AI models have clearly transformed the process of writing software. Nevertheless, jobs in the U.S. software industry have risen over the past year, according to the Bureau of Labor Statistics, while the 12-month moving average of workers in computer and mathematical occupations hit an all-time high in July 2026. Looking at a broader measure, the unemployment rate for young people (aged 20 to 24 years old) in July 2026 was only 7.1%, down from an average of 8.3% in 2025.
But while fears of an AI-driven jobs apocalypse have dissipated for the moment, new concerns have arisen. As open-weight models — many from China — have become more widely available and narrowed the gap with closed-weight frontier models, people are worried that American individuals and firms will become more vulnerable to dangerous cyberattacks.
At the same time, hundreds of companies and organizations, including Amazon, Google, and OpenAI, have signed the Open Weights and American AI Leadership letter, arguing in favor of open-weight models as an important part of a strong AI ecosystem. The letter called for “expanding access to AI, encouraging competition, robust application layers, and giving Americans greater control over the technology they rely on.”
In this context, enterprises and individuals are increasingly worried about “AI sovereignty,” a broad term that addresses fears of being excessively dependent on third-party AI models. According to the Foundation for American Innovation, 56% of business owners and C-suite executives are “concerned that their competitors will benefit from AI models trained on their organization’s proprietary data, workflows, or institutional knowledge.” Nearly all agree that transparency is important, and AI providers should disclose how customer data is used to improve AI systems.
AI sovereignty is also relevant for the ability of AI to tackle difficult problems in the physical world, including supporting national security, improving manufacturing and construction productivity, and lowering health care costs. Getting a positive return on investment in these areas requires the accumulation and protection of enormous new granular and proprietary data sets, which includes the slow and expensive collection of data on rare but important long-tail events.
The importance of AI sovereignty cannot be overstated, even though it’s tough to define exactly what it is. At one end of the spectrum, AI sovereignty can be defined as control over data, models, and infrastructure. For example, open-weight models running on self-owned hardware mean that there’s no possibility of leakage of private data and insights. At the other end of the spectrum, concerns about AI sovereignty are directly addressed by products such as Google’s Sovereign Cloud, Amazon’s AWS “digital sovereignty,” and Apple’s newly revamped Siri, which focuses heavily on privacy protection. Companies such as Palantir and Nvidia are building a “middle-layer” to allow enterprises to connect to AI models of their choice without exposing proprietary data.
Training on the broad corpus of the literate Web was necessary to bootstrap the first stage of the AI revolution. But that was the low-hanging fruit. Outside of coding, most executives are dissatisfied with their return on AI spend and the lack of business value delivered in real-world use cases and workflows. Even in the software domain, there’s a growing sense that enterprises are not yet getting enough bang for the buck.
The next stage—solving difficult real-world challenges — requires training on data that is typically located in different parts of the business and in different formats. Businesses need some application layer to make sense of that existing data, and put it into common language.
Moreover, it’s turning out in many areas that there’s a lot of data that simply doesn’t exist yet—physical data, material data, manufacturing data, clinical data. And giving enterprises the right incentives to spend heavily to collect that essential data will require some form of AI sovereignty to protect the data’s value.
Let’s be clear here. We do not foresee the popping of an AI bubble. Nor are we calling for heavy-handed regulation or government control. But we must realize the AI revolution is about to move into the next phase.
