Satya Nadella makes the case for AI independence
Welcome to AI Decoded, Fast Company’s weekly newsletter breaking down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company covering emerging technology, AI, and tech policy.
This week, I examine Satya Nadella’s argument that enterprises need a more balanced and less dependent relationship with frontier-model providers. I also look at a new warning from economists and technology experts about AI-driven job losses, as well as the ways states are beginning to regulate data centers’ water use.
Sign up here to receive this newsletter by email every week. And if you have comments on this issue or ideas for future ones, drop me a line at sullivan@fastcompany.com and follow me on X @thesullivan.
Microsoft’s Satya Nadella makes the case for sovereign AI
In the early days of the generative AI boom, enterprises hesitated to share their trade secrets with models controlled by third parties such as OpenAI and Anthropic. Artificial intelligence providers eased some of those concerns through service guarantees. But as Satya Nadella, CEO of Microsoft, points out in an essay posted on X, enterprises must still share significant amounts of institutional knowledge with the models to benefit from them. That can include user prompts, feedback given to the model, and the workflows followed by AI agents.
“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,” Nadella writes in his essay. “The better you want the model to perform, the more of that knowledge you have to feed it!”
Nadella argues that enterprises must rebalance this one-sided relationship. His proposed solution is sovereign AI, loosely defined as an arrangement in which organizations own and control their data, models, and AI infrastructure.
“[E]nterprises need a real trust boundary for their human capital and token capital to compound,” Nadella writes. “It is where an organization’s data, traces, evals, adapted weights, and memory accumulate and improve together.”
By “traces,” he means the steps and decisions AI agents generate as they work toward a goal. By “evals,” he means the feedback workers provide to steer a model toward useful and reliable answers. “Adapted weights” are the specialized parameters an enterprise tunes to make its models better suited to the company’s particular tasks and workloads.
All of these are direct expressions of how an enterprise solves problems and builds knowledge. Nadella argues that this valuable institutional learning should remain the sole property of the enterprise. “[A] company should be able to use a model without giving up the knowledge that makes it unique,” he writes.
If that knowledge flows into a third-party frontier model, the model could incorporate it in ways that ultimately benefit other companies, including the enterprise’s competitors.
Nadella’s sovereign AI ideal appears to conflict with OpenAI’s business model, which largely involves providing customers access to its models as a managed service. That tension is especially notable because Microsoft is OpenAI’s biggest financial backer and owns a large stake in the company.
Nadella’s most provocative argument is that enterprises will increasingly demand the right to use the outputs of third-party frontier models to train their own private models through a process called distillation. Frontier-model providers, including OpenAI, expressly forbid this practice.
Nadella finds that prohibition ironic. Frontier-model developers have relied on fair-use interpretations of copyright law to justify scraping enormous amounts of online content to train their models. They are far less tolerant when others seek to use their own models’ outputs in a similar way.
200 economists warn of major AI job disruption
One of the biggest near-term structural risks posed by AI is that the technology could replace workers or eliminate jobs. So far, economists have found little clear evidence that this is happening in a major way. An analysis last month from the Yale Budget Lab found no connection between “measures of AI usage” and “changes in employment or unemployment.” A recent PwC study of a billion job advertisements, however, found an erosion of junior-level positions, possibly because employers are using AI tools to perform work that would previously have gone to entry-level employees.
Economists have expressed widely varying views about AI’s likely effect on employment throughout the generative AI boom. That may be changing.
A new statement signed by more than 200 leading economists, prominent AI researchers, and Nobel Prize winners suggests that a broader consensus may be forming. The statement warns that unprecedented economic upheaval could be coming.
AI may become far more powerful during the next decade, the statement says, transforming the economy more radically and rapidly than the Industrial Revolution. That transformation could improve living standards while also causing “large-scale job displacement.”
The signatories say economists, policymakers, and technology leaders should move quickly to understand the economics of the transition and establish the “incentives, guardrails, and institutions” needed to steer AI toward complementing, rather than sidelining, human workers and benefiting society broadly. The statement bears more than 1,500 signatures.
A day later, on July 14, Google DeepMind’s CEO published an essay calling for an international body of scientists, policymakers, and AI industry representatives to evaluate major frontier models and determine whether they have adequate safeguards before they are released to the public.
How states are regulating AI data center water use
A new report from the University of Colorado Law School finds that state legislatures across the country are responding to the growing water demands of data centers with a mix of tax incentives, reporting requirements, and conservation mandates.
Daniel Anderson, a coauthor of the report, said artificial intelligence is driving data center development faster than traditional water-planning systems were designed to accommodate. Data centers require large volumes of water for cooling, the report says, and much of the recent growth in data center infrastructure has occurred in the western United States, where water supplies are already limited.
In the absence of federal standards, states have adopted varied approaches to regulating data center water use.
Some states require data center operators to track, verify, and publicly disclose their annual water consumption. Others restrict the use of potable water for cooling or require facilities to meet water-efficiency standards. Some make tax exemptions or development grants contingent on operators’ adopting low-water or waterless cooling technologies. States are also reinforcing existing water-rights rules as they apply to data centers.
The water consumption of AI data centers has become a contentious issue over the past year. Initial reports emphasized the enormous amounts of water required to cool the graphics processing units that run AI models. More recent reports have noted that while data centers may require large amounts of water when they begin operating, many use cooling systems that recycle the same water over and over again.
“Today’s total direct water usage by U.S. data centers is small relative to other industries’ consumptive water use totals, but the rapid additional strain and projected growth in certain communities is prompting state legislators to act,” the Colorado law school’s report states.
The report’s authors say state-level water laws addressing data centers remain at an early stage. The policy landscape will likely change as states assess the effectiveness of recently adopted laws. The authors also say policymakers need better water-use data and standardized reporting requirements to guide future decisions.
Meanwhile, public opposition to proposed data centers has grown and could become a significant issue in this year’s midterm elections.
The electricity demands of data centers may pose an even greater concern. Studies have found that new facilities can strain the public power grid and, in some cases, contribute to higher electricity rates.
The Federal Energy Regulatory Commission, however, told grid operators last month to accelerate electricity-interconnection requests from data centers and other large power users. The agency ordered six major grid operators to demonstrate that new data centers can “connect to the transmission system in a timely and orderly manner.” The data centers will be responsible for paying the costs of those interconnections.
More AI coverage from Fast Company:
- OpenAI’s fight with Apple is really about Silicon Valley’s war for talent
- The cover letter is officially dead: AI has created a new job-hunting paradox
- The many controversies of Meta’s AI glasses
- I built an agentic AI clone of my family to plan our summer travel
Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
