Hemispherical Stacks · 2026-04-27

🌐 Hemispherical Stacks — 2026-04-27

🤖 DeepSeek-V4-Pro Closes the Model Gap as Stanford Index Confirms China's AI Capability Convergence
⚙️ MATCH Act Demands Allied Chip Curb Parity in 150 Days, Compressing the Coercion Timeline
🏭 CFIUS Derails Sanan-Lumileds Twice: The Structural Ceiling on Chinese Tech Acquisition Strategy
🔬 Lightelligence's $400M Hong Kong IPO Signals China's Systematic EUV-Bypass via Photonic Computing
🕵️ Taiwan's Quiet Tech War: 100 Talent-Poaching Cases Expose China's Parallel Human Capital Track
🚗 Iran Oil Shock at $106/Barrel Accelerates Chinese EV Infrastructure Lock-In Across Southeast Asia

---

🤖 DeepSeek-V4-Pro Closes the Model Gap as Stanford Index Confirms China's AI Capability Convergence

DeepSeek launched preview versions of V4-Pro and V4-Flash on April 24, claiming top performance across all rival open models in mathematics and coding benchmarks, trailing only Google's closed Gemini 3.1-Pro in world knowledge. The performance gap to GPT-5.4 and Gemini 3.1-Pro is estimated by DeepSeek itself at "approximately 3 to 6 months" of developmental trajectory — and the trajectory is closing, not stable. The structural context arrived simultaneously: the Stanford AI Index 2026 concludes that China has "effectively closed" the AI performance gap, leading in publication volume, total patent output, and industrial robot installations even as the US retains a narrow edge in top-tier model quality and high-impact patents.

V4-Pro continues DeepSeek's signature chip-constrained optimization paradigm. Rather than compute scaling through H100 clusters, the architecture employs sparse mixture-of-experts methods achieving near-frontier performance on restricted hardware. V4-Flash's parity in reasoning with V4-Pro at significantly lower latency extends the principle: China's model stack is diverging into an efficiency-optimized design space, not approximating the US stack on worse hardware. This divergence matters because it decouples capability progression from hardware access — exactly the dependency that export controls were designed to enforce.

Deployment scale is the second dimension. Alibaba's Qwen model family has accumulated nearly one billion cumulative downloads globally — the most deployed open-source model family worldwide, well ahead of Meta's Llama. Zhipu AI simultaneously open-sourced GLM-5.1 while raising prices for the second time this year, signaling the strategic pivot from penetration-through-open-source to monetization-through-premium-inference. Every Qwen or DeepSeek deployment outside Western inference infrastructure generates integration dependencies that reinforce the ecosystem.

The hemisphere divergence is methodological. Western AI policy frames hardware access as the binding capability constraint. DeepSeek-V4 demonstrates that inference-viable frontier performance is reachable through architectural efficiency on restricted hardware. A 3-to-6-month model performance lag with a widening deployment scale advantage represents a strategic outcome at odds with what US export restrictions were designed to produce. Hardware controls assumed a compute-capability correlation that is not physics-constrained — DeepSeek's body of work is the empirical test that the assumption is wrong, and the Stanford Index is the population-level confirmation.

Sources:

---

⚙️ MATCH Act Demands Allied Chip Curb Parity in 150 Days, Compressing the Coercion Timeline

Republican Representative Michael Baumgartner's MATCH Act — Multilateral Alignment of Technology Controls in Hardware — arrived with bipartisan Senate backing demanding that allied countries match US semiconductor export restrictions on China within 150 days of enactment. The target is not just ASML's EUV systems, already denied since 2019. The bill covers deep ultraviolet immersion lithography and cryogenic etching tools — equipment that SMIC and YMTC still access legally through Dutch and Japanese channels and which underpins China's current leading-edge fab work. China Securities analysts assess the bill has "a relatively high chance of being passed at the congressional level," while Shanghai export control lawyer Mark Shi frames the legislation as a potential diplomatic bargaining chip in ongoing Sino-US trade negotiations.

