Since competition in the artificial intelligence industry, spearheaded by large language models, has entered a white-hot phase, and especially as the United States has grown increasingly anxious over the rise of China’s open-source large model cluster, bizarre claims about artificial intelligence originating from the U.S. have multiplied. For example, a U.S. think tank recently put forward a proposal with a straight face: to prevent China from catching up with or surpassing the United States in artificial intelligence, or from achieving AGI, the U.S. government should consider launching military strikes against Chinese data centers.
The Center for a New American Security (CNAS), a US think tank, recently released a highly provocative report, openly discussing “extreme measures” such as military strikes against China’s data centers to prevent China from being the first to achieve artificial general intelligence (AGI). This proposition was immediately criticized as “extremely reckless”: using hypothetical technologies that have not yet been implemented as a justification for taking action against China’s civilian infrastructure, pushing technological competition directly to the brink of a hot war.
According to a report by Hong Kong’s English-language media outlet South China Morning Post on September 4, Jacob Stokes, the deputy director of the Indo-Pacific Security Program at the center and a former national security staff member in the Obama administration, stated during an online event on Thursday that the U.S. Department of Defense, National Security Agency, and other agencies should begin assessing what kind of intelligence is needed to provide a basis for taking such actions.
Jacob Stokes
In his report, he listed a series of options, including “diplomatic, espionage, cyber, and kinetic measures”, among which were military strikes against data centers that train and operate China’s AGI systems. He even threatened to launch “preventive strikes” before China develops AGI.
While we have grown accustomed to this delusional anti-China performance, it is worth noting that U.S. anti-China think tanks are often funded by vested parties, such as the separatist authorities in Taiwan or certain military contractors. Strictly speaking, such reports carry little reference value; they should rather be understood as advertisements, designed to expand their own political clout.
Therefore, this plan to strike Chinese data centers is more than mere hype from U.S. anti-China think tanks. It should be interpreted as an attempt by U.S. AI giants to leverage the issue to gain greater political influence. Barring surprises, we will see more such reports in the future, and it would not be abnormal to encounter ones even more outlandish.
Firms such as OpenAI have already formed their own political action committees to advance AI-related policy agendas. In particular, Anthropic, a company known domestically for its combative rhetoric, constantly stirs up ideological disputes. It not only acts as a vanguard for anti-China messaging but has almost become the archetypal villain in a cyberpunk narrative. We may now readily draw a conclusion: in the United States of the future, this political force represented by artificial intelligence enterprises will exert tremendous influence over U.S. politics.
What we need to do now is to clarify the likely stances of future AI lobbying groups, a task of considerable importance for China. U.S. AI giants are of course the biggest competitors to China’s open-source AI cluster, and the proportion of anti-China voices among Silicon Valley right-wing circles is very high — otherwise they would not have concocted this shocking plan to strike China’s data centers. While we know the likelihood of this becoming reality is extremely low, it pays to be prepared.
Making Military Strikes Against China a Viable Option for U.S. AI Companies
U.S. artificial intelligence firms hold relatively consistent positions on China: nearly all support tightening export controls on China and cast doubts on China’s AI safety, differing only in degree. Take GPU exports as an example: apart from Jensen Huang, other U.S. AI industry players back stricter export restrictions targeting China. Beyond his commercial interests, Jensen Huang is more of a traditional tech entrepreneur than one of the new breed of AI magnates. These new AI tycoons are markedly more aggressive and exclusionary than traditional technology business leaders.
In the past, major U.S. tech giants, despite certain frictions with China, could often continue doing business, and these companies were adept at masking their hostility. AI firms are different. A string of statements by Anthropic on China are intended to openly display hostility and aggression, and the company even takes pride in this stance. Anthropic is fond of hyping claims that Chinese open-source models are distilling its models. Even Elon Musk criticized these assertions, yet Anthropic persists in repeating and smearing.
Thanks to Anthropic’s hype, “distillation” has evolved from a technical term in artificial intelligence into a political buzzword, and many people within China’s online space parrot this narrative. Such hype blatantly ignores fundamental principles of large models. First, distillation is not omnipotent and may even lead to overfitting. Second, distillation requires massive volumes of API calls and carefully crafted prompts; it cannot be achieved through casual conversations. Ordinary chat data simply cannot be used for distillation.
AI companies ought to understand this better than ordinary people. However, the so-called evidence previously released by Anthropic fails to explain this core issue. As Musk quipped, if we are talking about distillation, Anthropic itself would be guilty first. Furthermore, criticism of Silicon Valley AI firms represented by Anthropic has intensified across Europe and the United States, including accusations of copyright disregard, contributing to mental health crises, polluting the internet, not to mention competing with ordinary residents for water and electricity. Aware of their own moral vulnerabilities, these companies routinely deflect attention and seize the moral high ground by smearing and blaming China.
