screening obligation creates audit rights. Map what oversight architecture the conditions actually install versus what they restrict. Track whether verification capacity (BIS staffing, accredited third-party testers) is scaling proportionally to license volume. Current: ~40 BIS enforcement officers vs global semiconductor procurement networks. over: > Treating H200 case-by-case policy as either pure liberalization or pure restriction. The architecture is neither; it is conditional dependency — Chinese buyers gain access by accepting embedded US compliance oversight. Framing as "US relaxes controls" misses the surveillance architecture in the conditions. Framing as "US maintains restrictions" misses the commercial engagement that makes Chinese buyers willing to accept those conditions. because: > BIS January 13 rule requires: (1) applicant demonstrates no reduction in US customer capacity; (2) Chinese purchaser adopts screening procedures; (3) independent US third-party testing in the United States verifies performance and security. Applied Materials $252M penalty (Feb 12) — twice the $126M merchandise value, statutory maximum — signals maximum enforcement on non-compliant routes. arXiv:2604.08353 (Lee & Liu, April 9) argues US legal architecture preserves technical advantage through export restrictions; conditional access adds oversight architecture rather than removing restriction. Gupta et al. (2024): hardware-centric controls are inherently permeable; conditions-based access creates compliance infrastructure inside the buyer relationship that hardware prohibition cannot produce. breaks_when: > Verification infrastructure does not scale. If BIS cannot actually audit Chinese buyer compliance screening or enforce third-party testing requirements, conditions become nominal. Korea transshipment case shows compliance fails under commercial pressure even in allied nations; same risk applies to Chinese buyer self-certification. Also breaks if China develops domestic chip alternatives (Huawei Ascend 920, CXMT DRAM) that match H200 performance — conditional access loses its leverage when the controlled product is no longer at the performance frontier. confidence: high source: report: "Hemispherical Stacks — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1

- id: model-access-as-missing-export-control-layer domain: [AI-governance, dual-use-AI, export-controls, financial-infrastructure] when: > Frontier AI model capability (Mythos-class coding and vulnerability identification) diffuses through API access, not hardware shipment. Asian financial regulators (HKMA, ASIC, MAS, South Korea FSS) respond with emergency governance architecture to a US-origin model they cannot compel to restrict access. Export control frameworks (EAR, ITAR) govern physical exports and specific technologies; no equivalent framework covers frontier model inference access as dual-use capability. prefer: > Analyze model access as a missing layer in the export control stack. Current architecture: chip controls (EAR, BIS) restrict hardware; investment controls (CFIUS) restrict ownership; outbound investment rules restrict capital flows. Model inference access — any global organization with an API key can query Mythos-class capabilities — exists outside all three frameworks. Map which regulatory body would have jurisdiction: BIS (Commerce), State Dept (munitions list), Treasury (OFAC), or a new framework. Track whether HKMA's public-private taskforce framing (AI as financial stability risk) produces any formal request to US authorities for access controls. over: > Treating Mythos financial risk as a generic cybersecurity problem addressable through bank-level hardening alone. HKMA and MAS can require banks to patch faster and test defenses more rigorously — but cannot restrict Mythos access for malicious actors who are not regulated financial institutions. The governance gap is jurisdictional, not technical. Comparing to prior software vulnerability cycles misses the magnitude: previous tools required operator skill; Mythos-class models democratize exploitation capacity. because: > Reuters April 20: HKMA announces Cyber Resilience Testing Framework and dedicated public-private taskforce — a reactive governance architecture built in response to a capability produced outside HKMA's jurisdiction. Singapore MAS: "financial institutions need to redouble efforts to strengthen their security defences." South Korea FSS held emergency meeting with bank information security officers. Barclays CEO (Reuters April 17, G30/IMF): "there will be a Mythos 2 and a Mythos 3, and they'll come up with probably distressing frequency." arXiv:2604.08353: US export restrictions on models preserve advantage but API access bypasses physical export framework entirely. BIS January 13 rule conditions H200 access on security requirements — no equivalent conditions exist for model API access. breaks_when: > US government classifies frontier model inference capabilities under existing dual-use export control frameworks (requiring licensing for non-US API access). Alternatively: Anthropic and other frontier labs voluntarily implement access controls for critical infrastructure sectors in foreign jurisdictions. Also breaks if HKMA-style regulatory responses generate sufficient diplomatic pressure to create a US-Asian financial regulator dialogue on model access governance — which would formalize the missing layer through bilateral agreement rather than unilateral US export control. confidence: medium source: report: "Hemispherical Stacks — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1

- id: deployment-tempo-advantage-vs-restriction-architecture domain: [US-China-competition, embodied-AI, military-posture, industrial-policy] when: > China demonstrates deployed operational capability at public scale (humanoid robots beating human runners, second iteration after 2025 failures; carrier transits establishing operational precedent; satellite manufacturing district completing core infrastructure). US responds with restriction architecture (export controls, alliance building, industrial base reform legislation) calibrated to 5-10 year timescales. CSIS April 20 event frames the challenge as US needing to revitalize industrial base and accelerate technological competitiveness. prefer: > Separate the restriction architecture question (what can US prevent China from accessing) from the deployment tempo question (what is China already operating). Both matter but on different timescales and with different leverage points. Restriction architecture (BIS, CHIPS Act) operates on 3-10 year horizons; deployment tempo operates on 12-24 month iteration cycles. Chinese humanoid robot iteration: 2025 failures → 2026 performance competitive with human runners = one cycle. Mythos-to-Mythos-2 interval Barclays CEO warned about: "distressing frequency." Map which competitive domain is governed by which timeline and calibrate policy response accordingly. over: > Treating the competition as primarily an industrial base or manufacturing capacity problem. SIA/BCG: US fab capacity up 203% by 2032 (10→14% global share, 0→28% advanced logic). These numbers are real gains. But the robot race and Mythos deployment show China operating at deployment tempo that does not require winning the manufacturing race first. 2032 CHIPS Act targets don't address 2026 operational deployments. CSIS "Maintaining America's Edge" framing as defense strategy problem is accurate but insufficient if the competitive deficit is being built through commercial and public deployment, not military acquisition. because: > Chinese humanoid robots Reuters April 20: "whizzed past human runners, highlighting the sector's rapid advances after last year's inaugural edition was riddled with mishaps." One-year iteration from failure to competitive performance is deployment tempo, not R&D pace. Liaoning Taiwan Strait transit establishes carrier access precedent through operational fact, not policy negotiation. Beijing Satellite Town (Reuters April 18): core area completing H2 2026 — concentrated manufacturing cluster operational within 6 months. US SIA/BCG 203% fab capacity increase runs from 2022 baseline to 2032 target — a decade to deliver manufacturing resilience against 12-month operational iteration cycles. arXiv:2604.08353: China's legal environment advantages (data access, IP framework) favor "rapid model training at scale" — same dynamic in physical systems. breaks_when: > US demonstrates equivalent deployment tempo through autonomous systems at scale (Replicator initiative achieving production quantities, Palantir/Shield AI operational deployment in allied exercises). Also breaks if Chinese iteration cycles slow due to chip access constraints — specifically, if H200-class compute restrictions genuinely impede embodied AI training rather than being circumvented through domestic alternatives. Gupta et al. (2024) argues hardware-centric controls are already permeable; the break condition requires the controls to actually bind on embodied AI training compute, not just inference. confidence: high source: report: "Hemispherical Stacks — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1 `