Jacob Stokes spent years advising Joe Biden on national security before Biden became vice president. His new paper, published through the Center for a New American Security, argues the United States should start seriously preparing for a scenario in which stopping China from reaching artificial general intelligence first requires "extreme measures," a category that, in his own framing, includes military strikes on Chinese data centers. He wants the Pentagon and the NSA to begin assessing now what intelligence they'd need to justify taking that kind of action later.
It's worth being precise about what this actually is and isn't. CNAS is a real, well-connected Washington think tank, co-founded by Kurt Campbell, who went on to coordinate Indo-Pacific policy at the White House, and it has genuine influence on how US-China defense policy gets discussed in Washington. But a report from an outside think tank is not government policy, and nothing in the coverage of Stokes's paper suggests the Pentagon or the White House has adopted his recommendations. What it represents is closer to an idea being seeded into the policy conversation by someone with real institutional credibility, the kind of proposal that starts as a paper and, over years, sometimes migrates into something officials actually consider.
The premise underlying Stokes's argument is itself contested. AGI, artificial general intelligence capable of matching or exceeding human reasoning across the board, remains entirely theoretical; nobody has built one, and there's no scientific consensus it's achievable on any predictable timeline. Stokes writes as though its arrival is inevitable and treats the real question as only when, and who gets there first. Critics quoted alongside coverage of his report have pushed back on exactly that framing, noting the concept remains speculative and that, by most current benchmarks, American AI labs still lead their Chinese counterparts rather than trailing them. Stanford's 2026 AI Index found the US produced 59 notable AI models last year against China's 35, with American private AI investment running north of $285 billion compared to roughly $12 billion in China, though Stanford's own researchers caution that figure understates Beijing's real spending given how much flows through state-backed funds rather than private capital.
China's actual position in this race is more complicated than either "hopelessly behind" or "about to overtake" captures cleanly. Nvidia still commands roughly half the world's installed AI chip capacity and, by some measures, two-thirds of usable computing power, and Huawei, China's most credible domestic chip challenger, remains technologically behind Nvidia's latest generations. But Chinese chipmakers have been compensating by clustering more chips together rather than matching Nvidia's per-chip performance, and the harder, less-discussed front is software: Nvidia's CUDA platform has functioned for years as a kind of moat, letting developers write AI software that runs efficiently across generations of Nvidia hardware, and Chinese firms are now working seriously to build a comparable software layer of their own. If they succeed even just within China's domestic market, analysts at Bruegel have argued, the durable American advantage narrows considerably, and it happens on the software side, not the chip side that gets most of the political attention.
China's government response arrived fast and via its own state media rather than diplomatic channels, the Global Times ran a piece calling the proposal "indistinguishable from terrorism" and accusing CNAS of dressing up "tech-hegemony" ambitions as strategic necessity. That's obviously not a neutral assessment, it's Beijing's own state press making its case, but the underlying point about escalation risk isn't unique to Chinese messaging. Independent commentary flagged in coverage of the report has raised the same concern in plainer terms: an attack on infrastructure inside a nuclear-armed state, however narrowly framed as targeting computing capacity rather than people, carries a genuinely unpredictable risk of escalating into something neither government could fully control once it started. Stokes himself reportedly acknowledged as much, conceding the options he's describing "could create severe geopolitical tensions."
What actually matters here isn't whether this specific proposal gets adopted, most think-tank papers don't become policy verbatim. It's that a credentialed former White House adviser felt confident enough in the current political climate to put "military strikes on foreign data centers" into a public paper at all, and that the paper is getting serious pickup rather than being dismissed as fringe. Whether that reflects where Washington's China debate is actually heading, or simply one analyst testing how far the conversation can be pushed, is the more useful question than whether AGI itself is coming anytime soon.
