Satya Nadella warns enterprises pay for AI twice over
Microsoft's boss says firms using OpenAI or Anthropic models pay twice, in cash and in leaked know-how. For India's IT majors and sovereign-model hopefuls, the warning cuts deep.
The News
Microsoft chief executive Satya Nadella has told enterprises that every time they lean on a proprietary AI model, they quietly hand over their most valuable asset. In a blog post published on Monday, 13 July 2026, Nadella argued that firms buying intelligence from providers such as OpenAI and Anthropic pay a second, hidden bill on top of their token invoices.
"You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal," Nadella wrote. His point is that a model does not just answer a query. It absorbs the prompts a company writes, the tools its agents reach for, and above all the corrections staff make when the model gets something wrong.
Nadella calls this residue "exhaust", and warns it accumulates into institutional expertise a rival could never buy on the open market. "In consuming intelligence, you are creating intelligence. And what you create should belong to you," he wrote.
Why It Matters
The striking part is who is making the argument. Microsoft is OpenAI's largest backer, the two firms joined at the hip since 2019. For Nadella to caution buyers about the data-leakage risk of frontier models signals that the platform layer, not the model itself, is where he now believes durable value sits.
His prescription follows that logic. He wants companies to keep ownership of their data, prompts and feedback, to build private learning environments on cloud infrastructure, and to bolt on an orchestration layer that lets them swap one provider for another with minimal friction. He also jabbed at the industry, saying providers cannot credibly forbid distillation of their models while themselves training freely on the public internet.
This is no longer a fringe worry. Nadella joins investors such as Jason Calacanis and Palantir chief executive Alex Karp, who have raised versions of the same concern about knowledge draining into shared models. When the last big platform debate landed, over public cloud lock-in a decade ago, the winners were the firms that designed for portability early. The same instinct is resurfacing for AI.
Indian Angle
For India's technology services giants, this is close to an existential design question. Firms such as TCS, Infosys, Wipro and HCLTech wire client workflows on top of OpenAI and Anthropic models at scale, often fine-tuning on a customer's proprietary processes. If Nadella is right that the "exhaust" carries competitive value, Indian integrators need airtight guarantees that a bank's or insurer's data does not seep into a shared system.
It also sharpens the case for India's home-grown model builders. Sarvam and Ola-backed Krutrim have pitched sovereign models partly on data control, and Nadella's argument is, inadvertently, a marketing brief for them. An enterprise wary of feeding a US provider its trade knowledge has a cleaner story if the weights sit on domestic infrastructure.
Regulation adds a third layer. Under the Digital Personal Data Protection Act and MeitY's evolving guidance, Indian firms already face pressure on where customer data lives. Nadella's framing hands compliance teams a commercial argument to match the legal one: keeping prompts and corrections in-house is not just privacy hygiene, it is retention of hard-won intellectual property. For rupee-conscious buyers already paying for tokens in dollars, the idea of paying "twice" will land hard.
FAQ
What exactly did Nadella warn about?
He warned that enterprises using proprietary AI models surrender business knowledge alongside their money. Models learn from prompts, tool use and corrections, accumulating expertise rivals could indirectly benefit from through the same shared system.
Why is this surprising coming from Microsoft?
Microsoft is OpenAI's biggest investor, so a warning about proprietary-model risk reads as counter-intuitive. It signals Nadella sees lasting value in the orchestration and cloud layer rather than any single model.
What does he recommend companies do?
Retain ownership of data, prompts and feedback, build private learning environments on cloud infrastructure, and add an orchestration layer so providers can be switched easily instead of locking in to one vendor.
What does this mean for Indian IT firms?
Services majors like TCS and Infosys, which build on third-party models for clients, will need stronger data-isolation guarantees. It also strengthens the pitch for sovereign Indian models from Sarvam and Krutrim.
Where can I read the original?
The full post and analysis were covered by TechCrunch, linked in the attribution paragraph below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.
Sources & Citations
Frequently Asked Questions
What exactly did Nadella warn about?
He warned that enterprises using proprietary AI models surrender business knowledge alongside their money. Models learn from prompts, tool use and corrections, accumulating expertise rivals could indirectly benefit from through the same shared system.
Why is this surprising coming from Microsoft?
Microsoft is OpenAI's biggest investor, so a warning about proprietary-model risk reads as counter-intuitive. It signals Nadella sees lasting value in the orchestration and cloud layer rather than any single model.
What does he recommend companies do?
Retain ownership of data, prompts and feedback, build private learning environments on cloud infrastructure, and add an orchestration layer so providers can be switched easily instead of locking in to one vendor.
What does this mean for Indian IT firms?
Services majors like TCS and Infosys, which build on third-party models for clients, will need stronger data-isolation guarantees. It also strengthens the pitch for sovereign Indian models from Sarvam and Krutrim.