Unisound’s Half-Year Results Put Enterprise Agents in Focus

Introduction

As the initial excitement around AI agents fades, the market is asking a more practical question: can agents generate recurring revenue? Unisound’s reported 2026 interim results offer one example from China’s enterprise AI market. Enterprise intelligent services continue to provide the revenue base, while token-based model access is developing into a potential second curve.

Key figures

• First-half revenue reached RMB 562 million, up 38.7% year over year.
• Agent-related business generated RMB 478 million, or about 85.1% of total revenue, up 35.7%.
• Token revenue approached RMB 30 million, up roughly 760% year over year. Second-quarter revenue exceeded RMB 25 million, rising more than 500% sequentially, with gross margin above 60%.
• Repeat revenue accounted for more than 60% of total revenue. Contract value increased 65% year over year, and orders on hand exceeded RMB 1.5 billion.
• Gross profit rose 42% to RMB 186 million, while the loss attributable to owners narrowed 20.2% to RMB 237 million.

The figures also clarify what Unisound is selling today. Its core business combines agent applications, an agent platform and integrated solutions, rather than relying only on direct model access. The company says its systems have been deployed in healthcare, insurance, transportation and manufacturing. It has served more than 470 medical institutions, over 80% of which are tertiary hospitals, according to the cited report.

Turning delivery work into reusable capability

Enterprise AI projects are difficult because they must connect to data, permissions, workflows and legacy systems. Unisound’s approach combines its U2 model and related model portfolio with a platform for knowledge, tools and workflow orchestration. The final layer is a business solution designed to produce measurable outcomes.

The potential advantage is reuse. Knowledge and software components developed for one customer can be adapted for others in the same industry. Unisound reports that its agent platform has accumulated 1,773 instantiated agents, with one module reused as many as 119 times. It also says implementation costs fell from 21% of the relevant figure a year earlier to 17%.

Token access represents a different monetization route. Customers can call models through a public-cloud API and pay according to usage, allowing model capability to be sold separately from a full project. However, token revenue is still much smaller than enterprise services. Its rapid growth will need to be tested against customer retention, actual usage and inference costs.

Why it matters—and what remains uncertain

The broader lesson is that industrial AI commercialization may require three steps: start with frequent, costly and measurable business problems; convert project experience into reusable modules; then offer multiple purchasing options, including private deployment, platform services and token-based access.

There are clear limitations. Unisound has not yet reached profitability, and research spending remains substantial. Revenue acceleration and a narrower loss are encouraging, but they do not by themselves prove that the model is mature. Investors will need to watch recurring revenue quality, implementation efficiency, token growth and cash consumption.

For now, Unisound’s results are best viewed as a stage of validation. Agents become durable businesses only when they enter real workflows, deliver auditable results and persuade customers to keep paying.

Source: QbitAI