Open Weight Models

Open-Source AI Becomes a Battlefield — With Antitrust Implications

A philosophical and economic war is breaking out inside the artificial intelligence industry with stakes that extend well beyond tech company market share. OpenAI's policy chief published arguments this week characterizing open-weight AI — models whose underlying weights are made publicly available for anyone to run or modify — as a path toward, in the direct phrasing used, 'communism.' The framing is that open-weight models undermine the ability of American companies to maintain competitive economic advantages over adversaries, particularly China. The immediate counterargument came from investor Chamath Palihapitiya, backed publicly by Jack Dorsey, who argued the opposite: that restricting open-source AI would create a massive cost asymmetry where American companies pay premium prices for proprietary access while foreign competitors run open-weight models for near-zero cost.

Both sides are using American competitiveness as their framing and reaching opposite conclusions — which points toward a deeper antitrust dimension. The Sherman Act of 1890 prohibits monopolization and agreements that restrain trade, but having large market share is not itself illegal under U.S. law; what matters is whether exclusionary conduct was used to achieve or maintain dominance. Palihapitiya's argument maps onto this framework: he is essentially contending that OpenAI lobbying for open-weight restrictions constitutes an attempt to use regulatory capture as an exclusionary tool — paying hardware stores, in effect, not to stock competing mousetraps. The EU is grappling with the same dynamic in its AI Act implementation, and Google has separately pushed back against EU Android AI rules that the bloc argues are needed to prevent the company from leveraging its Android dominance to foreclose AI competition.

Moonshot AI's Kimi K3 model illustrates the physical constraint underlying all of this competition: the Chinese lab had to halt new signups because demand exceeded its GPU compute capacity. Morgan Stanley projected that Micron could achieve gross margins approaching 90 percent on high-bandwidth memory by 2027, underpinned by 16 supply contracts with floor pricing — a position that reflects the acute supply-demand tightness in the chips that make AI run. AI chip startup Etched is reportedly seeking a 20-billion-dollar valuation for application-specific integrated circuits optimized for transformer-based AI inference, and Nouriel Roubini argued publicly that AI-driven labor displacement will eventually force governments toward universal basic income or what he called 'socialism' — a structural claim that Hyundai's manufacturing workers are already living in practice as four-hour daily strikes began Monday over the planned deployment of Boston Dynamics' Atlas humanoid robots.