Proactive Elite Alignment Theory (PEAT) is an analytical framework developed by Charles Sun in his research on Chinese state-business dynamics and applied to China's AI sector in The Incentive Architecture Export Controls Cannot Reach (The Incentive Architecture Export Controls Cannot Reach (Sun, Lawfare, May 6 2026), Lawfare, May 6 2026). It identifies four structural variables that, according to Sun, predict when firms in China's technology sector preemptively align with state priorities without being directly ordered to, and argues that all four are active in China's current AI subsidy landscape.
The four variables
Sun frames PEAT around four structural variables.
Central strategic signaling refers to unambiguous central-policy direction. In the AI case, the State Council has designated AI as a national priority; eight central ministries have jointly issued AI development guidelines, such as the January 2026 "AI Plus Manufacturing" guideline led by MIIT; and "computing-electricity coordination" appeared in the 2026 Government Work Report. Sun argues that when the central signal is this clear, local governments do not merely implement it but compete to over-implement, each city layering additional incentives to attract AI firms before rival jurisdictions do.
Cascading local escalation resembles, in Sun's account, China's pandemic-era "dynamic zero" campaigns, in which central-policy direction triggered local escalation far beyond what Beijing initially mandated. In the AI case he cites Shenzhen's 4.5-billion-yuan combined digital-economy/tech package plus 200M-yuan March 2025 training-power vouchers; Beijing Yizhuang's 100M-yuan/year compute vouchers with chip-origin differentiation; and Hangzhou's 250M-yuan/year program. Each city's program, on this reading, is a bid to outpace the others rather than a response to a specific central directive.
Resource dependence operates through chip-origin-differentiated subsidy rates that lock firms into domestically produced compute. Beijing Yizhuang offers 40% reimbursement for domestic chips and 30% for non-domestic; Hangzhou offers 30% for domestic and 20% for non-domestic. Sun argues the differential is small per transaction, but over a year of model training a 10-percentage-point premium amounts to millions of yuan. He characterizes the dependence as structural rather than contractual: model architectures, engineering workflows, and cost structures optimize around the subsidized platform, so switching costs amount to reengineering an entire production pipeline rather than changing a contract clause.
Organizational capacity refers to the speed with which Chinese AI firms restructure their engineering pipelines around subsidized infrastructure. Sun describes this as the operational complement of resource dependence: firms must be capable of capturing the subsidy structure for the alignment dynamic to operate.
Export-control-induced feedback loop
Sun's central causal claim is that US export controls strengthen rather than weaken the system, through a four-step loop. First, export controls restrict the supply of advanced foreign chips to Chinese firms, the intended effect. Second, restricted foreign chip supply increases dependence on domestic compute, whose pricing is shaped by state subsidies and whose availability is channeled through government-backed programs, deepening the resource-dependence variable. Third, deeper resource dependence strengthens the incentive for firms to align with state priorities: by making domestic compute cheaper through differential reimbursement rates, the economically rational choice and the politically preferred choice point in the same direction, so firms align because the price structure makes alignment the path of least resistance. Fourth, aligned firms use every available pathway to acquire capabilities that unlock further state resources, and Sun treats distillation of American frontier models as one such pathway, framing it not primarily as espionage or defiance but as a rational response to an incentive structure that rewards capability acquisition from every available source. The loop closes when distillation triggers calls for tighter export controls, which further restrict foreign chip supply, deepening resource dependence and strengthening the alignment incentive.
Boundary condition
Sun argues the loop is not perpetual. It holds only as long as domestic computing is sufficient to produce competitive models; if the performance gap between domestic chips and restricted foreign alternatives becomes unbridgeable, the incentive structure breaks down, because resource dependence without resource adequacy produces failure rather than alignment. Citing Lennart Heim's RAND analysis, Sun argues that China can still develop competitive models despite compute constraints, because frontier training consumes only a fraction of total compute capacity, and concludes that the boundary condition has not been reached.
