Jefferies says ByteDance's cloud arm Volcano Engine is targeting annual revenue of more than $14.7 billion before 2030, drawing on a media interview with its president that pointed to rapid growth in AI model services.
The broker, recapping an interview given to outlet 36Kr by Volcano Engine president Tan Dai, said the unit now expected to hit that revenue milestone earlier than previously planned.
Much of the optimism centres on model-as-a-service, or MaaS, which provides AI models to enterprises through the cloud rather than as a standalone business.
Jefferies said industry MaaS revenue was expected to grow five to ten times each year, with Volcano Engine revising its own target upwards from the level set at the start of the year.
The note highlighted the role of Seedance, Volcano Engine's video generation model, which contributes more than half of MaaS revenue.
Jefferies said Seedance ranked first in the Chinese market and generated close to half its revenue from overseas, with applications spanning drama, film and advertising as well as embodied AI, smart manufacturing and intelligent driving.
Tan reportedly sees larger long-term opportunities in large language models and agents than in video generation, given lower current penetration.
The broker described a flywheel effect, in which the commercialisation of coding and agents creates value that in turn funds better models.
It said model intelligence increasingly determined pricing, since more capable models created greater customer value.
Jefferies noted that an industry price war in 2024 had reflected weaker value creation at that time.
As evidence of accelerating adoption, the broker said the number of enterprises consuming more than one trillion tokens had risen from over 100 in December to more than 200 by June.
The note also flagged the launch of a professional edition of Volcano Engine's Doubao AI assistant, with monthly subscription packages ranging from about $5.60 for students to $74 at the premium tier.
Jefferies expects rising token consumption to benefit Chinese cloud service providers more broadly.
It named Baidu, Alibaba, Tencent and Kingsoft Cloud, alongside AI labs and the Kling video model, as beneficiaries, holding buy ratings on all four of the larger groups.
Citing analysis from Artificial Analysis, the broker pointed to a narrowing intelligence gap between US and Chinese models.
It added that Chinese models benefited from a fraction of the API cost of US rivals, helped by techniques including mixture-of-experts architecture and linear attention, opening significant opportunities in coding, workplace tools and longer-form content
