Mark Cuban backs concerns that AI is diluting patent value, pointing to trade secrets instead

Microsoft CEO Satya Nadella has warned about “pay twice” dynamics in the age of artificial intelligence, a concern that intellectual property rights could be triggered both when content is used to train AI systems and again when outputs are monetized.

In a separate discussion carried by Yahoo Finance, Mark Cuban said that shift is showing up in the marketplace. He argued that the value of patents is declining as AI reduces the practical exclusivity that patents were intended to provide.

Cuban’s position, as summarized in the report, is that AI’s ability to ingest large amounts of data, including information tied to patents, makes it harder for patent holders to maintain the same kind of control they historically expected. The more widely patent-related information can be absorbed and used during model training, the less differentiated the patent may become in practice.

Rather than betting primarily on patents, Cuban suggested that companies may increasingly rely on trade secrets. Trade secrets are business-sensitive know-how, formulas, methods, or processes kept confidential to preserve competitive advantage rather than being publicly disclosed like patents.

The “pay twice” framing matters because it points to a broader economic tension in AI development. If IP owners feel they are being asked to license their rights at multiple stages, that can raise costs for model builders and, in turn, increase pressure on whether firms seek protection through patents, trade secrets, or other contractual arrangements.

For Microsoft, the issue intersects with its strategy across cloud computing and AI services. Microsoft sells enterprise software and platforms and also operates large-scale AI infrastructure through its Azure cloud, so questions about how IP can be used in training and how rights are cleared can shape both product design and legal risk management.

Even beyond Microsoft, the debate reflects a wider shift in how businesses think about IP in the AI era. Patents are meant to exchange disclosure for protection. Trade secrets, by contrast, try to avoid disclosure altogether. As AI systems become more data-driven, companies may adjust their IP “mix” to match how easily information can be replicated or learned from.

The report does not provide additional specifics on Nadella’s original “pay twice” claim beyond the general concept, and it does not lay out measurable evidence showing patent value has fallen. It also does not quantify how much of AI training data is derived from patented inventions or how courts and regulators are likely to treat these disputes in the near term.