The Manus AI case as coercive backstop
Sun presents PEAT as predicting that the incentive architecture has a coercive backstop for firms that reduce their economic pull, and treats the Manus AI case as the canonical operative example. Manus was built on Anthropic's Claude as its US foundation model, had a global subscription base, was headquartered in Singapore, raised US venture capital, and was the subject of a $2B Meta acquisition in December 2025. On April 27, 2026, China's foreign investment security review mechanism, an interagency body comprising the NDRC and Ministry of Commerce, ordered withdrawal of the deal, citing only "laws and regulations," and exit bans were placed on the co-founders. Sun's reading is that Manus had systematically minimized state-ecosystem dependencies, both by design through its US foundation model and global subscription base and by later strategic choice through its Singapore HQ and US VC, so that the clearest available case of an attempted exit ended in state veto, indicating the backstop is operative rather than merely latent.
Engagement with US policy debate
Sun frames his argument as a structural claim rather than a brief against export controls: the incentive system on the Chinese side, in his account, is not addressable by supply-side US tools. He argues that demand-side tools would have to operate on the Chinese internal price structure, outside the reach of the Entity List, IEEPA, and ECRA.
The framework engages explicitly with two prior positions. Joe Khawam (Just Security, Feb 2026) proposed phased sanctions escalation targeting Chinese AI distillation, treating distillation as behavior that sanctions can deter. Ryan Fedasiuk (War on the Rocks, April 2026) proposed regulatory tools to limit the diffusion of Chinese open-source models. Sun argues that both are correct about the tools available to the US but that neither engages the incentive system on the other side. PEAT is positioned in opposition to this "sanctions-as-deterrence" framing on the grounds that it treats distillation as bad-actor behavior and so misses the structural incentive.
PEAT relates to several adjacent positions. Against Export Controls (AI), Sun argues export controls deepen rather than disrupt the Chinese incentive architecture; he acknowledges the boundary condition of domestic adequacy has not been reached but argues the loop is nonetheless operative. Relative to the "multiple races" framework attributed to Kyle Chan (China's Not the Problem. We Are. (NYT Interesting Times, Douthat-Chan, May 14 2026)), Sun's account is offered as complementary, explaining why China can compete on the diffusion and applications races even without parity on the frontier-model race. In relation to Adversarial Distillation and Distillation, PEAT recasts distillation as a rational response to incentive structure rather than primarily espionage or defiance. In relation to Sovereign AI (Product Concept), PEAT identifies the resource-dependence dimension as the structural anchor of the sovereign-AI position.
Sun frames PEAT as generating falsifiable claims. He projects continued resource dependence even after relocation or exit attempts, citing Manus and anticipated future cases; increased subsidy-rate differentiation as export-control pressure intensifies, to be tracked through 2026-27 city-program updates; persistent distillation pressure so long as the domestic-adequacy boundary condition holds, to be tracked through Khawam- and Fedasiuk-style sanctions and their measured effect on distillation activity; and expanded da moxing bei'an scope as the registration system matures, to be tracked through CAC enforcement actions.
Relationships
- supports: The Incentive Architecture Export Controls Cannot Reach (Sun, Lawfare, May 6 2026) (canonical source)
- complements: China Generative AI Registration (da moxing bei'an / 大模型备案) (the regulatory gate that filters AI firms for state-content alignment — the structural anchor of the framework)
- complements: China's Not the Problem. We Are. (NYT Interesting Times, Douthat-Chan, May 14 2026) (Kyle Chan multiple-races framework) — both reject the framing of US-China AI competition as a single race
- contradicts: Khawam-Fedasiuk-style "sanctions-as-deterrence" framing — Sun argues these treat distillation as bad-actor behavior, missing the structural incentive
- related: Export Controls (AI), Adversarial Distillation, Distillation, Sovereign AI (Product Concept), AI Political Cleavages, US-China AI Competition: Different Races, Different Metrics
Sources
- The Incentive Architecture Export Controls Cannot Reach (Sun, Lawfare, May 6 2026) — Sun, The Incentive Architecture Export Controls Cannot Reach, Lawfare, May 6 2026 (canonical anchor)
- Sun's prior PEAT research (citation in the Lawfare essay) — foundational framework paper