U.S. Senator Bernie Sanders introduced legislation this week requiring the federal government to acquire 50% equity stakes in the country's largest artificial intelligence companies through a newly created sovereign wealth fund. The proposal argues that because AI models train on public data—scraped web content, government datasets, user-generated contributions—the American public holds a legitimate claim to corporate profits derived from those inputs. Sanders did not specify which firms would face mandatory dilution, but the structure would apply to any AI company crossing an undefined revenue or valuation threshold.

The bill envisions a federally administered fund holding non-voting or voting shares, depending on final language, with dividend streams redirected to a public account. Sanders' office cited Norway's $1.7 trillion Government Pension Fund Global as a structural precedent, though that vehicle holds diversified global equities purchased at market rates, not confiscatory domestic stakes. The proposal does not address purchase price, whether existing shareholders would be compensated at fair value, or how valuation disputes would resolve. It also leaves open whether the fund would require Senate confirmation for board seats or exercise governance rights. The draft relies on broad assertions about data provenance rather than specific intellectual property or licensing frameworks, a gap that invites constitutional challenge under the Takings Clause.

The immediate market implication is not passage risk—the bill lacks co-sponsors and faces a Republican House—but the migration of nationalization rhetoric from fringe commentary to formal legislative text. That shift compresses the Overton window. Allocators pricing U.S. AI equities now must model a non-zero probability that future administrations, particularly in a recession or after a high-profile algorithmic failure, revive dilutive equity mandates under different branding. The Nordic comparison is rhetorical cover; Norway's fund does not expropriate domestic companies, and its $1.7 trillion in assets came from oil revenue invested abroad over five decades, not legislative seizure. A U.S. analog would require either printing equity stakes into existence—directly inflationary and balance-sheet destructive—or purchasing them with deficit financing, which adds fiscal pressure without the hydrocarbon windfall.

Second-order effects concentrate in venture formation and domicile decisions. If a 50% public-stake mandate enters serious debate, early-stage AI founders will incorporate outside U.S. jurisdiction, favoring Switzerland, Singapore, or UAE free zones where sovereign interference remains theoretical. That accelerates the brain drain already visible in regulatory arbitrage around model training and data residency. Later-stage firms with embedded U.S. operations face a more complex calculus: they cannot easily redomicile, but they can freeze domestic headcount growth and route new R&D through foreign subsidiaries. The result is a bifurcated landscape where the most valuable AI infrastructure—model weights, training clusters, alignment research—migrates offshore, leaving the U.S. with application-layer companies that lack the foundational IP needed to sustain long-term competitive advantage.

Allocators should track three near-term developments. First, whether any Democratic co-sponsors with Banking or Finance Committee seats attach their names, which would signal a genuine legislative vehicle rather than a messaging bill. Second, how OpenAI, Anthropic, Google DeepMind, and Microsoft frame their opposition—technical rebuttals around data ownership versus direct political lobbying—and whether they coordinate a unified response or fragment. Third, the reaction from public pension funds, which hold large-cap tech exposure and would face conflicting incentives if a sovereign wealth fund promised dividend redistribution but cratered underlying equity values.

The bill will not pass this Congress, but it establishes a legislative baseline that future populist coalitions can reference when AI profits concentrate further and unemployment narratives sharpen. The version that eventually surfaces, if any, will likely involve tax surcharges or royalty structures rather than outright equity seizure, but the framing—public data as public property—is now embedded in Senate record.