Social Capital founder and venture capitalist Chamath Palihapitiya on Sunday backed Microsoft Corp. (NASDAQ:MSFT) CEO Satya Nadella’s warning that companies relying on closed artificial-intelligence models may effectively “pay for intelligence twice” by spending money while also exposing valuable institutional know-how.
Nadella Warns Of A Second AI Cost
“He’s right,” Palihapitiya wrote on X, responding to Social Capital’s summary of Nadella’s “Reverse Information Paradox.”
Nadella argues that enterprises pay once through token or subscription fees and again through the knowledge they reveal while making AI useful. “You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful,” he wrote in July.
That knowledge can appear in prompts, agent tool calls, evaluations and human corrections. Nadella said every correction can become “institutional know-how,” creating what Social Capital describes as a learning loop where repeated interactions become increasingly valuable. Palihapitiya has separately argued that cheaper open-weight models are rapidly closing the capability gap, raising questions over how much enterprises should pay closed-model providers.
Social Capital’s September research found leading open-weight models were roughly four months behind the best publicly evaluated closed frontier systems, while offering businesses more control over data, infrastructure and customization.
Enterprise Data Policies Complicate The Claim
The “pay twice” argument does not mean every enterprise prompt automatically trains a provider’s foundation model. Microsoft says prompts, responses and organizational data in Microsoft 365 Copilot are not used to train foundation models.
OpenAI similarly says ChatGPT Business, Enterprise and API data are excluded from training by default unless customers explicitly opt in. Reuters, in a report last week, said that OpenAI reiterated business data is not used for training without consent.
Data Sharing Could Reverse AI Economics
The economics can even reverse. OpenAI, according to its website, offers complimentary API tokens to customers that voluntarily share eligible inputs and outputs for model improvement, including up to one million daily tokens on certain larger models.
Palihapitiya has also earlier argued that open-source AI could give companies more control and lower costs.
Photo courtesy: Shutterstock
Login to comment