Author｜Yang Xi

Editor｜Gao Shan

While the market continues to speculate on whether AI will displace human jobs, Silicon Intelligence has boldly started to package and market its 'silicon-based workforce'.

On May 21, Nanjing Silicon Intelligence Technology Group Co., Ltd. (hereinafter referred to as 'Silicon Intelligence') resubmitted its prospectus to the Main Board of the Hong Kong Stock Exchange, with CMB International and DBS Group serving as sponsors. Its previous application, submitted in October 2025, had lapsed.

Silicon Intelligence positions digital human agents as a new form of labor, distinct from 'carbon-based' human labor. Yet, the company, a leader in selling tireless 'silicon-based labor', has itself suffered losses for three consecutive years. How far is it from achieving profitability?

Revenue Growth Without Profit Increase

Silicon Intelligence generates revenue by offering comprehensive and synergistic silicon-based labor solutions, encompassing silicon-based intelligent voice, silicon-based digital human videos, silicon-based digital human live streaming, silicon-based digital human intelligent interactions, and silicon-based fully automated content production. To date, Silicon Intelligence has deployed over 100,000 digital human agents across various industries.

According to CIC, based on the revenue from providing digital human agent solutions in 2024, Silicon Intelligence ranked first among all digital human agent providers in China, with a market share of 32.2%, and second globally, with a market share of 16.1%.

Despite its leading industry position and sustained revenue growth, the company continues to grapple with losses. The contradiction of revenue growth without corresponding profit increase remains pronounced.

From 2023 to 2025 (the 'reporting period'), Silicon Intelligence's revenue steadily increased, reaching RMB 531 million, RMB 655 million, and RMB 789 million, respectively. However, its annual profits were -RMB 95.907 million, -RMB 112 million, and -RMB 25.915 million, respectively. The company explicitly stated in its prospectus that it anticipates continued losses in 2026.

On one hand, this is attributable to additional R&D expenditures on technological upgrades and model training at Silicon Intelligence.

During the reporting period, Silicon Intelligence's R&D expenditures rose annually, reaching RMB 130 million, RMB 150 million, and RMB 163 million, respectively. However, their proportion of revenue showed a declining trend, at 24.4%, 22.9%, and 20.7%, respectively.

Notably, the cost of cloud services, reflecting the intensity of investment in cloud computing resources, exhibited sharply contrasting trends in the company's sales costs and R&D expenditures.

As revenue grew, the proportion of cloud service costs in sales costs decreased from 38.4% in 2023 to 14.1% in 2025. Conversely, cloud service fees in R&D expenditures surged from RMB 74.395 million in 2023 to RMB 137 million in 2025, with their proportion increasing from 57.4% in 2023 to 83.8% in 2025.

This indicates that the company is allocating more cloud computing resources to model training and technological iteration at the R&D end, rather than to scalable services at the delivery end. To some extent, this explains the underlying reasons for the company's sustained performance pressures: investments in technological upgrades have not been promptly and effectively translated into increased product value.

On the other hand, the pressures stem from market promotion efforts. Although the company's sales and marketing expenditures decreased annually during the reporting period, reaching RMB 110 million, RMB 82.57 million, and RMB 55.235 million, respectively, related investments are expected to rise again due to increasing demand for promoting businesses such as silicon-based labor solutions. In this IPO, Silicon Intelligence will allocate part of the proceeds for marketing and promotion.

Highly Concentrated Customer Base

Amid performance pressures, Silicon Intelligence's gross profit margin remains low, at 45.9%, 34.3%, and 34.5% during the reporting period, respectively.

The slight recovery in gross profit margin in 2025 was primarily due to the company's newly launched silicon-based fully automated content production business, which contributed a high gross profit margin of 54.9%, slightly boosting the overall level. However, with this business accounting for only 6.5% of revenue, it was insufficient to reverse the overall weak profitability.

In terms of revenue structure, silicon-based digital human intelligent interaction products contributed the majority of the company's revenue during the reporting period, at RMB 460 million, RMB 613 million, and RMB 723 million, respectively, accounting for 86.6%, 93.4%, and 91.6% of revenue.

During the reporting period, the gross profit margin of this product showed a declining trend, at 43.3%, 32.9%, and 33.3%, respectively. In response, Silicon Intelligence explained in its prospectus that the lower gross profit margin in 2024 was mainly due to the company's efforts to introduce relevant solutions to the market and offer discounts to attract customers. Despite its novel concept, it ultimately succumbed to trading price for volume.

