A survey finds the U.S.-China gap in AI performance has almost disappeared.

A SiliconANGLE report on April 13, citing Stanford University’s Human-Centered AI institute (Stanford HAI) and its 2026 AI Index report, said earlier reports showed the United States held a clear lead, but the two countries’ models now take turns occupying top benchmark positions. The United States still leads in capital, infrastructure and AI chips, while China holds the upper hand in patents, papers and autonomous robot development. Amid U.S. competition, South Korea drew attention by ranking first in the world for patent applications per capita.

The report also highlighted a trend of declining transparency as private companies expand their influence in the AI market.

Google, Anthropic and OpenAI are not disclosing the scale of training data or training time for their latest models, and 80 of 95 major models released last year were published without training code.

Public trust in AI also fell to a record low. Among U.S. citizens, 31 percent said they trust government AI regulation, the second-lowest among surveyed countries after China at 27 percent. In the European Union, 53 percent of citizens expressed trust, in contrast.

Generative AI has spread to the point that 53 percent of the world’s population uses it regularly. That is faster than PCs, the internet and smartphones. But the United States, despite being a leading country in AI development, ranked 24th globally, with a regular use rate of 28.3 percent.

The report also highlighted concerns about environmental costs. It estimates that carbon dioxide emitted in xAI’s training of its latest model Grok 4 is about 72,000 tons. Water used for GPT-4o inference workloads is enough for 12 million people to drink.

Scientists’ individual productivity has tripled with the adoption of AI tools, but the report also points to a decline in diversity as research scope tilts toward data-rich topics.