Unisound (09678.HK) released its 2026 interim results on August 28, and the report signals several positive shifts across its core financial metrics. The company reported faster revenue growth, margins expanding at a quicker pace than sales, sustained loss reduction, and a significantly stronger cash position, while its token-based business—a key new growth driver—has begun to scale up.

For the first half of 2026, Unisound recorded revenue of RMB 562 million, up 38.7% year-on-year. Gross profit climbed 42.0% to RMB 186 million, outpacing revenue growth and lifting the overall gross margin to 33.1%. Meanwhile, the company's net loss margin improved by more than 31 percentage points year-on-year. Notably, cash and cash equivalents reached nearly RMB 1 billion by mid-2026, a substantial 302.8% increase, while its order backlog surpassed RMB 1.5 billion, providing strong support for future revenue recognition. For an AI company still in a heavy R&D investment phase, the simultaneous improvement in revenue growth, loss margin, cash reserves, and order scale points to a notable shift in its operational fundamentals.

**Revenue growth accelerates to 38.7% with improving quality**

Revenue expansion is the most direct highlight of this interim report. In H1 2026, Unisound's revenue grew 38.7% year-on-year, marking a further steepening of the growth curve compared to the company's normalized pace over the past three years. At the same time, gross profit rose 42.0%, faster than revenue, indicating that business profitability wasn't sacrificed during expansion. From an operational efficiency perspective, the net loss margin improved by 31 percentage points year-on-year, which is particularly significant given the continued increase in R&D investment. During the reporting period, Unisound invested RMB 284 million in R&D, up 69.0% year-on-year, with R&D expenses accounting for 78.5% of total period expenses; the R&D team comprised 327 personnel, or 67.7% of all employees. In other words, the loss margin improvement wasn't driven by cutbacks in core technology investment but was achieved despite significant R&D intensification. This change suggests that as revenue scale grows and business reuse rates improve, Unisound's operating leverage is becoming more apparent. As of the reporting date, recurring revenue accounted for over 60% of total revenue. For an AI firm, a higher share of recurring revenue—as opposed to reliance on one-off project income—indicates stronger customer stickiness and revenue sustainability. The company's module reuse capabilities on its intelligent agent platform are also helping reduce duplicate development and delivery costs.

**Business reorganized into three segments, large-model commercialization revenue now separately visible**

Another notable change in this interim report is Unisound's first reclassification of its business by product format. The company has now organized operations into three major segments: enterprise intelligent services, large-model token-based services, and edge-side AI. Previously, Unisound disclosed revenue by industry verticals such as "smart healthcare" and "smart living." With this adjustment, project-based solutions, platform subscriptions, and API pay-per-call models are now clearly differentiated, offering a clearer view of the company's large-model commercialization revenue structure. In terms of revenue composition, enterprise intelligent services continue to play the foundational role. This segment is divided into intelligent agent applications, intelligent agent platforms, and industry solutions. In H1 2026, it generated RMB 478 million in revenue, up 35.7% year-on-year, accounting for more than 85% of total revenue. More importantly, this business is shifting from traditional one-off customization toward platform-based, modular reuse. Unisound describes its strategy as "strengthen base models, deepen applications." The core idea isn't simply to add more AI projects, but to consolidate proven capabilities in complex scenarios like healthcare, insurance, transportation, and high-end manufacturing into reusable modules, lowering the marginal cost of future deliveries. If this model continues to advance, it fundamentally changes the traditional AI business logic where "revenue growth must be accompanied by proportional delivery cost increases." Early signs of scaling are already visible in the order pipeline. As of the reporting date, Unisound's order backlog exceeded RMB 1.5 billion. Meanwhile, its medical large model is transitioning from individual hospital partnerships to regional, standardized replication. The company has served over 470 medical institutions, with more than 80% being top-tier Grade IIIA hospitals.

