IPO Interpretation | Denglin Technology: Revenue broke 200 million in 4 months, the inflection point ahead of the cost curve?

For a long time, the market was stuck with the impression that “the technology is OK, but commercialization is yet to be verified” for domestic general purpose computing power chips.

However, with Denglin Technology officially opening the Hong Kong Stock Exchange's IPO, a number of domestic computing power chip manufacturers are handing over real answers to technology industrialization based on shipments, revenue, and cash flow.

On October 25, according to the Hong Kong Stock Exchange, Suzhou Denglin Technology Co., Ltd. (hereinafter referred to as “Denglin Technology”) passed the Hong Kong Stock Exchange's main board listing hearing. CITIC Securities, CITIC Construction Investment International, and Bank of China International were co-sponsors. The company officially submitted the 18C Speciality Technology IPO application to the Hong Kong Stock Exchange on February 27. After more than seven months, it finally opened the doors of the Hong Kong Stock Exchange.

Established in 2017, the company is a general purpose graphics processor (GPGPU) provider that develops and sells GPGPU products to meet the AI computing needs of China's Qianxing Industry. According to Frost & Sullivan, in terms of revenue in 2025, Denglin Technology is the third-largest Chinese company in the AI GPGPU market, with a 1.0% market share.

Next, we might as well break out of the simple “technical myth” narrative and take a look at the true characteristics of this heterogeneous GPGPU manufacturer from the two dimensions of business performance and industry pattern, as well as the domestic computing power growth opportunities represented behind it.

Revenue broke 200 million in 4 months, and operating leverage began to be released

According to Zhitong Finance, Denglin Technology mainly focuses on the development and sale of hardware products (especially GPGPU boards) and supplies them to original equipment manufacturers (OEMs), system integrators, and AI solution providers. The company operates under the Fabless business model, which is the mainstream business model for companies focusing on chip innovation and architectural advancement in the semiconductor industry. This model allows the company to focus resources on the design and development of energy-efficient GPGPUs while leveraging the expertise of third party partners to manufacture them.

When it comes to Fabless chip design companies, many people fall into a misunderstanding: having an IP and a sample card equals success. However, the harsh truth of the chip industry is that there are quite a few chips that can be made, can be stably shipped in batches, and are paid for by real customers.

As the third largest Chinese company in the AI GPGPU market, it is clear whether Denglin Technology can form batch shipments.

But through a few sets of core financial data, we may be able to see more.

In 2023, Denglin Technology achieved revenue of about 0.28 million yuan; in 2024, revenue of 83 million yuan; in 2025, revenue reached 171 million yuan; and in the first four months of 2026 alone, revenue reached 200 million yuan, surpassing the full year of 2025. Shipments of corresponding GPGPU boards have also accelerated: 12,144 pieces will be shipped throughout 2025, and 10,344 will be shipped in the first 4 months of 2026, which is close to the level of the full year of last year.

Revenue and shipping are speeding up at the same time. This is not simply a “good number”; it means that orders are no longer a sporadic pilot, but rather that customers are shifting from “testing and verification” to “batch procurement.”

Looking at the profit side, the margin of improvement on the profit side of technology is also quite remarkable. In 2023, the company had a net loss of 332 million yuan, a net loss of 297 million yuan in 2024, and a net loss of 281 million yuan in 2025; in the first 4 months of 2026, the company's books still recorded a net loss of 0.16 million yuan; after excluding non-cash items such as share payments, adjusted net profit for the first 4 months of 2026 was corrected to 20.6 million yuan.

R&D expenses were also not reduced during the same period — from 2023 to 2025, the company's R&D expenses remained between 238 million and 258 million yuan, and the share of R&D expenses in revenue declined significantly in the first four months of 2026 due to a rapid expansion in the revenue base rather than an absolute reduction in investment.

What is more critical is the trend in gross margin. From 34.4% in 2023 to 49.2% in 2025, gross margin increased further to 50.9% in the first four months ending 2026. The structural rise in gross margin indicates that the company's products are not traded at low prices, but are entering customer deployment lists with a premium.

