AI Diplomacy: Whoever Writes the Rules Helps Shape the Game

In 2024, American institutions produced 40 notable AI models, China 15, and Europe combined 3, according to the AI Index of Stanford University (2025). But the same index records that the performance gap narrowed within a single year: the lead of American models over Chinese ones on the MMLU test was 17.5 points at the end of 2023, and 0.3 points at the end of 2024. This is what makes talk of "AI diplomacy" intelligible: when capabilities converge, competition intensifies over everything else.

Hard power, soft power, and a third?

Joseph Nye, the Harvard professor who died in 2025, defined soft power as the ability to get others to want what you want, through attraction rather than force or payment. Artificial intelligence draws on both kinds: hardware, chips and computing that can be withheld or granted, and standards, language and values that travel with every model in use. We might call this mixture "technology power", though that is our own interpretive label, not an established term.

The standard as power

In 2012 Anu Bradford showed, in what she called the "Brussels Effect", that the European Union spreads its influence through its legal institutions and standards, without needing international institutions or the cooperation of other states, so that legal frameworks in many markets are shaped by them. The European AI Act sorts systems into four risk levels. Will that effect recur in artificial intelligence? An open question, and it is likely to depend on the size of the market and on what other countries decide.

Platforms are multiplying. In August 2025 the United Nations established a scientific panel of 40 members and an annual forum for dialogue on AI governance. In February 2026, 88 countries and organisations endorsed the "New Delhi Declaration" at the AI Impact Summit in India, a non-binding declaration whose signatories include the United States and China. Even so, press coverage notes that the American delegation warned against centralised global control of generative AI, and that differences over global governance persist. Cooperation and competition proceed together.

And middle powers?

Having an influential presence does not require a state to excel at everything. It is enough to choose a place it masters in the value chain, and to diversify partners so as not to depend on a single source. In Saudi Arabia, the company HUMAIN offers an example of this direction: launched by the Public Investment Fund in May 2025 to build the links of the whole stack, from data centres and cloud to models and applications. The Fund's page says it develops ALLaM, an Arabic language model, in cooperation with the Saudi Data and Artificial Intelligence Authority (SDAIA), and that its partners include NVIDIA, Microsoft, AMD, Qualcomm, Amazon Web Services and Google Cloud. This is a description from the owning body, and we do not yet have an independent assessment of its results. Still, choosing the Arabic language as a field of specialisation, with diverse partners, accords with what is generally advised for middle powers. For more on the Arabic model, see the article "Arabic Artificial Intelligence: Do We Make or Consume? Sovereignty and Arabic Language Models".

What are decision-makers advised to do?

• Build skills before hardware; tools are bought, but the knowledge that runs them is built.
• Reserve a seat in standards bodies and international forums, for rules are drafted where people gather.
• Diversify partners between East and West in a way that preserves room for decision.
• Invest in a distinctive advantage, such as the Arabic language and authentic Islamic and Arabic content, rather than crowding with everyone on every front.

The aim in the end is not influence for its own sake but that artificial intelligence serve people, that the nation keep its place as a partner rather than a follower, and that the work rest on trustworthiness and service of the public good.