The Open Global Investment (OGI) model is a proposal by Nick Bostrom, set out in a 2025 essay for Digitalist Papers Vol. 2 (Vol. 2 essay), for governing the development of transformative AI. It envisions one or more lead ventures that are widely open to international shareholding and operate within a government-defined regulatory framework. Bostrom argues that the current status quo already approximates this model and that policy should refine it incrementally rather than move toward more exotic alternatives such as a Manhattan Project, a CERN-style consortium, an Intelsat-style public-private body, nonprofit stewardship, or full nationalization.
Definition
Bostrom characterizes OGI as one or more lead ventures with three properties:
- Widely open to international investment — shareholding is globally accessible.
- Operating under a government-defined regulatory framework — sovereign oversight, with enhanced government assurances and assistance.
- Structured to survive intelligence-explosion-scale strain — not only optimal in normal operation but robust to rapid capability change.
Variants
| Variant | Features |
|---|---|
| OGI-1 | Single lead AGI company |
| OGI-N | Multiple AGI companies |
| US-OGI | US-domiciled ventures |
| Geographically-neutral OGI | Any technically-capable host nation |
Relation to the status quo
Bostrom argues that the prevailing arrangement already roughly approximates OGI: Anthropic, OpenAI, Google DeepMind, and xAI are publicly-traded or privately-held ventures with significant international investment, operating under a US federal regulatory framework. His recommendation is to refine this arrangement toward the OGI ideal rather than move away from it toward alternative structures.
Comparison to alternatives
Bostrom assesses OGI against several alternative governance blueprints, weighing the advantages and drawbacks he attributes to each.
A Manhattan Project for AI offers speed, security, and national alignment, but in Bostrom's account monopolizes AGI in a single state's hands, prompts rivals to respond with their own crash programs, intensifies rather than eases race dynamics, and strips legitimacy through nationalization. This option has been explicitly proposed by Hendrycks, Schmidt, and Wang and some US policy voices.
A CERN for AI would provide international cooperation, scientific legitimacy, and neutral ground, but Bostrom argues it would be slow and consensus-bound, unable to move at frontier speed, and ill-fitted to labs structured as commercial ventures rather than science projects.
An Intelsat for AI offers a public-private partnership model with sovereign participation, but carries regulatory-capture risks; Bostrom notes that Intelsat is mostly privatized today and that the model fits infrastructure better than research and development.
A nonprofit or public benefit corporation offers mission alignment and governance flexibility, but faces capital constraints at frontier scale, the risk of board drift (Bostrom cites the OpenAI events of 2023), and unclear sovereign accountability.
Full nationalization offers democratic accountability and public-benefit alignment, but Bostrom argues it eliminates international shareholding, thereby reducing legitimacy and the available capital pool, and worsens race dynamics relative to a Manhattan Project.
Criteria for governance
Bostrom holds that any AGI governance structure is judged against many simultaneous desiderata: democratic oversight, national security, international cooperation, legitimacy, economic efficiency and dynamism, equitable benefit sharing, AI safety, free enterprise, wise stewardship, scientific advancement, operational security, and responsible deployment. He argues that no structure optimizes all of these, that any design will be subject to criticism, and that the choice is between imperfect options.
Robustness constraint
Bostrom places particular weight on the requirement that governance structures maintain integrity through intelligence-explosion-scale change rather than being merely satisfactory under stable conditions. He frames this through three tests: speed-matching, whether governance can respond at the pace of technical development; incentive compatibility, whether key actors such as labs, investors, and sovereigns will continue to support the structure; and adoptability, whether it can be agreed to in the first place.
Tensions with alternative proposals
Bostrom and adjacent proposals locate OGI relative to other governance frameworks. Compute-governance-centered approaches, including Compute Governance and BIS export controls, emphasize sovereign control of chips as the governance lever; Bostrom presents OGI as complementing rather than replacing them. Containment is a ten-layer technical-and-institutional framework while OGI is a structural corporate-governance one, and the two operate at different levels of the stack.
OGI also intersects with proposals for US-China coordination. Graylin's Beyond Rivalry essay assumes bilateral US-China coordination, and OGI-N with shared international investment is presented as a possible mechanism for it, with Chinese entities holding shares in US-domiciled AGI firms. The game-theory essay by Abraham, Kavner, Moon, and Matheny notes that race dynamics depend on the game actors believe they are playing; on this reading OGI changes the game, because if AGI firms are globally owned the notion of "winning" loses meaning.
Relation to policy as of 2026
Bostrom's argument can be measured against several features of the 2026 landscape. On ownership, the major labs have internationally accessible investment: Anthropic counts Google, Amazon, and sovereign-wealth-fund investors, and OpenAI's backers are led by Microsoft alongside others. A US regulatory framework is in place through executive orders (EOs), state laws, and anticipated federal legislation, and public-benefit components exist in forms such as Anthropic's PBC structure and OpenAI's capped-profit arrangement.
Two elements of the OGI ideal remain less developed. The "government-defined framework" with enhanced assurances that Bostrom describes is underdeveloped relative to his proposal. His call to resist proposals that move away from the model also runs against the direction of current US policy, which has drifted toward more sovereignty-centered framings in documents such as America's AI Action Plan and EO 14320 on AI exports.
Relationships
- supports: The Digitalist Papers (Stanford, Volumes 1–2) — Bostrom Vol. 2 essay.
- supports: Superintelligence: Paths, Dangers, Strategies — Bostrom (2014) — book-length framing.
- related: Containment (Suleyman) / Digital Geneva Convention / AI Tech Accords — complementary governance layers.
- related: Compute Governance / BIS Framework for AI Diffusion — Interim Final Rule (Jan 13, 2025) — adjacent policy levers.
- contrasts-with: Superintelligence Strategy (Hendrycks, Schmidt, Wang) — Manhattan-Project-for-AI alternative.
- contrasts-with: America's AI Action Plan / Executive Order 14320 — Promoting the Export of the American AI Technology Stack — US sovereignty-centered drift.
- related: Nick Bostrom — entity page for the author.