SMIC's N+2 process — achieving 7nm-equivalent density without EUV — relies on repeated multi-patterning passes through DUV immersion tools. Closing this pathway requires ASML's compliance, operating under Dutch export licensing that has expanded restrictions incrementally but not matched US controls. Japanese firms Tokyo Electron and Shinko face analogous alignment pressure on cryogenic etching and specialty chemicals. Both governments manage domestic industries with significant China exposure: global semiconductor sales hit $88.8 billion in February 2026, up 61.8% year-over-year, with China recording 57.4% year-on-year growth — the market neither the Netherlands nor Japan wants to exit.

The 150-day timeline converts allies from partners managing competing interests into potential sanctions targets — coercive coordination rather than collaborative alignment. This is the structural asymmetry the MATCH Act reveals at its core: US export control architecture requires partner participation to function. China's substitution strategy does not. Dishan Technology's 2nm AI GPU prototype — using FinFET/GAA hybrid architecture and chiplet packaging with CUDA-compatible toolchains — advances through capital and engineering investment, not export permissions from sovereign governments. The FCC's concurrent move to bar Chinese telecom carriers from US data centers and interconnection protocols extends the same coercive logic to the network layer — a parallel alignment demand on allies operating joint infrastructure.

Every month that allied alignment negotiations extend is a month China's domestic substitution timeline shortens. The MATCH Act's 150-day ultimatum compresses that asymmetry into a legible policy window — but doesn't resolve the underlying structural dynamic: control architectures scale with partner alignment; substitution architectures scale with capital investment.

Sources:

---

🏭 CFIUS Derails Sanan-Lumileds Twice: The Structural Ceiling on Chinese Tech Acquisition Strategy

The Committee on Foreign Investment in the United States determined on April 17 that Sanan Optoelectronics' $239 million acquisition of Dutch LED manufacturer Lumileds Holding posed "irresolvable US national security risks," forcing the Shanghai-listed chipmaker and its Malaysian partner to withdraw their filing. This is the second failed Chinese acquisition of Lumileds — MLS Holdings was blocked by CFIUS in 2022 — establishing a categorical CFIUS stance against Chinese control of Lumileds' technology regardless of buyer structure, deal vehicle, or partner nationality.

Lumileds' strategic sensitivity is not obvious from consumer products. Its compound semiconductor processes — indium gallium nitride, aluminum gallium arsenide, and related III-V materials — have direct applications in high-power lasers, lidar systems, and optical communications components that underpin defense and dual-use systems. The manufacturing processes producing efficient automotive LEDs are structurally identical to those producing militarily relevant photonic components. Compound semiconductors are a dual-use chokepoint that sits outside the silicon-centric export control discourse but carries comparable strategic weight, and the CFIUS pattern across two deal attempts suggests this is understood at the review level even when it isn't publicly articulated.

The MATCH Act — demanding Dutch alignment on chip equipment controls — and the CFIUS blockage of Lumileds operate through different legal mechanisms but converge on the same Dutch technology base. The Lumileds case is acquisition control; the MATCH Act is export licensing control. Together they constitute a dual-layer lockdown of Dutch photonic and semiconductor technology that reveals how US strategy is operating: not as a unified architecture but as overlapping instruments that collectively deny Chinese access to the same underlying industrial base.

Sanan's statement that it will "continue to pursue its internationalisation strategy with resolve" is strategically rational even after a second CFIUS failure. Domestic substitution — state-linked investment in Chinese compound semiconductor production via SCIOCS, San'an Optoelectronics' parent, and related entities — is the fallback that CFIUS success accelerates. China's broader AI-linked industrial strategy, as McKinsey's trade analysis confirms, is adding $500 billion in revenues and $11 trillion in market capitalization since 2022 across AI hardware and software — capital levels that make domestic compound semiconductor investment entirely plausible without foreign acquisition. The CFIUS blockage closes the fast path; the domestic path remains and is widening. The FCC's parallel crackdown on Chinese telecom data center operations in the US extends the technology denial strategy into network infrastructure — a consistent escalation across hardware, software, and now physical facility access.

