The global balance of power in AI is shifting, according to Stanford University’s 2026 AI Index, with the U.S. no longer holding a clear lead over China in model performance.
The report, produced by the Stanford Institute for Human-Centered Artificial Intelligence, notes that U.S. and Chinese systems now frequently trade top positions in benchmark rankings. In some cases, the difference between leading models is marginal, measured in single-digit percentage points.
While the U.S. retains advantages in funding, advanced chips, and large-scale infrastructure, China has built strength in research output, patent filings, and robotics deployment.
The competition has also broadened across the globe. Countries including South Korea are emerging as aggressive innovators, particularly in patents per capita, while dozens of nations are investing in domestic computing capacity to build tech independence.
The global AI push has produced a growing network of state-backed supercomputing systems, though the expansion is uneven. Regions such as South America and parts of the Middle East risk falling behind, raising concerns about a widening technological divide tied to access and investment.
Private Companies Take Control
The report notes that control of AI development is consolidating within private companies. More than 90 percent of significant models now originate from these for-profit companies. As a result, capability is increasing but transparency is declining. Many leading developers no longer disclose key details about how their systems are trained, and most recent models have been released without supporting code.
This shift coincides with rising political influence from major AI firms. Their presence in policy discussions has grown sharply, while independent academic participation has diminished. Public confidence has moved in the opposite direction. In the U.S., only about one-third of respondents say they trust government oversight of AI, among the lowest levels recorded globally.
Despite these concerns, adoption continues at a rapid pace. Generative AI tools have reached more than half of the global population within three years, surpassing earlier technologies such as the internet and smartphones.
Still, usage varies widely. The U.S. ranks well below several Asian markets in this regard, where expectations for AI’s impact are significantly higher.
Corporate investment has surged into the hundreds of billions of dollars annually, and consumer value from AI services continues to climb. But here too, benefits are not evenly distributed. Early signs of workforce disruption are emerging, particularly among younger employees in fields most exposed to automation.
Experts vs. Public Opinion
The Stanford report notes a growing disconnect between expert and public opinion. A majority of specialists expect AI to improve employment outcomes, while far fewer members of the public agree. That skepticism reflects real shifts in hiring patterns, where entry-level roles in software and customer service have already declined.
Technically, AI systems are advancing quickly, though not consistently. Models now perform at or above human levels on certain scientific and reasoning benchmarks, including graduate-level problem sets. Yet they struggle with basic tasks such as interpreting analog clocks or completing multi-step real-world activities.
Researchers describe this uneven capability as a jagged frontier, where progress in complex domains coexists with persistent gaps in common-sense reasoning. The pattern extends into science and medicine. AI is contributing to research output and clinical workflows, though rigorous validation remains limited, with many studies relying on simplified or synthetic data.
Environmental costs are also rising alongside capability. Training large models requires vast amounts of energy, with emissions comparable to tens of thousands of vehicles annually. Data center demand now approaches the power consumption levels of entire countries, while water usage for cooling systems has reached levels comparable to the needs of millions of people.
“AI’s capabilities are advancing quickly; less so, our ability to measure and manage them,” the report states.
