When AI Access Becomes a Weapon, Everyone Builds Their Own

Six governments on three continents, all now building AI infrastructure they control. India is manufacturing data-center components domestically and backing roughly 20 foundation models through a national AI mission ^[1]. The European Commission awarded a 180-million-euro sovereign cloud contract exclusively to European providers and is pushing to restrict foreign cloud companies from public tenders in defense, healthcare, and energy ^[2][3]. The UK launched a £500 million sovereign AI fund ^[4]. Japan's SoftBank, NEC, Sony, and Honda jointly founded a domestic AI model initiative with up to a trillion yen in government backing ^[5]. China's AI-chip self-sufficiency has climbed from 10 percent in 2021 to 41 percent this year ^[6]. Ukraine is building military-grade AI computing inside its own borders ^[7]. These governments have different reasons. The UK and Japan are close American allies pursuing competitive autonomy. Ukraine needs military computing it can trust in a war zone. The EU wants to shield public infrastructure from foreign law. But underneath all of them runs a shared signal, sent by Washington: AI access is conditional, and it can be revoked. The clearest case is India. A US government export-control directive forced Anthropic to cut off all foreign nationals from two of its models, disrupting major Indian firms including TCS and Infosys ^[8]. India's response was not to find another American vendor. NITI Aayog began reviewing the entire AI ecosystem; the IndiaAI Mission is expanding toward localized data centers and GPU clusters; and the startup Sarvam AI is building indigenous foundation models ^[8]. Sarvam's framing captures the logic the cutoff set in motion.

For AI users, it is clear that you should not confuse access with ownership, or adoption itself as an advantage. — Sarvam AI

The cutoff told Indian firms that renting access to someone else's model is not the same as owning the capability. And once that distinction becomes visible, it cannot be unseen. The two executives on opposite sides of the chip war agree on what the controls are producing. Jensen Huang, whose company sells the chips the controls target, argues they backfire.

China's AI moves on with or without U.S. chips. It has to compute to train and deploy advanced models. The question is not whether China will have AI, it already does. The question is whether one of the world's largest AI markets will run on American platforms. — Jensen Huang

Huawei's rotating chairman Xu Zhijun, whose company is the intended target, makes the same argument from the other side.

If the US hadn't forced our country, our companies, and our industry, we wouldn't have done something like this. — Xu Zhijun

When the seller and the buyer both tell you the same thing is happening, it is probably happening. China's domestic chip self-sufficiency has quadrupled in five years, and China has reportedly refused to buy licensed Nvidia H200 chips, preferring domestic alternatives ^[9]. Both sides are now choosing sovereignty over interdependence. The US is trying to play both sides of this. Even as it restricts model access through export controls, American tech giants have committed $80 billion to Indian AI infrastructure — Amazon alone plans $48 billion by 2030 ^[10]. The strategic logic seems to be: restrict the sensitive technology, flood the ally with capital. But the signal and the money work against each other. You can promise $80 billion in infrastructure and still have a government directive cut off your models overnight. India's sovereign AI push coexists with deepening US tech partnership; it does not depend on it. Not every sovereign-AI initiative traces to US export controls. The UK's tech secretary framed the £500 million fund as a matter of national security and economic prosperity, not fear of American restrictions ^[4]. Japan's push is about closing the gap with the US and China. Ukraine's is about surviving a war. And not every firm builds from scratch — Infosys, after the Anthropic cutoff, pivoted to OpenAI's Codex rather than going fully indigenous ^[11]. The sovereign AI movement has many fathers. But US weaponization is the accelerant. It takes governments that were already worried and gives them proof that the worry is justified. It turns a hypothetical risk — what if we lose access? — into a demonstrated one. And once access has been shown to be conditional, owning the stack becomes the rational hedge regardless of how much capital flows in alongside the restrictions. Nokia's CEO has warned that Europe's investment pace may not be enough to keep developers from migrating to where the capacity is ^[12]. That may be right, and the outcome may bifurcate between powers that can afford genuine sovereignty and those that cannot. But the political will to try is now locked in. The irony is not lost on Washington. Trump's military AI integration order mandates "sourcing from multiple vendors to avoid single points of failure" and creates an AI National Security Strategic Reserve of outside experts ^[13]. The US government is applying the same anti-dependency logic to its own AI procurement that its export controls have provoked everywhere else. It doesn't want to depend on a single AI firm. Neither, now, does anyone else. China has begun mirroring the approach — restricting overseas access to its own advanced AI models, blocking model-weight downloads, and designating AI theft as a national security offense ^[14]. Both great powers are now using AI model access as a strategic lever. That symmetry is what forces every third party to choose or hedge — and the hedging is exactly what the original controls were meant to prevent.