Superpowers and AGI: How U.S.-China Competition Could Be Shaped by the Emergence of Artificial General Intelligence (CNAS, August 2026)

A 32-page report published on August 27, 2026 by the Center for a New American Security, written by Jacob Stokes, a senior fellow and deputy director of the CNAS Indo-Pacific Security Program who previously served on the national security staff of then-Vice President Joe Biden and as a professional staff member of the U.S.-China Economic and Security Review Commission.

The report's stated method is to bracket the technological question. Rather than judging whether or when AGI will arrive, it "stipulates a world where AGI is imminent or already exists" and then applies "informed speculation rooted in the current contexts and histories of U.S.-China strategic competition, international relations, and global security." Its framing claim is that building AGI would not by itself confer dominance: "the state with AGI would have to apply it to translate technological capability into wealth, power, and influence," in ways that are cost-effective against alternatives, and "sometimes that AGI capabilities-to-power conversion could prove unworkable."

On definitions, the report treats AGI as contested — quoting Sam Altman that the concept is "weakly defined" — and adopts a working sense of an AI system reaching human levels of performance across a wide variety of tasks, noting Google's addition of generalization ability and common-sense knowledge. It distinguishes AGI from superintelligence, a system that "greatly exceeds the cognitive performance of humans in virtually all domains of interest," and from transformative AI, which is defined by societal effect rather than capability. It records timeline disagreement, from Mustafa Suleyman's position that AI is "already superhuman" through industry estimates of one to three years to independent estimates in the early 2030s or mid-2040s, and argues emergence is unlikely to be binary.

Four lenses on whether China is pursuing AGI

The report's central empirical section states that "China is not currently pursuing AGI as a formal, state-directed goal," noting that Beijing uses a different term — "general artificial intelligence," or "general-purpose AI." It then sets out four readings, concluding that China's approach "might be best understood as a messy and still-developing mix of all these lenses," and that "AGI is not the North Star for the AI community in China the way it is in Silicon Valley — at least not yet, and it might never be."

The report notes that tracking is hard because AI's physical constituents are common commercial items and the Chinese system is opaque, making verification harder than for nuclear activities.

Five mechanisms and four phases

The report proposes five mechanisms by which AGI could affect the balance of power: boosted economic growth and productive capacity; consolidated influence over the global information environment; new military capabilities; large-scale misalignment or loss of control; and downstream political effects. It presents each with limits attached, describing the aim as "inject[ing] some realism into a debate that often veers into the fantastical."

It then examines the choices facing Washington and Beijing across four phases: both states pursuing AGI; one state on the cusp; one state possessing AGI while the other does not; and both states having it. A fifth possibility is that the phases blur, if emergence proves "neither discrete nor observable," which would change policymakers' calculus.

Four categories of conflict risk are identified: preventive action by either power to stop the other crossing an AGI-related threshold (more likely China as attacker given the current US lead in frontier models, though a Chinese surprise breakthrough could invert this); a power transition if China reaches AGI first and gains first-mover advantages or the ability to deny the technology to the United States; escalating racing dynamics in which "safety engineering could fall by the wayside"; and inadvertent conflict from new capabilities testing operational norms, or from AGI malfunctioning "in a way that hurts both the United States and China."

Recommendations

Nine recommendations are addressed to US policymakers. They are positions advanced by the report, not established findings.

1. Balance near- and long-term AI bets — pursue and prepare for AGI without overconcentrating resources in moonshot bets, since "the U.S.-China AI race is not a singular one but rather multiple simultaneous contests" and the gap between AGI development and effective application would give a laggard state a window to become a fast follower.
2. Ensure US resources and technology do not aid China's AI development — keep, restore and where possible tighten export controls on advanced chips and chip-making equipment, and impose costs for distillation of US models by PRC-affiliated entities. The report notes China taking reciprocal steps, including blocking a China-founded company from merging with a US firm, regulating outbound investment, and restricting travel by Chinese AI engineers and executives.
3. Create diplomatic channels with allies and with China — two mechanisms stood up by the Secretary of State, using the May 2026 Trump–Xi agreement to establish a US-China AI channel as the vehicle while avoiding confinement to economic and trade matters.
4. Track inputs to AGI and performance thresholds — direct the intelligence community to produce integrated assessments covering compute available to Chinese labs and PLA units, the pace of algorithmic breakthroughs, and relative US-Chinese model performance, centralized in an IC-wide task force modeled on the CIA's Alec Station or the National Counterterrorism Center, and written into presidential intelligence-priority directives.
5. Conduct a scenario exercise on intervening to stop China from developing AGI — and use the findings to develop a strategy based on realistic assumptions about available intelligence and the diplomatic, espionage, cyber or kinetic capabilities that could be used.
6. Prepare to reverse engineer or steal AGI technology if China gets it first — the report argues US intelligence has not been organized for technological espionage for decades and would need to overhaul its approach, justifying such operations by reference to China's record of intellectual-property theft, reciprocity, and the risk of an enduring power advantage, while stating that authorities "should be narrowly scoped and should not enable intellectual property theft beyond PRC targets linked with AGI."
7. Limit and counter the reach of authoritarian digital infrastructure — incentivize non-PRC platforms at home and abroad, which the report calls "the technological heir to the internet freedom debates of an earlier era."
8. Provide AGI alternatives compatible with democracy — arguing that AGI engineered to reflect democratic values would carry more legitimacy than one instrumentalized for narrow US interests or for China's system.
9. Build programs to cushion AGI's social effects — creating shock absorbers in advance, on the argument that "resilience and societal cohesion is the bedrock on which American power stands."

Across these, the report's framing is that "strategic rivalry and the need for some level of U.S.-China coordination will exist in a state of constant tension," and that policymakers "will have to manage that tension instead of trying to resolve it by picking either unmanaged competition or unrealistic cooperation."

Relationships

• related: Center for a New American Security (CNAS) — publisher
• depends-on: Artificial General Intelligence (AGI) — the contested definition the report works from
• related: US-China AI Competition: Different Races, Different Metrics — the competition the report analyses
• related: AI Race Dynamics — the racing dynamic the report's third risk category describes
• related: Export Controls (AI), Compute Governance — the instruments recommendation 2 turns on
• related: Xi Jinping, Sam Altman — figures quoted