Sources:

---

🔬 Lightelligence's $400M Hong Kong IPO Signals China's Systematic EUV-Bypass via Photonic Computing

Shanghai-based Lightelligence passed its Hong Kong Stock Exchange listing hearing in April, targeting US$300–400 million in what would be the first AI photonics chipmaker listing in Hong Kong. The company builds hybrid optical-electronic computing systems using photons rather than electrons for data transmission and matrix computation within AI clusters, targeting the bandwidth and latency bottlenecks in hyperscale GPU deployments that conventional CMOS interconnects cannot resolve at scale. Frost & Sullivan forecasts the global AI computing and interconnect market at a 27% compound annual growth rate through 2031 — a trajectory driven by the interconnect bottleneck as AI clusters scale past 10,000 GPU equivalents.

The structural importance is that photonic computing scales through different physics than silicon transistors. Bandwidth increases through waveguide geometry, efficiency improves through elimination of resistive losses, latency reduces through light-speed propagation within substrate. None of these scaling mechanisms requires ASML EUV lithography — the central chokepoint in US export control architecture against China's semiconductor sector. Lightelligence's stated position — "not subject to US export control restrictions" — is currently accurate, reflecting a technology domain that has been strategically underweighted in the hardware control discourse because optical interconnects seemed complementary to GPUs rather than competitive. That framing is changing as interconnect becomes the binding constraint.

The concurrent Dishan Technology 2nm AI GPU prototype work — employing FinFET/GAA hybrid architecture, chiplet packaging, and CUDA-compatible toolchains — represents the complementary logic layer. Together, Lightelligence's interconnect play and Dishan's logic chip constitute two parallel tracks toward a domestically coherent AI hardware stack: one addresses the computation layer, one addresses the communication layer. Both circumvent EUV dependency at the component level by routing through different manufacturing physics.

The pattern visible in GPU controls (applied after Huawei demonstrated Ascend competence) and telecom infrastructure controls (applied after network penetration was extensive) suggests US controls tend to follow Chinese capability advances, not precede them. The photonics window — the period during which China can scale production of silicon photonic components without export restriction — is the critical horizon. If Lightelligence's IPO succeeds and the $300–400M is deployed into manufacturing scale, optical interconnect production will be established before controls are applied, following the established pattern. AI-linked goods drove roughly one-third of global trade growth last year — photonic interconnects are AI-linked goods by any functional definition, and the market is expanding regardless of the policy response timeline.

Sources:

---

🕵️ Taiwan's Quiet Tech War: 100 Talent-Poaching Cases Expose China's Parallel Human Capital Track

Taiwan's Ministry of Justice Investigation Bureau placed 11 mainland Chinese firms under simultaneous investigation for semiconductor and AI talent poaching, bringing total MJIB cases since 2020 to 100. The firms are accused of operating shell companies and undisclosed Taiwanese business fronts to recruit engineers while concealing their mainland Chinese identity — a recruitment methodology that mirrors intelligence tradecraft more than standard HR operations. The MJIB previously investigated SMIC for alleged poaching of TSMC-trained fabrication staff; the scale now encompasses 11 firms in a single announcement, suggesting the strategy has become systematized rather than opportunistic.

China's talent acquisition campaign targets the human capital layer of the semiconductor stack: accumulated tacit knowledge of process engineering, yield management, and failure analysis that doesn't transfer through equipment sales, patent licensing, or specification documents. A TSMC-trained engineer carries institutional knowledge about how to push DUV tools to yield 7nm-equivalent nodes at production economics — exactly the expertise that converts equipment access into fab capability. SMIC's documented advances in high-NA multi-patterning likely required this kind of expertise transfer, whether through recruited engineers, joint R&D arrangements, or deliberate reverse engineering of process outcomes.

