Public AI Infrastructure Model Gains Traction as Alternative

Public AI Infrastructure Model Gains Traction as Alternative

As debates intensify over AI regulation and ownership, a distinct approach is emerging that rejects both laissez-faire corporate control and government equity stakes in favor of treating AI as public infrastructure.

B Cavello, director of emerging technologies at the Aspen Institute, advocates for what she calls "public AI"—open-source systems centered on public access, accountability, and sustainable public goods. This framework differs sharply from proposals floated by OpenAI CEO Sam Altman and Sen. Bernie Sanders that would give Americans financial ownership stakes in major AI companies.

Why it matters

The public infrastructure model addresses a critical gap in current AI policy debates: how to ensure broad access and accountability without either accepting corporate dominance or relying solely on government ownership. With data centers straining electrical grids and AI capabilities concentrated in a handful of companies, the infrastructure framing offers policymakers a tested regulatory template.

The Infrastructure Analogy

Cavello draws parallels to electricity regulation, envisioning AI as "a reliable, publicly accountable utility that has universal service requirements to make sure that everyone can be connected." Under this model, a foundational infrastructure layer would be publicly regulated while allowing private innovation on top—similar to how the electrical grid supports diverse applications.

The approach prioritizes local representation and capability distribution. CurrentAI, a public-private partnership, recently launched its AIPotluck initiative to build a complete open-source AI stack accessible to the public. Cavello notes similar movements toward "AI sovereignty" across Africa and Europe seeking independence from American AI providers.

Practical Implementation

Cavello outlined several concrete steps toward public accountability:

• Requiring transparency in data center operations, challenging the assumption that tech companies can remain secretive
• Mandating that 20 percent of data center capacity be allocated to publicly accessible projects
• Strengthening federal agencies like the IRS to enforce accountability
• Expanding public library access to AI tools as an alternative to corporate platforms

She contrasts this with Sanders' partial ownership proposal, which she views as "accepting defeat on other forms of regulation" by making the public "just passengers on this ride."

Path to Policy

Cavello compares the challenge to the renewable energy transition, which required both technological maturity and policy incentives like rebates. She expects local-level demonstrations to precede federal action, given that Congress typically operates as a reactive institution.

The Windfall Trust network is working to secure pre-commitments on policy triggers before crises like mass unemployment occur, when political pressure might force hasty decisions.

Cavello acknowledges the utility model isn't perfect—current fights over data centers straining electrical grids illustrate existing infrastructure governance problems. But she argues the framework provides a starting point for ensuring AI serves public interests rather than solely corporate ones.

These details were first reported by Mother Jones.