To The Moon

The recent approval of the IPO application for Shanghai SENSENOVA Co., Ltd. marks a significant step for China's AI chip industry, potentially adding its sixth listed company.

On June 15th, the Shanghai Stock Exchange announced that SENSENOVA's initial public offering application had been approved by the listing committee. As an AI chip company backed by Tencent Holdings Ltd. (HKG: 0700), SENSENOVA's listing process has progressed relatively swiftly. However, among the so-called "Four Domestic GPU Dragons," it is the last to receive such approval, following earlier listings by peers such as Moore Threads, Muxi, and Biren Technology.

The company aims to raise 6 billion yuan through its IPO to fund R&D and industrial projects for its fifth and sixth-generation AI chip series, alongside an advanced AI software-hardware co-innovation project. This fundraising target is lower than the combined totals of Moore Threads and Biren but exceeds those of Muxi and TianShu ZhiXin.

Driven by surging global demand for intelligent computing power, continuous technological iteration, and expanding product deployment, SENSENOVA has reported consistent revenue growth in recent years. Its primary income streams are AI accelerator cards/modules and intelligent computing systems/clusters. From 2023 to 2025, its operating revenue grew from 301 million yuan to 990 million yuan, representing a compound annual growth rate of 81.32%.

Despite this growth, the company remains unprofitable due to sustained high R&D expenditures. It projects that it may achieve profitability on a consolidated basis in 2026 or 2027. The company's founders noted in an investor statement that while revenue is growing rapidly as products are commercialized, the domestic AI computing industry is still nascent, and cloud AI chips require significant R&D investment with long customer validation cycles, necessitating further sales volume to reach profitability.

In terms of profitability, SENSENOVA's main business gross margin lags behind the average of five listed AI chip peers. This is primarily because over 80% of its revenue from AI accelerator cards/modules comes from inference products, which have relatively lower gross margins.

A notable aspect of its business is its increasing reliance on related-party transactions with Tencent. In 2025, sales to Tencent subsidiaries surged, accounting for over 80% of its total operating revenue. The company stated that, given Tencent is a major domestic demand source for AI computing power, it expects this high sales concentration to persist for the foreseeable future.

Industry observers note that AI chip companies backed by major internet firms have a natural advantage in deploying products across diverse, large-scale scenarios. However, constraints in capacity and resources often mean these companies prioritize their backer's procurement needs, which can create a shortcoming in expanding their clientele beyond the internet sector. Conversely, independent AI chip firms may find it easier to secure clients in non-internet fields but face significant challenges in becoming suppliers to domestic internet giants.

For companies seeking a listing on the STAR Market, losses are not necessarily a critical flaw; regulators are often more concerned about heavy reliance on a single client. The fact that SENSENOVA's IPO was approved despite its significant revenue dependence on Tencent reflects regulatory tolerance for different AI technology pathways within the context of domestic substitution, according to market analysts.

Revenue Growth Amidst Persistent Losses

Similar to other listed AI chip companies, SENSENOVA has seen sustained revenue growth in recent years due to the global explosion in intelligent computing demand. However, it has yet to turn a profit because of continuous R&D investment.

Over eight years, the company has self-developed and iterated four generations of architecture, resulting in five cloud AI chips, building a complete product system encompassing AI chips, accelerator cards/modules, intelligent computing systems/clusters, and an AI computing and programming software platform.

From 2023 to 2025, its three main revenue sources—AI accelerator cards/modules, intelligent computing systems/clusters, and IP licensing/others—showed different trends. Revenue from AI accelerator cards/modules grew consistently, rising from 186 million yuan to 856 million yuan over the period.

In 2024, both sales volume and average selling price for these products increased, driving revenue growth of over 60%. The company attributed this to the mass production of its third-generation AI accelerator card, which offered higher performance and a higher average price. In 2025, despite a 6.5% drop in average price, a 198% surge in sales volume, driven by the third-generation card's deployment in internet clients' AI business scenarios, led to a 178% year-on-year revenue increase for this segment.

In contrast, revenue from intelligent computing systems and clusters fluctuated significantly, surging nearly threefold in 2024 before dropping over 60% in 2025. The company explained that the 2025 decline was due to focusing resources on meeting the procurement needs of major internet clients, considering its own product inventory, which limited growth in this business line.

For the first quarter of 2026, revenue skyrocketed 1,475% year-on-year to 287 million yuan due to further product deployment at client sites. Based on Q1 performance and existing orders, the company forecasts H1 2026 revenue between 1.06 billion and 1.15 billion yuan, representing growth of 258.68% to 289.13%, and expects to match its full-year 2025 revenue in the first half of 2026.

