China’s Cheaper AI Models Face a Monetization Test

TechNode framed China’s AI race around a harder commercial test: cheaper models and wider adoption are no longer enough unless companies can turn lower token costs into paying products.

The Sept. 1 UBS Securities briefing behind the article put three questions at the center of China’s large-model sector: capability, token return on investment and monetization.

UBS Securities China internet analyst Xiong Wei described a market moving from “token-maxxing” toward “token optimization,” with enterprise buyers weighing how much model intelligence each task actually needs.

That shift changes the competitive value of lower-cost Chinese models.

Xiong put development spending for some leading domestic systems below one-tenth of comparable overseas projects, and placed typical API charges at roughly 10% to 20% of international rival prices.

The advantage is meaningful for repetitive or lower-risk workloads, but the source makes clear that cheaper inference does not automatically create revenue.

UBS internet research head Kenneth Fong pointed to a structural limit for the country’s biggest online platforms.

Mobile user traffic and time spent are no longer expanding much, leaving AI to improve content production, advertising and recommendation systems inside a more crowded attention market.

More automated output still has to compete for the same consumer minutes.

Mango TV’s new AIGC drama shows how that tension is moving from model economics into media operations.

The Later Journey to the West opened Aug. 31 across Mango TV and Hunan Satellite TV’s evening schedule, giving China its first long-form AIGC series placed in a satellite-channel prime-time window.

The production plan is ambitious.

Season one is planned as 30 installments, each around 40 minutes, and the workflow depends on Mango Lingchuang, the broadcaster’s internal AIGC production system.

By mid-2026, Mango Lingchuang had been used by more than 40,000 professional users across over 3,900 projects.

The series also uses generated production assets at scale.

For the show, Mango Lingchuang produced 109 character elements and 143 scene elements, while the team is testing a parallel workflow that keeps later episodes moving through production as earlier episodes go through review and broadcast.

That process points to one possible business case for AI: shortening the distance between production, approval and release.

Early viewing data gave the experiment a stronger starting point, though not a final answer.

During the Aug. 31 launch window, real-time ratings put the premiere ahead of other provincial satellite channels, and ChinaTimes recorded 27.57 million Mango TV plays by Sept. 2.

Before broadcast, Hunan’s regulator had asked the project team to prove two things at once: a viable production route for AI-assisted long-form storytelling and a business model for seasonal AIGC drama.

The practical question for China’s AI market is therefore shifting from whether models can do more to whether cheaper production can create durable business value.

Mango TV now offers a visible test case: if faster and less expensive content pipelines do not translate into sustained audience attention and monetization, lower model costs will remain an efficiency gain rather than a full commercial breakthrough.