Tech leaders back open AI, but deeper openness needed—data access key

Jensen Huang's Friday letter represents a major shift: the Trump administration's protectionist stance against open-source software unexpectedly unified nearly the entire tech industry (with Anthropic as the sole holdout) in support of open AI—a reversal that signals both geopolitical anxiety about Chinese AI competitiveness and recognition that openness itself drives innovation. The author endorses this momentum but argues it is incomplete. The letter addresses open weights and open source, but sidesteps the data problem: proprietary AI labs have trained models on vast quantities of copyrighted material—books, academic papers, code repositories, videos—often without explicit permission, citing transformative fair use as legal cover. Open-source developers, by contrast, lack comparable access to these datasets, creating an asymmetry that prevents fair competition. The author frames this as a moral and strategic question: if closed models can legally use pirated and copyrighted material for training, why should open models not have the same fair-use right, especially in service of national technological sovereignty against China? The author acknowledges three hard challenges—recruiting talent away from private AI companies, securing compute infrastructure through industrial collaboration (citing Chinese examples like DeepSeek and Moonshot that trained models for under ten million dollars), and overhauling copyright law—but argues all are surmountable given industry consensus and the precedent of open-source innovation in earlier decades.