Furthermore, the high customer concentration further amplifies profitability risks. During the reporting period, revenue from the company's top five customers accounted for 57.7%, 78.9%, and 57.4% of total revenue, respectively. Among them, the largest customer (Customer A, a leading telecommunications operator) contributed 36.8%, 64.4%, and 41.3% of revenue, respectively.

From an industry perspective, the proportion of revenue from telecommunications industry customers increased annually, reaching 69.3% in 2025, while the proportion from internet technology industry customers shrank from 44.9% in 2023 to 25.0% in 2025.

To forge partnerships with large customers such as commercial banks and telecommunications operators, Silicon Intelligence often adopts competitive pricing models, resulting in lower gross profit margins and, to some extent, compressing product profit margins.

Silicon Intelligence's heavy reliance on a single industry and a single customer means that any changes in its cooperative relationship with core customers will significantly impact the stability of its revenue.

More worryingly, Silicon Intelligence's ability to acquire new customers continues to weaken. During the reporting period, the number of new customers decreased sharply from 890 in 2023 to 461 in 2024 and further to 254 in 2025. Meanwhile, the total number of customers also declined annually, from 1,009 in 2023 to 680 in 2024 and further to 431 in 2025.

With the loss of existing customers and weakened new customer acquisition, pressures at the customer acquisition end continue to intensify. The average cost of acquiring a customer also climbed from RMB 123,000 in 2023 to RMB 222,000 in 2025.

Cash Flow Pressures

Amid a highly concentrated customer structure and deteriorating new customer acquisition, Silicon Intelligence's operating cash flow also remains unfavorable.

Since Silicon Intelligence primarily serves large customers through locally customized project deployments, the proportion of local deployments increased from 91.4% to 98.7% during the reporting period, while the proportion of cloud deployments continued to shrink. This revenue model, characterized by heavy delivery and slow repayment, further exacerbates the company's cash flow pressures.

During the reporting period, Silicon Intelligence's trade receivables increased annually, reaching RMB 147 million, RMB 232 million, and RMB 282 million, respectively. Among them, 56.8%, 77.4%, and 62.5% came from Customer A. Meanwhile, the trade receivables turnover days were 70, 106, and 119 days, respectively.

According to the prospectus, Silicon Intelligence's operating cash flow has experienced net outflows for three consecutive years. During the reporting period, the net cash flow from operating activities was -RMB 50.808 million, -RMB 82.216 million, and -RMB 122 million, respectively, with the outflow scale expanding annually.

The company's cash reserves are highly dependent on external financing. In 2025, Silicon Intelligence's net cash inflow from financing activities reached RMB 171 million, driving the year-end cash and cash equivalents back up to RMB 218 million.

Meanwhile, Silicon Intelligence had granted special rights such as redemption rights to investors. As of the end of 2023, Silicon Intelligence's redemption liabilities were as high as RMB 913 million, directly driving up the net current liabilities for the year to RMB 583 million. In 2024, due to the unconditional termination of redemption liabilities, they were terminated and reclassified as equity, allowing the company's financial position to recover.

However, such operations, which rely on the removal of non-recurring provisions to beautify the balance sheet, cannot conceal the essence of insufficient profit-generating capacity in the core business. If financing conditions tighten or the listing process is disrupted in the future, the company's liquidity pressures will suddenly intensify.

Perhaps due to its novel concept, Silicon Intelligence has indeed been highly sought after in the capital market.

From December 2017 to June 2025, Silicon Intelligence successively completed eight rounds of financing, including Pre-A, A, A+, A++, B, B+, C, and D rounds, raising a total of RMB 869 million. The financing lineup includes well-known capital firms such as Linzhi Tencent, Sequoia Zhisheng, CMB Fund, Guoxin Fund, Pengpai Capital, Hejun Capital, and Shenzhen Gongying.

The latest round of financing was completed in May 2025, with the company securing RMB 200 million in Series D financing from Jiaxing Hi-Tech. After this round, the post-money valuation reached RMB 3.15 billion.

Silicon Intelligence's second prospectus submission, despite being supported by its industry-leading position and star-studded capital backing, faces persistent losses and a weak operating foundation, leaving its profitability turning point elusive. As the AI halo fades, the real test may have only just begun.