Like its peers, SENSENOVA remains loss-making, though the net loss has narrowed. From 2023 to 2025, its net loss attributable to shareholders decreased from 1.665 billion yuan to 1.164 billion yuan. The Q1 2026 net loss was 444 million yuan, a year-on-year increase. The company projects an H1 2026 net loss between 577 million and 608 million yuan, which would represent a narrowing compared to the previous year. It anticipates achieving consolidated profitability in 2026 or 2027, barring major unforeseen impacts like international trade friction.

Comparing Profitability with Peers

SENSENOVA's gross margin trails the average of its listed peers, primarily due to its product strategy of "gradually exploring training products while rapidly advancing inference products." This has resulted in a high proportion of revenue from lower-margin inference products.

From 2023 to 2025, its main business gross margins were 22.60%, 30.59%, and 31.78%, respectively, below the average of approximately 57% for five listed AI chip companies including Cambricon. The gross margin for its AI accelerator cards/modules also showed a significant gap compared to the average of similar products from peers.

In 2024, a more than 60% increase in the average selling price of its AI accelerator cards/modules, far outpacing a 13% rise in unit cost, lifted the segment's gross margin. In 2025, a decline in average selling price coupled with rising unit costs led to a year-on-year drop in product gross margin.

The company explained that over 80% of its AI accelerator card/module revenue comes from inference products, whereas for domestic peers, over 50% comes from training or training-inference integrated products. Due to clients' extreme cost requirements for inference scenario deployment, SENSENOVA's inference products have appropriately trimmed hardware configurations and performance compared to training/integrated products. This product structure difference results in lower unit prices and gross margins for its AI accelerator cards/modules compared to peers.

In response to regulatory inquiries, SENSENOVA detailed gross margin differences between its training and inference cards across generations. The high proportion of inference products stems from its strategic product iteration path. The company stated this is a pragmatic choice, as inference scenarios allow for faster commercial deployment and self-funding to support ongoing investment in more complex training products.

When its third-generation product was initiated, generative large language models were still in early stages. Based on the market demand at the time, which focused on traditional AI and search-advertising-recommendation models, the company launched only an inference-oriented accelerator card (S60) for that generation. Its fourth-generation product, the L600 training-inference integrated accelerator module, has completed chip tape-out but has not yet entered mass production and delivery. The company is collaborating with clients on R&D for AI large model training applications based on the fourth-generation product but widespread deployment has not yet occurred.

SENSENOVA asserts that its current product structure, focused on inference accelerator cards while steadily advancing training-inference integrated products, aligns with the current stage of domestic AI computing market demand, matches the product direction of peers, and is consistent with the long-term roadmap of international mainstream manufacturers, thus fitting the development trend of the AI chip industry.

Future Prospects with Tencent Backing

Transactions with related party Tencent were a key focus in the regulatory review. According to the prospectus, Tencent subsidiaries are the largest shareholders, holding a 20.2580% stake. From 2023 to 2025, sales to Tencent subsidiaries grew significantly, reaching 830 million yuan in 2025 and accounting for 83.79% of total revenue that year.

Notably, in 2025, sales to Tencent increased by 557 million yuan year-on-year, while revenue from other clients decreased by approximately 290 million yuan. The company stated that the transaction prices with Tencent, determined through negotiation considering its status as a long-term strategic partner, are fair, even though they are lower than prices charged to non-related third parties for similar products.

SENSENOVA began collaborating with Tencent in 2019, jointly investing significant resources in areas like business scenario adaptation and optimization. Its products have been deployed at scale across many of Tencent's AI business scenarios, forming a deep cooperative relationship. The company noted that Tencent's demand for AI accelerator cards currently exceeds SENSENOVA's supply capacity, and it has focused its limited resources on Tencent to rapidly enhance product and technological capabilities within the massive, high-concurrency, demanding real-world business scenarios of a leading internet client.

Given that Tencent is a primary domestic demand source for AI computing power, SENSENOVA expects the high proportion of sales to Tencent to continue for some time. However, it also warned of risks: if its products fail to meet Tencent's technical needs, if domestic AI technology/applications slow down leading to reduced capital expenditure by internet giants, or if the cooperative relationship is replaced by other suppliers, it could lead to reduced procurement volumes or lower pricing from Tencent, adversely affecting future business and performance.

In its disclosures, SENSENOVA stated that, considering industry supply conditions and its own capacity and R&D resource constraints, it prioritizes core clients with large demand potential and strong cooperation stickiness, while steadily developing other high-quality clients. Among internet clients, for a key potential client, its fourth-generation product has passed initial hardware system and model matching tests and is entering the stage of finalizing a grayscale testing plan. Grayscale testing is expected to commence in 2026, with potential for small-scale delivery by December 2026 and mass delivery starting in 2027.