Can $500 Billion Win the AI Race? | Anton Leicht

Strategic Dilemma for Middle Powers:

• The default outcome for nations not building frontier AI is capturing all societal risks while minimizing AI benefits.
• Middle powers face a "second best" equilibrium of becoming useless allies or shutting markets, rather than integrating into a broad alliance supply chain.
• The optimal strategy involves middle powers occupying specific bottlenecks (e.g., robotics, manufacturing, legacy industries) and integrating them with US AI labs.

The "Compute for Access" Strategy:

• Middle powers should build data centers to alleviate US inference crunch in exchange for guaranteed access parity with the US commercial market.
• This model relies on private US labs (e.g., OpenAI, Anthropic) rather than US government treaties to avoid political volatility and reneging.
• Successful deals require offering 1 gigawatt of compute capacity within 18–24 months.
• Existing deals (e.g., UAE, Norway) have not yet fully secured access parity; Norway's "Stargate" deal collapsed partly due to OpenAI reneging on terms.

Sovereignty vs. Integration Debate:

• Building a sovereign frontier model requires a coalition investment of approximately $500 billion over 4–5 years.
• A sovereign project would require a private vehicle structure with direct government funding and a "czar" layer for political-technical translation.
• Participants should include the EU, UK, Canada, Australia, New Zealand, Japan, and South Korea, potentially excluding the US to mitigate retaliation.
• Critical barriers include securing access to US chips (requiring leverage via semiconductor supply chain bottlenecks like ASML) and US coding agents.
• The "fast follower" open-source strategy (lagging by 4 months) is deemed unsustainable; the US could restrict access to the frontier, widening the gap and allowing US labs to outpace followers via recursive self-improvement.

Economic and Labor Market Implications:

• Relative disempowerment is a greater threat than absolute economic decline; a US lead compounds over time via better tools for R&D and model improvement.
• Middle powers must protect "AI-compatible" industrial assets (e.g., robotics factories, data-generating manufacturing) from foreign acquisition by US firms (e.g., Bezos's Prometheus fund).
• Labor markets must become more flexible to allow workers to move along the "jagged frontier" of AI capabilities; protectionism risks long-term firm failure and mass layoffs.
• To offset displacement, governments should implement wage guarantees or junior job subsidies funded by increased corporate income taxes rather than specific "token taxes" on AI usage.

US Government and Policy Risks:

• US "soft nationalization" is expected to occur through informal pressure, embedded observers, and threats of export controls rather than formal legislation.
• A "pause" on AI development is politically unlikely to succeed due to fragmented coalitions, the difficulty of verifying Chinese compliance, and the asymmetry of US geopolitical leverage.
• A "decisive strategic advantage" for a single nation is less desirable than a shared Western alliance advantage constrained by checks and balances.
• Involving middle powers in AI governance creates a more stable world, reducing the density of "stupid decisions" caused by concentrated power in a volatile single actor.

Forward-Looking Statements and Urgency:

• The window to secure "compute for access" deals is closing rapidly; delays could increase the cost of a sovereign project from $500 billion to $1 trillion within a year.
• Political awareness among policymakers is lagging the necessary moves by months, increasing the risk of a suboptimal equilibrium.
• Without a coordinated allied approach, the world risks a chaotic future defined by AI misuse, mass migration, and conflicts exacerbated by unequal AI access.
• The "outside game" of policy analysis remains valuable despite increasing information silos within labs and governments.