For a long time, the market has viewed domestic general-purpose computing power chips with the impression that "the technology works, but commercialization remains unproven." However, as Denglin Technology formally knocks on the door of a Hong Kong Stock Exchange IPO, a host of domestic computing power chip makers are delivering real evidence of technological industrialization through shipments, revenue, and cash flow.
On October 25, according to HKEX disclosures, Suzhou Denglin Technology Co., Ltd. (hereinafter referred to as "Denglin Technology") passed the main board listing hearing of the Hong Kong Stock Exchange, with CITIC Securities, CITIC Securities International, and BOC International serving as joint sponsors. The company officially submitted its 18C specialist technology IPO application to HKEX on February 27, and after more than seven months it finally knocked on the door of the Hong Kong Stock Exchange.
Founded in 2017, the company is a general-purpose graphics processing unit (GPGPU) provider that develops and sells GPGPU products to meet AI computing needs across a wide range of industries in China. According to Frost & Sullivan data, based on 2025 revenue, Denglin Technology is the third-largest company in China's enterprise AI GPGPU market, with a 1.0% market share. Next, let us move beyond the simple narrative of a "technology myth" and examine this heterogeneous GPGPU maker's true quality from two dimensions: operating performance and industry landscape, as well as the domestic computing power growth opportunity it represents.
Revenue Breaks RMB 200 Million in 4 Months, Operating Leverage Begins to Release
According to Zhitong Finance, Denglin Technology mainly focuses on the development and sales of hardware products (especially GPGPU boards), supplying original equipment manufacturers (OEMs), system integrators, and AI solution providers. The company operates under a fabless business model, which is the mainstream business model for companies in the semiconductor industry focused on chip innovation and architectural advancement. This model allows the company to concentrate resources on the design and development of energy-efficient GPGPUs while leveraging the expertise of third-party partners for manufacturing.
For fabless chip design companies, many people fall into a misconception: having IP and a sample board means success. But the harsh truth of the chip industry is that while many can make chips, very few can ship stably at volume and be paid for by real customers. As the third-largest company in China's enterprise AI GPGPU market, whether Denglin Technology can achieve volume shipments is obviously not in doubt. But through several sets of core financial data, perhaps we can see more.
In 2023, Denglin Technology achieved revenue of approximately RMB 28 million; in 2024, revenue was RMB 83 million; in 2025, revenue reached RMB 171 million; and in the first four months of 2026 alone, revenue already reached RMB 200 million, exceeding the full year of 2025. Corresponding GPGPU board shipments also accelerated: 12,144 units shipped in the full year of 2025, and 10,344 units shipped in the first four months of 2026, approaching last year's full-year level. Revenue and shipments accelerating in tandem is not simply "good-looking numbers"; it means orders are no longer sporadic pilots, but customers shifting from "testing and validation" to "volume procurement."
Turning to the profit side, Denglin Technology's improvement in profitability was also quite significant. In 2023, the company recorded a net loss of RMB 332 million; in 2024, a net loss of RMB 297 million; in 2025, a net loss of RMB 281 million; in the first four months of 2026, the company still recorded a net loss of RMB 16 million on its books. After excluding non-cash items such as share-based payment, the company achieved positive adjusted net profit of RMB 20.6 million in the first four months of 2026. R&D expenses during the same period were not compressed either—from 2023 to 2025, the company's R&D expenses remained between RMB 238 million and RMB 258 million. In the first four months of 2026, R&D expenses as a proportion of revenue declined significantly, due to the rapid expansion of the revenue base rather than a reduction in absolute investment.
More critically, the trend of gross margin is key. It rose from 34.4% in 2023 to 49.2% in 2025, and further increased to 50.9% in the first four months of 2026. The structural uplift in gross margin indicates that the company's products are not trading low prices for volume, but entering customers' deployment lists with a premium. This is also the true meaning of operating leverage—that is, the fixed costs previously invested in architecture and software stack are being diluted by rapidly expanding revenue. This coincides with a statement in Denglin's prospectus: "The turnaround from loss to profit was mainly driven by revenue growth and operating leverage effects, rather than by cutting R&D expenses." It can be seen that Denglin's profitability inflection point is one where the revenue side has run ahead of the cost curve.
As of April 30, 2026, Denglin Technology had RMB 449 million in cash on its books and had raised approximately RMB 1.5 billion in cumulative financing, with industrial capital such as Qualcomm China, China Internet Investment Fund, and Hikvision Fund on its shareholder list. Its capital reserves cannot be said to be ample enough to rest easy, but combined with IPO fundraising, they are sufficient to support the iteration window for the next two to three generations of products. It is understood that Denglin Technology expects to raise approximately USD 300 million to USD 400 million, which will be used for iterative upgrades and commercialization of existing GPU product lines, development of new product categories, software toolchain development, and market channel expansion.
From the above performance, it can be seen that Denglin Technology focuses on energy-efficient GPGPU design under a fabless model, and its financial data is validating its transition from "making chips" to "volume delivery." The simultaneous acceleration of revenue and shipments, gross margin rising to 50.9%, and adjusted net profit turning positive all mean that the company's growth and profitability trajectory is releasing positive signals.
