Jacob Stokes, a former Obama administration national security official and now a senior fellow with the Indo-Pacific Security Program at the Center for a New American Security, argued in an Aug. 27 report that the United States should be ready to use cyber operations, and potentially military strikes, to destroy data centers in China if Beijing reaches artificial general intelligence first.
The report, titled Superpowers and AGI, pushed the discussion well beyond export controls or computing restrictions. It asked whether cyberattacks or even missile strikes should be considered against an opponent’s computing infrastructure at a time when AGI itself remains scientifically unsettled.
Report lays out a four-stage AGI race
Stokes divides the United States and China into a potential “leading state” and “lagging state,” and describes four stages of an AGI competition: both countries are pursuing it, one country is close to a breakthrough, one country has it while the other does not, and both countries have it.
He wrote that once a country concludes it is behind, its choices could escalate step by step. Those choices begin with accelerating its own program, then accepting a lagging position while moving quickly to catch up, including through technology theft, and ultimately trying to disrupt the leading country’s AGI systems. According to the report, that disruption could involve physical infiltration or offensive cyber operations, explicitly including cyberattacks. As a historical comparison, Stokes pointed to the Stuxnet attack against Iranian nuclear facilities.
Military action was included as the highest-risk option
The report does not stop at cyber operations. Stokes wrote that the most dangerous option, and the one carrying the greatest risk of escalation, would be “kinetic attacks,” meaning missiles and bombs used to destroy targets.
The targets he identified were large-scale data centers that provide the computing power needed for AI training and operation. The report also cited a March 2026 Iranian strike on Amazon data centers in the United Arab Emirates and Bahrain, describing it as a precedent for the first kinetic attack on data centers during a conflict.
Stokes says AGI is harder to track than nuclear development
Even while drawing a comparison to the nuclear era, the report acknowledges that the analogy does not fully hold. The International Atomic Energy Agency can track uranium, plutonium, centrifuges and delivery systems, and nuclear tests are difficult to hide. AI inputs, by contrast, are described as ubiquitous, more civilian than military, and constantly in use. That makes it difficult to determine whether a rival is close to AGI, and impossible to calculate a “breakout time” in the same way policymakers once tried to do for nuclear programs.
The report also recognizes a deeper uncertainty: AGI has not yet been invented, and no one knows what key inputs it actually requires or where the threshold lies. Futurism pushed back on the premise as well, arguing that the United States, not China, is currently ahead on most large language model benchmarks, adding that if anyone is really chasing AGI, it may be the U.S. itself.
Stokes also warned that if a leading state were to obtain AGI first, it might use that position to lock a trailing state out. In that scenario, the lagging country could face pressure to use force rather than accept permanent disadvantage.
Report calls for U.S. scenario exercises
Stokes acknowledged that the United States previously considered bombing Soviet and Chinese nuclear programs on two occasions, but ultimately backed away because the risk of escalation into full-scale war was too high. He added that the U.S. did in fact invade Iraq and strike Iran in efforts tied to nonproliferation.
In the policy section, he argued that Washington should conduct a scenario exercise in advance so that the Department of Defense and national security agencies can assess what level of intelligence would be sufficient to justify extreme measures, including preventive strikes.
Backlash followed wider media attention
The report was published on Aug. 27, but broader attention came after the South China Morning Post covered it on Sept. 4. That was one day after OpenAI released GPT-6 Astra on Sept. 3 and President Greg Brockman said at a press conference, “Welcome to the age of AGI.”
Criticism followed quickly. William Hartung of the Project on Government Oversight told Investing.com that such a strike would be “extremely reckless” and would risk a war between two nuclear-armed states.
Former Global Times editor-in-chief Hu Xijin also responded on X on Sept. 4, writing: “If the United States really dares to launch military strikes on China’s data centers, I believe China’s retaliation would not be limited to U.S. military targets in the Asia-Pacific region, but would directly strike the U.S. mainland with missiles.”
Online reaction was even more blunt. Some users said Stokes fundamentally misunderstands what AGI is and called him a war hawk. Others asked how Washington would react if Beijing ran the same kind of war game and started calculating which American data centers should be attacked, especially if the U.S. is already ahead on current benchmarks.