Tech CEOs warn enterprises are handing AI labs their crown jewels

Microsoft CEO Satya Nadella published a blog post this week arguing that companies using AI models from OpenAI and Anthropic are effectively "paying twice" for their services. The first payment is the token cost—the direct fee for using the model. The second, invisible payment is proprietary knowledge: the prompts employees write, the corrections they make when the model is wrong, and the tools agents use on their behalf. "Models learn from 'exhaust,' the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong," Nadella wrote. "Every correction is distilled into institutional know-how." He argues this creates a dangerous asymmetry: enterprises hand over the data that trains the models and improves them, while the AI labs learn to compete with those very enterprises. Meanwhile, the labs train freely on public internet data but restrict enterprises from doing the same with their own models.

Nadella's proposed fix is for companies to build "proprietary learning environments" in the cloud and deploy "orchestration layers" that allow them to switch between AI vendors without lock-in. (It is worth noting that Microsoft sells exactly this type of platform.)

He is not alone. On CNBC earlier this month, Alex Karp, CEO of Palantir Technologies, made a similar case, arguing that enterprise clients are privately "livid" about AI vendors who extract proprietary data and competitive edge while charging premium token prices for tools that often fail to deliver commensurate value. Karp accused frontier AI labs of prioritizing "tokenmaxxing" over solving real business problems. Palantir responded with a nine-point "AI sovereignty" manifesto declaring that "data retention is your treasure." Former White House AI and Crypto Czar David Sacks called Karp "exactly right" on his All In podcast, arguing that OpenAI and Anthropic have formed a duopoly leaving enterprises with insufficient leverage over their own data and infrastructure.

The backlash is already shifting behavior. Companies are increasingly turning to open-weight models—systems whose underlying parameters are published publicly, allowing anyone to download, run, and modify them. These models offer enterprises more control and transparency, often at a fraction of the cost of frontier models. According to Vercel, open-weight models now account for 29% of traffic through its AI gateway. Amazon's CTO Werner Vogels recently told the publication that he was seeing companies shift from frontier model providers to open-source options both to control costs and gain transparency over the technology they embed in their operations. The U.S. government's recent decision to block access to Anthropic's Fable 5 models has accelerated this trend, as companies worried about workload continuity have begun diversifying away from single vendors.

The critique from Nadella, Karp, and others represents a significant shift in how established enterprise leaders are framing the relationship between Big Tech and frontier AI labs. While both Nadella (whose company has invested billions in OpenAI and Anthropic) and Karp (whose firm sells data-protection software) have obvious business incentives to push this narrative, the underlying concern taps into a genuine tension: enterprises are handing over sensitive data to train models they do not control, and those same labs are simultaneously improving their ability to compete in the enterprise software market.

The emergence of open-weight models as a credible alternative—particularly from Chinese labs that the article notes are now leading the open-weight space—has created a real option for companies wanting to reduce dependence on any single vendor. The U.S. government's block on Anthropic's Fable 5 models has reinforced this incentive: companies relying on critical AI workloads can no longer assume a vendor's models will remain available. The shift toward open models is already measurable (29% of Vercel's AI gateway traffic) and aligns with the broader push from figures like Karp and Vogels for data sovereignty and reduced vendor lock-in.

No comments yet. Be the first to share your thoughts!