Track Growing at 58.1%, Denglin Breaks Through with a Differentiated Route
If viewed in isolation, Denglin's financial inflection point is merely the operational improvement of a small company. Placed in the industry coordinate system, it has stepped onto a track that is rapidly widening. According to Frost & Sullivan data, China's enterprise AI GPGPU market grew from RMB 1.9 billion in 2021 to RMB 11.8 billion in 2025 at a compound annual growth rate of 58.1%. During the same period, China's public cloud AI GPGPU market grew from RMB 21.6 billion in 2021 to RMB 111.8 billion in 2025 at a compound annual growth rate of 50.8%. From 2026 to 2030, the enterprise AI GPGPU market is expected to maintain a high compound annual growth rate of 61.3% and reach RMB 108.5 billion in 2030, while the public cloud AI GPGPU market is expected to grow at a compound annual growth rate of 40.5% and reach RMB 494.2 billion in 2030. Driven by both segments together, China's overall GPGPU market is expected to achieve a compound annual growth rate of 43.3% from 2026 to 2030.
This is a window where both total volume and structure are favorable. A more important structural change is occurring on the demand side. IDC's judgment is that 2026 is a key year for China's computing power demand to transition from "cloud training" to a dual-engine model of "training plus inference." The training market pursues extreme computing power and cluster scale, which is where Nvidia's moat is deepest; the inference market is far more fragmented, and customers are much more sensitive to energy efficiency ratios, deployment costs, private adaptation, and total cost of ownership than to an obsession with peak computing power. The scenarios Denglin focuses on—enterprise private clouds, smart cities, industrial quality inspection, education, and edge video analytics—precisely fall within this fragmented range.
But this track also has very high barriers. The biggest obstacle is often not the hardware itself, but the software ecosystem. After more than a decade of accumulation, the CUDA ecosystem has formed enormous user inertia; for domestic manufacturers, if customer migration costs are too high, even good hardware specifications will find it hard to truly achieve large-scale deployment. This is also why not all chips can reach volume delivery: hardware can catch up, but software stacks, driver libraries, and model adaptation require year after year of "polishing scenarios" together with customers, ISVs, and integrators.
From this perspective, the opportunity in the track is clear: AI inference demand continues to explode, and self-controllability brings structural dividends; but the challenges are equally obvious—there is no shortcut to ecosystem building, and it can only be built bit by bit over time and with implementation cases. Against this development backdrop, Denglin Technology has also carved out its own differentiated position through GPU+ Minsky on-chip heterogeneity.
On the technical route, Denglin's GPU+ (Minsky) architecture integrates programmable general-purpose computing units, tensor acceleration engines, and high-bandwidth on-chip storage into a single chip, replacing traditional discrete multi-chip solutions with on-chip heterogeneity, reducing latency and power overhead from data movement. Paired with its self-developed Hanming software stack, it achieves compatibility with CUDA and mainstream AI frameworks, greatly reducing the code rewrite cost for customer migration. For enterprise customers, "not having to start over" is itself a scarce purchasing reason.
In terms of scenario selection, Denglin's commercialization is concentrated on the inference side and cost-sensitive industry scenarios, rather than general large-model training clusters. Markets such as smart cities, enterprise private clouds, and industrial quality inspection may not have impressive single-order sizes, but their dispersed demand, short decision chains, and heavy weight on localized service precisely give domestic GPUs room to penetrate with cost-effectiveness and close-at-hand service. The company claims to be the first domestic GPU enterprise to achieve large-scale commercial deployment, and its Gaolin and Nashi series have already been mass-produced and introduced by leading enterprises in multiple industries such as the internet and smart cities.
Of course, advantages do not equal victory, and we also need to objectively see its current shortcomings—on the one hand, its overall market share remains low, at about 1.0% of the domestic market based on 2025 revenue, leaving a long way to catch up with industry leaders; on the other hand, software ecosystem building is a protracted war, and compatibility does not equal "complete equivalence." It still needs to continuously invest massive adaptation work to cover more niche industry models.
From the above, it can be seen that Denglin Technology is breaking out through heterogeneous architecture, deeply cultivating the high-growth blue ocean of inference computing power, and using continuously high revenue growth and steadily rising gross margin to prove the commercial viability of its differentiated route. Although its current market share still needs to improve and ecosystem building still requires deep cultivation, industry dividends continue to be released, and the wave of domestic substitution is unstoppable.
Conclusion
As national digital transformation moves from single-point informatization to systematic intelligence in government and enterprises, industry, and urban governance, computing power is no longer just the business of a few companies, but a secure and controllable industrial foundation. The market does not only have room for one or two leading companies; it needs a group of domestic computing power suppliers that can deliver stably, be invoked by real scenarios, and stay close to customers in cost and service.
Denglin's value lies in its use of GPU+ on-chip heterogeneity and the Hanming software stack to enter the fragmented inference-side market, using a "no need to start over" migration experience and close-at-hand service to push domestic GPUs from "technically feasible" to "commercially usable." What it represents is not the financial improvement of a small company, but a realistic sample of multiple routes and multiple scenarios for domestic computing power landing. But listing is only a ticket to a new stage, not the climax of the story. Whether Denglin can truly break out ultimately depends on three things: whether it can continuously deliver hardware with stable performance and controllable yield; whether it can polish its software ecosystem from "compatible" to "easy to use," enabling more models, ISVs, and integrators to adapt at low cost; and whether it can turn scattered projects into replicable industrial orders, forming customer repeat purchases and ecosystem compounding.
Comments