Readers following AI news will know Anthropic recently made another bizarre claim, aggressively alleging that Chinese open-source models route user queries to its own models. In other words, Anthropic claims Chinese open-source models do not process requests independently but forward them to Anthropic’s systems, which supposedly explains their strong performance. Once again, certain voices online in China eagerly parrot this claim.
This assertion is utterly baffling. What would be the purpose of such an operation? To acquire training data? If for pre-training, the volume of data cited by Anthropic would be grossly insufficient. If for post-training, though less data is required, the bar for data quality rises higher. Most data generated from everyday usage cannot serve post-training purposes; data for post-training must mostly be purpose-built, and ordinary data is often unusable.
If the goal were to advertise the prowess of its own models, Anthropic’s steep pricing undermines that logic. If this were truly happening, only a small number of users would gain access to superior models, yielding no public promotional benefit. What is more, these Chinese models are open-source by design. Once deployed within closed private networks, how could there be any “routing” to external systems? It is true that in the complex model marketplace of model aggregators, there are cases where vendors mix in expensive premium models to deceive paying customers. But is Anthropic unaware of the practices of model aggregators? It certainly knows, yet it chooses to attack and slander Chinese open-source models anyway.
The hostility these U.S. AI firms harbour toward China is more complex than that of U.S. politicians. Hostility stemming from market competition is easy to grasp, yet for firms like Anthropic, there is an abstract dimension to their animosity: they believe China holds a different understanding and worldview of artificial intelligence, and that a China with such a worldview ought not to possess advanced AI.
Within the worldview of today’s Silicon Valley tech right wing, artificial intelligence is treated as an omnipotent deity. In China, artificial intelligence is merely a powerful tool. This clash of worldviews is often no less severe than conflicts over commercial interests. After all, the current AI boom in Silicon Valley has taken on highly religious overtones. China may well dismantle this religious fantasy within a few years, an outcome intolerable to those Silicon Valley megalomaniacs who imagine they can govern humanity’s future.
For the time being, these AI firms can console themselves with the belief that China lags behind the U.S. in chips and computing power, yet these gaps will keep narrowing over time. Under such circumstances, firms like Anthropic have strong incentives to embrace more aggressive options. The proposition of launching military strikes against Chinese data centers may genuinely become a lobbying priority for these corporations. China should therefore prepare in advance. Judging by statements from Silicon Valley right-wing figures across many issues, their moral floor is extremely low.
In addition, Northeast Asia hosts the world’s densest concentration of advanced semiconductor supply chains. Driven by American-style persecution paranoia, these companies constantly fear that China will disrupt global semiconductor supplies. Before the AI boom, the United States might have had a chance to achieve self-sufficiency in advanced semiconductors, yet the massive demand unleashed by the current AI frenzy has rendered that aspiration increasingly unattainable.
We can therefore predict that these AI interest groups hold highly aggressive attitudes on issues surrounding the Taiwan Strait and the Korean Peninsula. They would likely lean toward risking conflict with China in order to “protect semiconductor supply chains.” This possibility must be taken into consideration.
The Silicon Valley right-wing military contractors involved in the wars in Ukraine and the Middle East are cut from the same cloth as today’s Silicon Valley artificial intelligence companies. We have already witnessed the poor ethical standards of Silicon Valley defense firms, and Silicon Valley AI companies will not prove more virtuous. Remember that the U.S. military is a major client of these AI companies. In pursuit of their own interests, they would likely not rule out military force as an option.
The Internal Rivalry Logic of U.S. AI Giants
Nonetheless, the aggression and exclusionism of these firms carry negative consequences. While they ostensibly reach consensus on many topics, their actual stances diverge dramatically, posing challenges to the formation of a unified AI interest bloc. Geopolitical hostility toward China may be one of the few propositions on which they can achieve full agreement, for almost none of these companies can view the development of Chinese artificial intelligence objectively.
Like traditional Silicon Valley tech firms, AI giants have inherited their consistent policy inclination: opposition to regulation, though with nuances. Hype around AI technology has stoked widespread public fear of artificial intelligence across American society, a major driver of the current anti-AI backlash. Hence we observe a paradoxical situation: these AI giants continue loudly promoting AI doomsday narratives (they will not abandon this fear-based messaging), oppose regulation, yet are simultaneously compelled to accept and voluntarily participate in regulatory frameworks.
This does not mean these companies intend to lose out to regulation. For one thing, U.S. government agencies currently suffer weak governance and legislative capacity; relevant measures cannot keep pace, so restrictions on AI giants will have limited effect. For another, AI safety is a convenient weapon to attack competitors. Every time OpenAI and Anthropic hype AI safety concerns, they are implicitly taking aim at one another. Using AI regulation to attack Chinese models also offers a more respectable approach, since directly demonizing China has become an inefficient tactic within today’s U.S. media landscape.