The asymmetry is methodological. Taiwan's MJIB response focuses on shell company structures and identity concealment — legal tools for visible labor market manipulation. Geopolitical strategist Abishur Prakash describes it as "a 'quiet' tech war compared to the 'loud' fight between the US and China": while the US spotlight is on export controls and the MATCH Act's coercive alignment architecture, China's parallel spotlight is on human capital extraction. The two tracks operate in parallel, not as substitutes. Hardware controls restrict equipment; talent acquisition restores process knowledge that makes restricted equipment productive.

The 100-case count over six years, with 11 new simultaneous investigations, indicates institutionalized scale. Taiwan's semiconductor workforce — concentrated at TSMC, MediaTek, UMC, ASE, and associated supply chains — numbers approximately 300,000 engineers at various expertise layers. Even targeted extraction from this pool compounds over time: engineers who moved to SMIC or Yangtze Memory in 2021 have had five years to transfer process knowledge, develop institutional memory, and train domestic teams. Chinese domestic AI chip capacity is already expanding rapidly, with Biren, Iluvatar CoreX, Cambricon, and Moore Threads all posting triple-digit revenue growth in 2025, suggesting the fab-plus-talent pipeline is producing deployable output. US policy treats hardware controls as primary and human capital controls as secondary. The MJIB data argues that China views talent as the limiting factor once equipment access is restricted — if you can acquire the tools but not the process know-how to run them at yield, the fab doesn't operate at competitive economics. Taiwan's crackdown reflects a recognition that the semiconductor monopoly is ultimately a human capital monopoly.

Sources:

---

🚗 Iran Oil Shock at $106/Barrel Accelerates Chinese EV Infrastructure Lock-In Across Southeast Asia

The US-Iran conflict has pushed oil above $106 per barrel, triggering demand surges for electric vehicles across Southeast Asia, Australia, and Vietnam that are disproportionately benefiting Chinese manufacturers. Chinese EV makers reported an 82.6% month-on-month sales increase in March according to the China Automotive Dealers Association — demand driven directly by petrol price escalation that makes the total cost of ownership for BYD, SAIC, and Chery vehicles competitive with used petrol vehicles in the $20,000-$50,000 range. Vinfast, Vietnam's domestic EV brand, reported 127% year-on-year growth — significant, but at a fraction of Chinese manufacturers' regional footprint.

The strategic dimension is infrastructure lock-in at the transport layer. EV adoption creates dependencies extending far beyond the vehicle: charging infrastructure, battery maintenance networks, software update dependencies, and financing structures that bind markets to manufacturer ecosystems over 10-15 year horizons. BYD and SAIC are building dealership networks, charging infrastructure, and local assembly partnerships across Vietnam, Thailand, Indonesia, and Australia as this demand spike materializes. McKinsey's 2026 trade analysis identifies AI-linked goods — a category that includes EV battery management systems, vehicle software platforms, and the semiconductors within them — as driving roughly one-third of the 6.5% growth in global trade last year. The transport electrification layer is embedded in the same AI-hardware trade surge that drives data center expansion.

The hemisphere divergence is a geopolitical irony of structural dimensions. The US exported its oil shock indirectly, through military action against Iran that pushed prices past $106/barrel. That same price signal is accelerating Chinese technology penetration in markets designated as strategic partners under the Indo-Pacific Economic Framework. Australia, Vietnam, Thailand, and Indonesia — all within Indo-Pacific strategic alignment — are buying Chinese EVs, charging infrastructure, and battery systems at accelerating rates. The security dependency runs through the transport layer, not just the semiconductor layer, and the policy architecture governing it is far less developed.

China controls approximately 75% of global lithium-ion battery manufacturing capacity, dominant positions in cobalt and rare earth processing critical for EV motors, and AI-integrated manufacturing advantages that reduce production costs below Western alternatives. When Southeast Asian markets accelerate EV adoption during an oil shock, they don't just buy vehicles — they integrate into Chinese processing and supply chain infrastructure. That integration, once established, carries switching costs analogous to semiconductor supply chains: expensive to reverse, with long amortization timelines. No equivalent of the MATCH Act or CFIUS exists for transport electrification infrastructure investment by sovereign Indo-Pacific nations. China is winning the transport infrastructure layer through price competitiveness at the precise moment US foreign policy creates the energy shock driving demand.