**Token business scales rapidly, over 60% gross margin emerges as a key highlight**

While enterprise intelligent services anchor Unisound's current revenue base, the token business represents the incremental opportunity that the market is watching more closely. In H1 2026, Unisound's token business generated approximately RMB 29.77 million in revenue, with Q2 alone surpassing RMB 25 million, reflecting a sequential quarterly growth of over 500%. In absolute terms, nearly RMB 30 million remains modest relative to overall revenue, but when assessed on growth trajectory, revenue model, and gross margin, this business carries significance far beyond its current revenue share. The standout metric is its gross margin exceeding 60%. According to company disclosures, token business growth is primarily driven by US-dollar revenue, with gross margin above 60%. In a competitive large-model API market where token prices are continually declining, such a high margin suggests the company isn't relying on low-price volume expansion, but rather targeting higher-value model invocation scenarios. This marks one of the most structurally significant changes in the interim report. A key market question regarding large-model commercialization has been whether model calls can genuinely become high-quality revenue, rather than just "large volume with thin margins." Unisound's current data offers an initial signal: token revenue isn't just growing rapidly—it's also maintaining robust margins. The company frames this logic as "intelligence density × token value." Intelligence density refers to achieving sufficient model capability at the lowest possible inference cost, while token value emphasizes whether model outputs ultimately enter high-value scenarios like healthcare and speech, translating into customer revenue gains, cost reductions, or risk mitigation. This also explains why the company hasn't prioritized low-price competition for its token business, instead focusing on medical, speech, and complex enterprise use cases.

**From selling projects to selling capabilities, the revenue model is transforming**

An AI company's valuation logic largely depends on the replicability of its revenue. If revenue mainly comes from one-off project deliveries, the business model resembles traditional software integration; if models, platforms, and modules can be repeatedly invoked, it suggests revenue growth becomes less dependent on incremental headcount. This is another underlying theme in Unisound's interim report. On one hand, enterprise intelligent services are steadily improving module reuse rates and customer repurchase frequency. On the other, the token business naturally operates on a pay-per-call basis. Combined, Unisound's revenue model is evolving from one-time deliveries toward a mix of "project revenue + platform revenue + call-based revenue + recurring operational revenue." The company's healthcare progress is especially notable. During H1 2026, its medical business expanded further into provincial medical insurance, regional healthcare, and leading Grade IIIA hospitals. In January, the company won the Jiangsu Provincial Medical Insurance Large Model project; in May, it secured successive contracts with Henan Provincial Cancer Hospital, Changzhou Traditional Chinese Medicine Hospital, and the Eighth Affiliated Hospital of Sun Yat-sen University. The company has now served over 470 medical institutions cumulatively. Additionally, its service boundary is extending from individual enterprises to regional AI public service platforms, with related operations already deployed in cities like Xiamen and Nanning. The significance of these projects extends beyond new orders—they validate the company's "standardized, platform-based, replicable" business model.

**Cash reserves grow to nearly RMB 1 billion, strengthening the safety buffer amid heavy R&D spending**

For large-model companies at this stage, cash levels are a critical metric for market assessment of sustained competitiveness. As of mid-2026, Unisound held nearly RMB 1 billion in cash and cash equivalents, up 302.8% year-on-year. At the same time, the company maintained intensive R&D investment, spending RMB 284 million in H1, which represented 78.5% of total period expenses. From a financial standpoint, these figures indicate that while the company remains in a technology investment and product iteration cycle, its cash reserves now provide a solid safety cushion. The order backlog exceeding RMB 1.5 billion further offers visibility into future revenue growth. For the market, this may be more significant than focusing on a single quarter's revenue gain: whether an AI company can navigate the sustained high-R&D-investment phase and ultimately improve its profit model through revenue scaling depends largely on whether it simultaneously holds technology, customers, orders, and cash. Based on this interim report, Unisound is concurrently strengthening all these indicators.

**Large-model commercialization enters the phase of financial results scrutiny**

Over the past few years, competition in the large-model industry has centered on parameters, benchmarks, and model capabilities. But entering 2026, the market is increasingly focused on a different question: can technology investment translate into revenue, gross profit, and cash flow? This interim report suggests Unisound is attempting to provide an answer. The token business offers the most compelling upside. Nearly RMB 30 million in semi-annual revenue is still in its infancy, but Q2's over 500% sequential growth and gross margin above 60% indicate the company's large-model operations are moving from technical capability to independent commercial revenue. With products like U2-Med, U2-ASR, U2-TTS, and medical imaging multimodal models being rolled out successively, the token business's scenario coverage is broadening further. Rather than interpreting this interim report as merely a revenue increase, it's more accurate to see Unisound undergoing a structural transformation of its business model: enterprise intelligent services provide scale and cash flow stability, the token business unlocks higher-margin, more scalable growth potential, and edge-side AI continues to contribute steady incremental business. From the 38.7% revenue growth and the 31-percentage-point improvement in loss margin to the rapidly scaling token business with over 60% gross margin, Unisound is progressing from "proving AI can be deployed" toward "proving AI can generate scalable, high-quality revenue." For an AI company still in its large-model investment phase, this is perhaps the most noteworthy takeaway from the interim report.