Huawei projects that autonomous artificial intelligence agents will dominate global AI activity by 2035, accounting for more than 90 percent of AI token usage and driving significant increases in computing power demands. The company detailed these expectations in its report titled “Intelligent World 2035: Turning Vision into Action,” released on Wednesday.

AI tokens refer to the units processed by models when they receive input or generate output, and Huawei anticipates annual token consumption will surge over the next decade. This growth is attributed to autonomous agents that continuously perceive their environments, reason through problems, make decisions, and interact with external tools without human intervention.

The report outlines the need for substantial advancements not only in computational capacity but also in security frameworks to manage the risks posed by independent software operating autonomously. Huawei emphasizes the importance of developing new systems that can safeguard these agents as they become increasingly capable and prevalent.

Meanwhile, leading AI researchers and industry figures in the United States have raised concerns about the rapid pace of AI development, particularly regarding autonomous agents. They warn that future systems capable of independently controlling hardware and acquiring resources could pose control challenges and unpredictable risks. Some experts have called for a temporary slowdown in advancing frontier AI technologies to address these concerns.

Huawei’s report, however, does not advocate for slowing development. Instead, it frames the ongoing evolution of AI agents as an opportunity to reshape computation and expand AI applications at scale, emphasizing innovation and deployment over cautionary measures.

As autonomous AI agents become more integrated into technology ecosystems worldwide, the industry faces a growing tension between harnessing the technology’s potential and addressing the new security and ethical considerations that arise with increasing autonomy and complexity.