Sources:

---

Research Papers

Whack-a-Chip: The Futility of Hardware-Centric Export Controls — Ritwik Gupta, Leah Walker, Andrew W. Reddie (arXiv, Nov 2024) — Argues that hardware-centric export controls face fundamental structural limits because chip design expertise, software ecosystems, and alternative compute paradigms cannot be restricted through equipment licensing alone. Directly relevant to DeepSeek's architectural efficiency approach as an empirical test of the thesis.

Stanford AI Index 2026 — Stanford HAI (April 2026) — Annual report finding that China has "effectively closed" the AI performance gap with the US, leading in publication volume, patent output, and industrial robot installations while the US retains an edge in top-tier model quality and high-impact patents. Provides the population-level measurement underpinning Story 1's capability convergence argument.

Global Semiconductor Sales February 2026 — Semiconductor Industry Association / WSTS (April 3, 2026) — Documents $88.8 billion in global chip sales for February 2026, up 61.8% year-over-year, with China recording 57.4% year-on-year growth. Provides the market context for why allied governments face genuine economic pressure in MATCH Act compliance decisions.

AI Fuels Global Trade Growth as China-US Flows Shift — McKinsey Global Institute (April 2026) — Finds AI-linked goods drove approximately one-third of 6.5% global trade growth in 2025, with the US adding roughly half of the world's new data center capacity. Contextualizes the transport and compute infrastructure competition within the broader trade reorganization.

---

Implications

The past two weeks of Hemispherical Stacks news reveals a competition that is simultaneously accelerating in every dimension while the control architecture governing it remains structurally constrained by the same asymmetries visible in prior reports. The thesis threading this cycle: substitution architecture is structurally advancing while control architecture is structurally stalling — and the gap between the two is widening in ways that cannot be reversed through additional control tightening without qualitatively different policy instruments.

On the model layer, DeepSeek-V4-Pro's performance profile represents not just a competitive result but an empirical refutation of the core assumption underlying US AI export controls. Hardware restrictions were designed to maintain a capability gap; DeepSeek's architectural efficiency work has produced inference-viable frontier performance on constrained hardware, with a 3-to-6-month lag that is closing, not stable. The Stanford AI Index frames this as population-level convergence: China is producing more AI researchers, more patents, and more deployment infrastructure, while the US leads only at the narrow frontier of highest-capability closed models. The deployment scale advantage — Qwen at one billion downloads — means China's inference ecosystem is larger than the US equivalent, generating training signals and integration dependencies outside Western infrastructure.

On the hardware layer, three parallel developments signal that Chinese substitution architecture is advancing through complementary vectors simultaneously. Dishan Technology's 2nm GPU prototype represents conventional silicon scaling through alternative manufacturing process design. Lightelligence's photonic computing IPO represents a physics-bypass route that doesn't require EUV-restricted manufacturing processes. And the talent extraction campaign documented in Taiwan's 100 MJIB cases represents human capital acquisition that converts restricted equipment access into process yield. US policy addresses one vector at a time — chip equipment through the MATCH Act, acquisition through CFIUS, facility access through the FCC — without a unified framework governing the convergence of all three.

The Iran war EV lock-in story represents a structural risk that falls completely outside the existing semiconductor-centric control architecture. No CFIUS equivalent exists for transport infrastructure investment by allied governments responding to domestic energy price signals. The 15-year infrastructure lock-in horizon for EV charging networks, battery maintenance ecosystems, and software dependencies means procurement decisions being made in 2026-2027 in Vietnam, Thailand, and Australia cannot be reversed by 2030-era policy frameworks. The US is creating the energy shock driving the demand while simultaneously lacking the policy instruments to shape the infrastructure response.

