Trump’s AI plan: Race against China that trips over itself

President Donald Trump’s “America’s AI Action Plan” is pitched as a decisive break from past US policy—a full-throttle push to outpace China in artificial intelligence. But as Foreign Policy points out in its recent article, when viewed alongside the administration’s broader trade, immigration, and regulatory agenda, the plan often works against itself. 

While it contains promising ideas, such as accelerating innovation, expanding infrastructure, and promoting open models, these measures are systematically undercut by tariffs, immigration restrictions, deregulation, and politicised oversight that risk stalling America’s AI ambitions before they can even take off.

At its core, the plan seeks to remove federal and state-level regulations that might limit AI development, with federal funding steered away from states with rules deemed too restrictive. It also envisions exporting a bundled “American AI stack,” combining hardware, models, cybersecurity, and applications from companies like Nvidia, AMD, Oracle, and OpenAI, backed by federal financing. The stated goal is to ensure that US children are not living under “algorithms of the adversaries,” namely China’s. 

Yet, as the analysis notes, American algorithms have already caused plenty of problems at home—from cyberbullying and misinformation to mental health crises—undermining the idea that AI is best understood as a geopolitical race.

Trump’s portrayal of regulation as the enemy of innovation sets up a false choice. Well-designed rules addressing bias, misinformation, privacy, and data standards can lower compliance costs, build public trust, and open new markets. In their absence, foreign buyers may hesitate to purchase AI systems they see as guardrails-free and dominated by a small group of US tech giants.

Beyond AI-specific policy, the administration’s broader agenda undermines its own stated goals. A chaotic tariff regime disproportionately hurts AI start-ups, which depend on imported specialised hardware and lack the reserves of large companies to absorb higher costs. A proposed 100 per cent tariff on semiconductors—without clarity on what qualifies as “US-built”—threatens to sharply raise GPU prices. 

Immigration restrictions choke the talent pipeline, despite the fact that 77 per cent of leading US AI firms were founded or co-founded by first- or second-generation immigrants. These moves undercut any claim to global leadership in AI.

The plan’s emphasis on expanding AI infrastructure in the US, a vital pillar for training advanced systems, also collides with other policy choices. If tariff exemptions on imports from Canada, Mexico, and China expire, costs for steel, aluminium, fiber-optic cables, and cooling systems will rise by 15 to 20 per cent, driving up data center construction costs and potentially pushing projects overseas.

The export strategy built around the American AI stack carries its own risks. While bundling hardware, models, and software could simplify deployment, it also limits customisation and locks buyers into a single supplier. Many customers prefer to mix and match components, integrate local data, and avoid dependence on one provider. 

Trust will be an even bigger hurdle: the administration plans to certify only “truth-seeking” and “ideologically neutral” AI, explicitly excluding what it calls “woke” models. This politicisation is likely to deter some foreign customers. Environmental deregulation could add another handicap, potentially excluding US AI providers from green-tech alliances and procurement in environmentally conscious markets.

China’s AI strategy provides a sharp contrast. Rather than chasing abstract goals like artificial general intelligence, Chinese companies focus on practical, commercially viable applications, which makes their products appealing to buyers with immediate needs. 

The launch of DeepSeek-R1 earlier this year showed that Chinese firms can innovate around US export controls through software efficiency and alternative modeling approaches. China’s dominance in rare-earth elements, essential for semiconductor production, gives it leverage to shape US export policy; it has already used that advantage to persuade the Trump administration to ease restrictions on Nvidia’s H20 AI chips.

The plan’s embrace of open AI models stands out as one of its most promising features, with the potential to democratise access, cut costs, and spur innovation. Yet, in a deregulated environment, open models heighten the risk of misuse and security breaches. Moreover, promoting open models while simultaneously pushing to lock in global buyers to a proprietary US stack sends mixed signals, diluting the credibility of both initiatives.

In the absence of coherent policy alignment, the burden will fall on AI developers, adopters, and industry groups to hedge against political and economic volatility. That means managing costs through more efficient computing methods, adopting voluntary safety frameworks to build trust, seeking talent globally even if it means expanding abroad, and working with international partners to shape common AI standards. 

In markets where convergence on standards is impossible, companies will need to prepare to comply with multiple regulatory systems, from US “bias-free” requirements to Europe’s risk-based tiers.

Open models can still be a powerful asset if accompanied by transparency, allowing for audits and regulatory compliance, especially in tightly regulated industries. But the opportunities they create will only be realised if organisations understand and address the associated risks, from data leakage to malicious use.

Trump’s AI Action Plan is billed as a bold new course to beat China. In reality, its mix of deregulation, protectionism, politicisation, and contradictory trade and immigration measures may hobble US efforts while giving Chinese competitors space to advance.

The plan’s best elements—open models, infrastructure investment, and national security prioritisation—require a coherent foundation of trade, talent, and trust-building strategies. Without that, as Foreign Policy concludes, the plan risks not only breaking from past policy, but breaking from itself.

By Sabina Mammadli