The model in one paragraph
In the OGI model, AGI development is led by one or more corporations that are widely open to international investment and operate inside a government-defined regulatory framework. Anyone in the world — individuals, sovereign wealth funds, even the governments of geopolitical rivals — can buy shares, and with them both profit participation and a measure of voting power. The host government (Bostrom's running example is the U.S.) regulates safety and security, while credibly committing not to expropriate or nationalize the company. Today's status quo, with publicly traded Alphabet, Meta, Nvidia, TSMC and institutional stakes in the private labs, already roughly approximates a weak version of the model.
Variants and mechanisms
The model comes in flavours: a single lead company (OGI‑1) or several (OGI‑N), domiciled in the U.S. (US‑OGI) or elsewhere. Bostrom sketches concrete machinery for a full-fledged version:
- Multiple share classes that partially separate profit participation from voting rights — e.g. Class A open to everyone worldwide, higher-voting classes restricted to citizens of countries that join a responsible-AI framework.
- Foreign participation, including by rival governments, encouraged up to caps (say 20% per country).
- Government assistance to the lead project — subsidies, regulatory waivers, antitrust forbearance for mergers — in the OGI‑1 version.
- Expropriation resistance: legal structures, treaty commitments, geographically dispersed data centres, and even multi-country cryptographic kill-switch schemes that let coalitions of investor nations halt the data centres if their contractual rights are violated.
Why Bostrom finds it attractive
Three motivations run through the paper. First, OGI leverages the world's best-established institutions — property rights, corporate law, capital markets — rather than inventing a novel governance scheme under time pressure. Second, wide ownership is incentive-compatible: powerful actors with a stake in the company have reasons to defend its property rights rather than expropriate it, and rivals offered a share of the upside have (at the margin) less reason to race desperately or sabotage. Third, it is implementable now, without treaties or new international organisations — which matters if timelines are short.
On fairness, the paper's most striking numbers: a U.S.-nationalized AGI project would exclude roughly 95.8% of the world's population from stake and influence, yielding a distribution with a Gini coefficient around 0.96 — more unequal than today's already very unequal global stock ownership (≈0.90–0.92). An open investment structure, especially with taxation of AGI profits in many shareholder countries, does better on every distributional measure Bostrom considers.
Against the alternatives
The paper compares OGI to a "Manhattan Project for AGI" (exclusionary, secrecy-bound, historically subject to almost no democratic oversight), to idealized international projects like a "CERN for AI" (poor incentive-compatibility for the actors who would need to fund and join them, slow to negotiate, operationally untested), and to closely held private companies (even more concentrated than nationalization in some respects). OGI is offered not as first-best idealism but as a robust second-best: "The choice will be one between imperfect options."
How this connects to Diversifying AI Ownership
The two papers approach the same goal from opposite ends. Bostrom's OGI describes the destination — an ownership-and-governance structure for AGI development that is open, diversified, and expropriation-resistant. Diversifying AI Ownership supplies the policy instruments that move the world in that direction: sovereign wealth funds and central banks passively holding the global market portfolio, pension reform that gets ordinary households out of concentrated home-country exposure and into world equities, and foreign-investment and tax reforms that lower the barriers to cross-border ownership.
The connection is explicit: Bostrom cites the paper (as Svarc & Hillebrandt 2025) alongside Barnard & Hillebrandt (2025) on debt-financed public investment in the AI sector, and notes that OGI "could also facilitate the implementation" of sovereign-wealth-fund proposals for transformative AI. If OGI is the governance ideal type, broad-based diversification policy is how you buy the world a seat at the table.
Read the full paper: Open Global Investment as a Governance Model for AGI (PDF) ↗