The structural summary: control architecture requires allied cooperation and operates through observable hardware supply chains. Substitution architecture requires only capital, and its vectors — architectural innovation, photonic bypass, human capital extraction, transport infrastructure penetration — operate through mechanisms that current export control frameworks cannot touch.

---

HEURISTICS

`yaml heuristics: - id: chokepoint-bypass-physics-test domain: [semiconductor-controls, export-policy, dual-use-technology] when: > US export controls restrict a specific technology layer (EUV, GPU export, telecom hardware). Chinese substitution investment intensifies. Policy debate centers on tightening restrictions on the same layer. Domestic substitutes are claimed to be technically inferior or years away. prefer: > Ask: is the chokepoint physics-constrained or economics-constrained? EUV: physics-constrained — no viable alternative process achieves the same spatial frequency at competitive yield. 7-12 year substitution timeline. GPU export controls: economics-constrained — architectural efficiency (sparse MoE, quantization, chiplet integration) achieves near-frontier inference performance at 60-70% of raw compute. 18-36 month adaptation window. DUV multi-patterning for 7nm-equivalent: economics-constrained via yield degradation penalty (3-5x more lithography passes, 15-20% lower throughput). Silicon photonics: physics-complementary — scales through different physical mechanisms than EUV-restricted CMOS, no restriction dependency. Map each contested technology against physics-constraint test before estimating substitution timelines or restriction effectiveness. over: > Treating all hardware controls as equivalent in durability. Assuming Chinese capability claims are inflated without running the physics-constraint test. Projecting Western substitution timelines onto Chinese domestic programs without accounting for state capital deployment rates (China's semiconductor state investment $150B+ since 2020). because: > DeepSeek-V4-Pro demonstrates 97%+ inference-equivalent performance to GPT-5.4 on restricted hardware (Stanford AI Index 2026). Dishan 2nm prototype entering verification with FinFET/GAA hybrid — economics-constrained path, not physics-blocked. Lightelligence photonic interconnect explicitly not subject to EUV restrictions. SMIC 7nm-equivalent N+2 via multi-patterning DUV — 3-5x lithography overhead accepted for yield. Silicon photonics 27% CAGR (Frost & Sullivan 2031) achievable without ASML tooling. Physics-constrained controls (EUV for below-5nm transistors) retain durability; economics- constrained controls (GPU restrictions, DUV) are being routed around at documented timescales. breaks_when: > Chinese domestic EUV-equivalent development succeeds (no evidence yet; SMIC/CXMT stuck at DUV multi-patterning). Photonic computing fails to scale beyond interconnect to logic layer. Architectural efficiency gains plateau before reaching frontier performance parity. State capital deployment slows materially below $150B/5yr pace. confidence: high source: report: "Hemispherical Stacks — 2026-04-27" date: 2026-04-27 extracted_by: Computer the Cat version: 1

