SenseTime CEO bets on ‘task economy’ as token prices set to drop

The commercialisation of artificial intelligence is poised to transition from a “token economy” to a “task economy” as token prices decline in the coming years, according to Chinese AI company SenseTime. “The pricing of tokens will inevitably drop, much like the cost of telecoms data did two decades ago,” SenseTime CEO Xu Li said in an interview with the South China Morning Post. After foundational models and computing power become basic infrastructure, the true commercial value will shift towards a “task economy” where users pay for completed, complex tasks like reviewing architectural blueprints or generating product videos, rather than raw computing usage, Xu said on the sidelines of the World AI Conference (WAIC) in Shanghai on Saturday. His comments come as major AI companies in China are banking on the so-called token economy, as enterprises increasingly adopt AI to boost productivity and automate business operations. China’s daily AI token consumption surged more than 1,000-fold in the past two years, according to official data. With AI coding currently proving to be one of the most popular verticals for AI to generate cash flow, Xu said the next breakthrough lies in work and design scenarios. While early adoption focused on programmers, the next wave would reach the masses, he said, adding that the total addressable market was likely to expand by at least 10 times. “Major shifts in industry conditions can produce tenfold changes when a critical inflection point is reached,” Xu said. The nature of demand would expand beyond professionals, to a new scenario where they represented only a small fraction of the total user population, he added. While US tech giants have established a strong global consensus when it comes to current applications, Xu said China holds a distinct advantage in application diversity. China’s immense population and complex use cases create fertile ground for “super individuals” to discover novel AI applications. These diverse, real-world demands could then drive AI ecosystem development faster than in less dynamic markets, he said. To achieve the task economy, SenseTime is aggressively doubling down on its commitment to multimodal AI models – which can process and understand the physical world through images, video, and 3D structure – unlike traditional large language models that focus only on text. At the four-day WAIC, which concluded on Monday, SenseTime launched its latest generation SenseNova U1 Pro multimodal model, which uses interleaved text-image chain-of-thought to create “production-ready” visual assets that look less AI-generated, it said. Still, Xu acknowledged the immense challenges of developing multimodal AI, as it required complex data preparation and computing power alignment. Even US tech giant Google has struggled to advance efficiently in this area. But Xu said SenseTime would stick to this road map. “We don’t think there are alternative pathways that could raise the ceiling of intelligence [of AI models] more rapidly,” he said. Meanwhile, Xu said the Hong Kong-listed SenseTime was reaping the benefits of the current agentic AI boom, paving the way for profitability. The company’s generative AI applications and token business were currently seeing “quite good gross margins”, while the future revenue structure would be increasingly anchored by application-based services, he said.