URL: nytimes.com Date: 2026-05-14 (during the Trump-Xi Beijing summit) Class: foundational (interview advancing a specific framework: China is running multiple races, not the AGI race)
A New York Times "Interesting Times" interview, published 2026-05-14 during the Trump-Xi Beijing summit, in which columnist Ross Douthat questions Kyle Chan, a Brookings foreign-policy fellow who studies China and AI. Chan argues that China is not running the United States' AGI race but a parallel set of races on efficiency, diffusion, open source, and robotics applications, that Beijing "is not A.G.I.-pilled," and that the binding policy problem is U.S. recklessness driven by the AGI-race framing. Chan locates the medium-term risks worth taking seriously in cyber and biosecurity rather than in AGI takeover, which he describes as "overblown."
Chan's "multiple races" framework
Chan's central argument distinguishes the race the United States is running from the races China is running. The U.S. effort, in his account, is oriented toward AGI or superintelligence, which Douthat glosses as "almost machine god"; the roughly $1 trillion valuations attached to OpenAI and Anthropic make that bet legible. China, Chan argues, is instead competing on several other dimensions at once:
- Efficiency — smaller, cheaper, easier-to-deploy models, a focus he attributes in part to chip constraints.
- Diffusion — open-source release, by labs including DeepSeek, Qwen, and Moonshot AI, to maximize global adoption.
- Applications — especially robotics, including autonomous delivery, restaurant waiters, hotel service, and drone food delivery.
Within China's lab landscape, Chan names DeepSeek as the closest equivalent to the American frontier labs, with Z.ai (Zhipu) and Moonshot AI as smaller startups in the roughly $40-50 billion market-capitalization range.
"Beijing is not A.G.I.-pilled"
Chan grants that individual Chinese tech founders sometimes sound like their U.S. counterparts, but argues that the state and the labs are not running an AGI Manhattan Project. His main evidence is a revealed preference: China said "thanks, but no thanks" to H200 access, and, in Chan's reasoning, if Beijing believed AGI were imminent "they would've gobbled up those chips as quickly as possible." Douthat offers the line "The closer you are to the machine god, the more its voice whispers in your ear," with which Chan agrees.
China's structural strengths and weaknesses
Chan frames two structural strengths and two structural weaknesses. On energy, he describes clean-energy capacity (solar, wind, batteries) being deployed at pace, data centers being sited in western provinces to leverage renewables, and a deliberate geographic redistribution of compute toward that capacity. On workforce, he argues that China's birthrate, around two-thirds of the U.S. rate, combined with a manufacturing labor shortage, gives humanoid robots a role filling labor gaps rather than displacing workers.
The weaknesses are chips and the economics of intellectual-property theft. China is cut off from the TSMC and ASML supply chain, and Huawei's domestic chips remain, in Chan's account, not as good as Nvidia's. On IP, Chan describes distillation from Anthropic and OpenAI models via proxy accounts as useful for catching up but insufficient to build a frontier model from scratch.
Cyber and biosecurity as the medium-term risks
Chan reorders the risk hierarchy: "A.G.I. risk has been overblown… The cyber-risk and the biosecurity risk… have been underestimated up until recently." These, in his view, are the risks for which export controls are worth maintaining. He cites Mythos as the concrete near-term concern, arguing that a deployment lead of a few months on cyberattack capability is operationally meaningful.
Engagement versus confrontation
Chan rejects the "all-out race" framing on the ground that it drives U.S. recklessness. He sees engagement with China as worth pursuing on open-source risks, non-state-actor arbitrage across U.S. and Chinese models, and biosecurity. He sees engagement as not worth pursuing, for now, on binding constraints, verification, and treaties, which he calls "way too early" given structurally low trust between Washington and Beijing. He adds the caution that "We might be waiting for [a major cyber or bio incident] to happen before we take action."
China's social policy on AI
Chan describes Beijing increasingly discussing AI-driven welfare-state needs, an echo of the U.S. welfare-state debate. He notes that China is already regulating "A.I. boyfriends and A.I. girlfriends," which he attributes to the same Confucian-productivity logic that drove the earlier video-game and ed-tech crackdowns. He describes the "iron rice bowl" era as gone and AI-job-displacement anxiety as rising, citing young-graduate unemployment in China at around 17 percent, roughly double the U.S. rate, with more than 12 million new college graduates entering the market this year.