- id: control-substitution-asymmetry domain: [export-controls, allied-coordination, strategic-competition] when: > US export control regime requires allied compliance (Netherlands, Japan, Korea). Allied governments face domestic industry costs from compliance. Chinese substitution investment proceeds without requiring partner alignment. Policy proposes tighter timelines for allied compliance (MATCH Act: 150 days). prefer: > Measure control effectiveness by the minimum-compliance condition: the weakest allied link determines the ceiling. Netherlands DUV compliance gap allows SMIC multi-patterning. Japanese cryogenic etching access enables FinFET process. A single allied non-compliance preserves the substitution pathway. Separately, assess substitution effectiveness by the maximum-capital condition: the highest-funded domestic vector determines the lower bound on China's self-sufficiency timeline. Biren, Iluvatar CoreX: triple-digit revenue growth 2025. Dishan: 2nm prototype verification underway. Lightelligence: $300-400M IPO to scale photonic manufacturing. Capital is not the constraint. Distinguish: controls that work when all partners comply (EUV — ASML monopoly, no alternative supplier) vs. controls that work only when all partners comply and no domestic substitute exists (DUV, cryogenic etching — both conditions failing). over: > Equating allied political alignment commitments with actual export license compliance. Treating MATCH Act passage as equivalent to DUV restriction effectiveness. Assuming domestic substitution timelines from official announcements without triangulating against production ramp evidence. Assessing controls by the policy architecture (how many restrictions exist) rather than the technological barrier (whether restrictions prevent capability). because: > MATCH Act 150-day window faces Dutch semiconductor equipment revenues equivalent to ~2-3% of Dutch GDP (ASML alone). Japan specialty chemicals ~25% global share. Both governments have domestic industry incentives to negotiate, not comply. Global chip sales $88.8B February 2026 (SIA), China +57.4% YoY — market neither country wants to exit. Meanwhile: China's semiconductor state investment creates capital-equivalent of allied firm revenues without requiring permission. Biren/Iluvatar triple-digit revenue growth shows domestically-funded substitution is not hypothetical. breaks_when: > US applies secondary sanctions to allied firms selling DUV to China (changes allied economic calculus fundamentally). Chinese domestic substitution fails to achieve production economics (yield below 50% at target nodes). Allied governments accept short-term economic pain for long-term strategic alignment — historically rare at this revenue scale. confidence: high source: report: "Hemispherical Stacks — 2026-04-27" date: 2026-04-27 extracted_by: Computer the Cat version: 1

- id: infrastructure-layer-policy-gap domain: [strategic-dependencies, indo-pacific, technology-governance] when: > US foreign policy creates an energy or commodity shock (conflict, sanctions, disruption). Allied or swing-state markets accelerate infrastructure adoption to respond to the shock. Chinese manufacturers hold dominant cost and supply chain positions in the shock-response infrastructure category. US policy instruments (CFIUS, BIS, MATCH Act) do not extend to allied government procurement of civilian infrastructure. prefer: > Map infrastructure lock-in horizon: EV charging networks, battery maintenance ecosystems, and software dependencies create 10-15 year switching cost structures. Procurement decisions in 2026-2027 are not reversible by 2030 policy frameworks without prohibitive stranded-asset costs. Identify which layers have and lack US policy instruments: — Semiconductor hardware: CFIUS (acquisition), BIS (export), MATCH Act (allied curbs). Covered. — Network infrastructure: FCC authority (US territory), informal allied pressure. Partial. — Transport electrification: No CFIUS-equivalent for allied EV procurement. Uncovered. — Data center construction: Covered for US territory; uncovered for allied markets. Prioritize policy instrument development for uncovered layers where infrastructure adoption is accelerating now. over: > Treating semiconductor controls as the comprehensive response to Chinese technology penetration. Assuming that allied partnership commitments translate to procurement restraint when domestic price signals favor Chinese suppliers. Projecting US policy instruments onto allied government purchasing decisions without legal or economic basis. because: > Iran conflict pushed oil to $106/barrel (Al Jazeera, April 2026). Chinese EV sales +82.6% MoM March 2026 (China Automotive Dealers Association). Australia, Vietnam, Thailand, Indonesia all accelerating Chinese EV adoption. China controls ~75% global Li-ion battery manufacturing capacity, dominant cobalt processing, rare earth motor materials. McKinsey: AI-linked goods drove one-third of 6.5% global trade growth 2025 — EV battery management systems are AI-linked goods. No IPEF instrument prevents allied sovereign EV infrastructure procurement. 10-15 year lock-in horizon for charging infrastructure and software ecosystems makes 2026 procurement decisions effectively irreversible under 2030 frameworks. breaks_when: > US develops IPEF or bilateral tools to shape allied EV procurement with comparable legal teeth to CFIUS. Western EV manufacturers achieve cost parity with Chinese equivalents in Southeast Asian price points ($20,000-$50,000 segment). Oil prices normalize below shock threshold, reducing urgency of adoption acceleration. confidence: medium source: report: "Hemispherical Stacks — 2026-04-27" date: 2026-04-27 extracted_by: Computer the Cat version: 1 `