The same applies to attitudes toward open-source models. Recently, all major U.S. AI companies apart from Anthropic announced their support for open sourcing. This may look like a welcome shift, yet it is merely part of internal corporate infighting. Anthropic is genuinely reluctant to open source its models, while most other U.S. AI firms will at most release dessert-tier models. They will never open-source their core production-grade models, which have limited utility for real-world workflows, given that most commercial workloads cannot be handled by a small 20B parameter model.
These companies express support for open-source chiefly to claim the moral high ground and criticize Anthropic for its poor ethics. This also illustrates that Anthropic has become a target of widespread condemnation within the United States, standing out for its exceptionally low moral standards.
Driven by U.S. AI interest groups, we will most likely witness this scenario: voices advocating open-source and systematic AI regulation in U.S. public discourse will grow louder, yet concrete actions will fail to curb the expansion of AI corporations. Instead, regulation will become a weapon for companies to attack each other.
AI Interest Groups: America’s Largest Emerging Political Force
These firms share another common domestic objective: suppressing anti-AI voices within the United States. Opposition to artificial intelligence is growing louder in America and is nearly coalescing into a political movement. U.S. political analysts have even begun comparing the anti-AI movement to the historic Tea Party Movement.
Though the Tea Party Movement did not directly transform America’s political landscape, there would have been no Trump or MAGA without it. Barring unexpected turns, the anti-AI movement will exert long-term impacts on U.S. politics, a prospect AI giants strongly wish to avoid. This problem currently has no simple solution. Leaving other factors aside, rising water and electricity costs, noise pollution, and job displacement are tangible realities, so these corporations will push to crack down on the anti-AI movement.
Interestingly, both left and right populists in the United States oppose artificial intelligence. We will likely see AI giants align with the U.S. establishment to suppress these so-called populist forces. From another angle, we can see that the priorities of these AI giants diverge sharply from those of the U.S. establishment. The U.S. establishment fundamentally seeks to preserve America’s existing political institutions, whereas these AI giants genuinely see themselves as cyber-corporate behemoths.
While these companies do not harbour grand political ambitions at this stage, conflicts and evolving circumstances will inevitably push them to expand their political power, a process bound to disrupt America’s political balance. Although the U.S. establishment often compromises with large corporations, it will ultimately reject attempts by firms like Anthropic to cast themselves as administrators of the United States, or even the entire world. Viewed from this perspective, America’s radical right wing and AI giants hold greater potential for long-term cooperation.
The dilemma facing AI giants in choosing partners between the establishment and the radical right will be a major feature of U.S. politics for a long time to come. However, given inconsistent stances across different firms, these corporations may end up partnering with different political factions.
The analysis above follows the behavioural patterns of traditional corporate lobbying groups. Corporate titans primarily intervene in politics through publicity, sponsoring politicians, advertising, and influencing think tanks and opinion leaders. But this is not the only way to shape U.S. politics.
Another model of political influence relies on religious and grassroots organizations. Such groups may not command vast financial resources, yet they can directly mobilize large numbers of voters. Many pastors wield influence comparable to members of Congress. A similar phenomenon may emerge in artificial intelligence. Although most Americans currently dislike AI and data centres, a minority are utterly infatuated with artificial intelligence, viewing it as magic capable of solving all human problems. Artificial intelligence discourse in the United States has already become highly religious. Major corporations release pronouncements akin to divine edicts, which KOLs then interpret like preachers. This already constitutes the core communications workflow for AI today.
American electoral politics has repeatedly proven that organizational strength matters more than raw numbers. The fervour of AI supporters in the United States is remarkable. If harnessed properly, AI giants can leverage these enthusiasts to secure substantial political resources, introducing major uncertainties into U.S. politics.
We may now draw a conclusion: AI giants are among the few interest groups that possess two distinct means of interfering in U.S. politics. Coupled with America’s enormous current investment in artificial intelligence, this emerging interest group will expand far faster than competing lobbies.
We can readily predict that AI giants will gain growing sway in U.S. politics, and they will clearly refuse to abide by America’s established political rules. AI leaders likely sincerely believe artificial intelligence is humanity’s ladder to the next evolutionary stage. Once that step is taken, all prior rules and sacrifices cease to matter — naturally including America’s present political institutions. This mindset is profoundly reckless, yet a surprising number of people genuinely believe these doomsday prophecies.
Undoubtedly, AI giants holding such beliefs will, even acting with benign intentions, exert destabilizing effects on U.S. politics as they seek to shape policy. Their hostility toward China is only the comparatively straightforward dimension of this challenge. Therefore, while China needs to prepare for these developments, it is America’s own political system that faces a greater need for readiness.
Editor: zhangyixincq