Key claims
| Claim | Confidence | Notes |
|---|---|---|
| Chinese models are 3-9 months behind the U.S. frontier | high | Standard analyst consensus; matches Epoch ECI data. |
| China declined H200 chips to bolster Huawei | high | Reported widely in May 2026; Trump quoted. |
| Beijing is "not A.G.I.-pilled" | medium | Chan's argument from revealed preferences (chip decline); not directly stated by PRC officials. |
| China's energy build-out is structurally faster than the U.S.'s | high | Multiple sources; AI-policy/IEA data align. |
| Open-source diffusion is China's deliberate strategy | high | DeepSeek, Qwen, Moonshot AI release behavior is consistent. |
| Distillation from U.S. models via proxies is occurring | medium | OpenAI and Anthropic have alluded; Chan confirms; not fully audited. |
| U.S. recklessness driven by AGI-race framing | low-medium | Chan's editorial framing — contested by Pentagon/AISI side. |
Reception and relation to competing framings
The interview functions as an entry point for the argument that China is not running the AGI race, and it sits in contrast to the AGI-race framing advanced in Situational Awareness: The Decade Ahead and Anthropic Machines Of Loving Grace. Where those sources imply a sprint backed by export controls and AISI scrutiny, Chan's account implies maintaining controls, engaging on cyber and bio, and deprioritizing the AGI race. The two framings carry an empirical test on chip acquisition: the AGI-race reading predicts Beijing buys all available Nvidia chips, while the multiple-races reading predicts Beijing declines the H200. On the May 14 summit, Trump publicly said China "chose not to" buy, which Chan reads as support for the multiple-races account.
Chan's reordering of risks toward cyber and biosecurity parallels the intermediate-risk emphasis of the Veilleux-Lepage framework. His position on export controls supports maintaining current controls because of cyber and bio risk rather than because of AGI, situating him in the middle of the export-controls debate. He names the JD Vance Paris-speech frame — "We should not have hand-wringing over AI safety slow down… American AI development" — as the policy posture he believes is now backfiring as cyber and bio risks materialize.
Several tensions remain unresolved within the interview. Douthat presents a 2-4-6-year-takeoff scenario, and Chan acknowledges he cannot refute the timeline uncertainty. Chan's own emphasis on a few months' cyber lead implies that some race framing survives; he threads this by arguing for racing on cyber and bio but not on AGI. And his "Beijing isn't A.G.I.-pilled" claim rests on revealed preference and could change if China alters its chip-acquisition posture.
Relationships
- supports: US-China AI Competition: Different Races, Different Metrics, AI Diffusion, Open-Source AI / Open-Weight Models, Embodied AI vs AGI (China's Race Redirect)
- contradicts: Situational Awareness: The Decade Ahead (AGI-Manhattan-Project framing); aspects of 2028: Two Scenarios for Global AI Leadership (Anthropic) (which argues a "decisive 12-24 month lead" is necessary and possible)
- depends-on: Export Controls (AI), Scaling Laws, Distillation
- related: Kyle Chan (Brookings; entity created), Ross Douthat (NYT columnist; entity created), Beyond Misuse: Artificial Intelligence, Grievance, and the Future Landscape of Political Violence — Yannick Veilleux-Lepage (Combating Terrorism Center at West Point, April 2026) (intermediate-risk framework parallel), DeepSeek, Moonshot AI, Zhipu AI, JD Vance (Paris speech reference)
- regulated-by: Bureau of Industry and Security (BIS) (export-controls implementer)
Wiki Folding
Updates US-China AI Competition: Different Races, Different Metrics (Chan-framework section and May 14 H200-decline empirical test), AI Diffusion (China's open-source strategy framing), AI Political Cleavages (Trump-Xi summit framing reference), Embodied AI vs AGI (China's Race Redirect) (China robotics-focus thesis), and creates entity pages for Chan and Douthat.