By default, countries that do not develop their own frontier AI models will experience all the negative societal disruptions, such as job displacement, while capturing minimal economic or strategic benefits. They get the risks without the rewards, the opposite of the US and China.
Instead of competing on model development, middle powers can secure a vital role by dominating physical bottlenecks in the AI supply chain, such as advanced manufacturing, robotics, or pharmaceutical production. This creates a mutual dependency with AI leaders like the US, ensuring their participation in the future economy.
A key strategy for middle powers is to offer fast, efficient data center construction to leading US AI labs. In return for alleviating the labs' 'inference crunch', these nations can negotiate guaranteed access to new frontier models at the same level as the US commercial market, ensuring they aren't left behind.
Direct deals with the US government are susceptible to volatility and linkage with unrelated political issues. It is more effective to partner directly with AI labs, whose acute need for compute incentivizes them to become powerful lobbyists for their international partners within Washington D.C.
Proposing an open-source model that quickly follows the US frontier is a flawed strategy. It antagonizes the US on two fronts: it threatens national security by promising to release dangerous capabilities to the world within months, and it commercially undermines trillion-dollar US companies by open-sourcing their technology.
Building a competitive frontier AI model is not a solo endeavor for a middle power. It would require a coalition of allied nations (like the G7) to commit roughly $500 billion over five years—a highly speculative and politically challenging investment that no single nation's treasury would likely approve.
To justify the immense public cost of a sovereign AI project, it must be framed as a critical strategic capability, analogous to an aircraft carrier. It is a national security asset you must possess, not a commercial enterprise expected to generate a financial return on investment.
Beyond restricting chip exports, the US holds a second, less obvious lever of control: access to its superior coding agents (e.g., Anthropic's Claude Code, OpenAI's Codex). Without these tools, any foreign attempt to build a frontier model is significantly slower and less competitive from the start.
To counter US threats of restricting chip access, an allied coalition can leverage its own critical positions in the upstream semiconductor supply chain. By controlling assets like ASML's manufacturing equipment or Japanese/Korean high-bandwidth memory, they can create the conditions for a mutually beneficial deal on chip allocation.
Middle powers are currently undervaluing their legacy industrial assets (e.g., factories, manufacturing plants). This "awareness gap" creates a window for savvy foreign investors to acquire these future AI-economy bottlenecks for cheap, before their true strategic value is widely recognized.
The correct response to AI-driven job displacement is counterintuitive: make labor markets more flexible. This allows workers to quickly reallocate to tasks where humans still hold a comparative advantage. Protecting old jobs with rigid regulations only makes firms uncompetitive, leading to worse economic outcomes.
The 'nationalization' of US AI labs will not be a formal government takeover. Instead, it will manifest as a continuous, soft back-and-forth where the administration uses veiled threats and its wide range of regulatory powers to informally pressure labs into aligning with its strategic goals.
A movement to pause AI development relies on a broad but fragile coalition of groups with different motivations. This coalition might succeed in passing a simple domestic pause but is likely to fracture immediately after, leaving no political capital for the complex and essential international treaties needed to make a pause effective and safe.