This is also the true meaning of operating leverage — that is, the fixed costs invested in the architecture and software stack in the early stages are being diluted by rapidly expanding revenue. However, this also coincides with a statement in the prospectus that “turning a loss into profit is mainly driven by revenue growth and operating leverage effects, not by cutting R&D expenses.” As can be seen, the profit inflection point is that the revenue side is ahead of the cost curve.

As of April 30, 2026, Denglin Technology's book cash was 449 million yuan, with a cumulative financing of about 1.5 billion yuan. The shareholder list includes industrial capital such as Qualcomm China, China Internet Investment Fund, and Hikvision Fund. Capital reserves are not sufficient enough to rest easy, but cooperating with IPOs to raise capital is enough to support the iteration window for the next two to three generations of products. According to reports, Denglin Technology is expected to raise about 300 million to 400 million US dollars, which will be used for iterative upgrading and commercialization of existing GPU product lines, development of new product categories, software tool chain development, and market channel expansion.

From the various performances described above, Denglin Technology focuses on energy-efficient GPGPU design using the Fabless model, and financial data is verifying that it is moving from “making chips” to “batch delivery.” However, performance such as the simultaneous acceleration of revenue and shipment, a rise in gross margin to 50.9%, and a correction in adjusted net profit also mean that the company's growth and profit trajectory are already sending positive signals.

The number of racetracks increased by 58.1%, and the breakthrough was achieved through differentiated routes

If you look at the financial inflection point in isolation, it's just an improvement in the operation of a small company. In the industry's coordinate system, it is stepping on a racetrack that is getting wider at an accelerated pace.

According to Frost & Sullivan data, the Chinese enterprise AI GPGPU market increased from 1.9 billion yuan in 2021 to 11.8 billion yuan in 2025 at a compound annual growth rate of 58.1%. Over the same period, China's public cloud AI GPGPU market increased from 21.6 billion yuan in 2021 to 111.8 billion yuan 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 CAGR of 61.3% and reach 108.5 billion yuan in 2030, while the public cloud AI GPGPU market is expected to grow at a CAGR of 40.5% and reach 494.2 billion yuan in 2030.

Driven by the two segments, China's overall GPGPU market is expected to achieve a CAGR of 43.3% between 2026 and 2030. This is a window where both overall and structural benefits are beneficial.

More significant structural changes occurred on the demand side. According to IDC's judgment, 2026 is a critical year for China's computing power demand to shift from “cloud training” to “training+inference” two-wheel drive. The training market pursues extreme computing power and cluster size, and is the deepest part of Nvidia's moat; the inference market is much more fragmented, and customers are far more sensitive to energy efficiency, deployment costs, privatization adaptation, and total cost of ownership than their obsession with peak computing power. The focused scenarios of enterprise private clouds, smart cities, industrial quality inspection, education, and edge video analysis fall within this fragmented range.

But this track also has a very high threshold.

The biggest obstacle is often not the hardware itself, but the software ecosystem. After more than ten years of accumulation, the CUDA ecosystem has formed huge user inertia; for domestic manufacturers, if customer migration costs are too high, even if the hardware parameters are good, it is difficult to actually implement it on a large scale. This is why not all chips can be delivered in batches: hardware can catch up, but software stacks, driver libraries, and model adaptations take year after year to “polish the scenario” with customers, ISVs, and integrators.

From this perspective, the opportunities on the racetrack are clear: the demand for AI reasoning continues to explode, and autonomy and control bring structural dividends; but the challenge is also on the surface — there are no shortcuts to ecological construction; it only takes time and implementation cases to pile up little by little.

In this context of development, Denglin Technology has also opened up its own differentiated position through GPU+ Minsky's in-chip heterogeneity.

In terms of technology, Denglin's GPU+ (Minsky) architecture integrates a programmable general-purpose computing unit, tensor acceleration engine, and high-bandwidth on-chip storage into a single chip, replacing traditional multi-chip discrete solutions with in-chip heterogeneity, reducing data handling delays and power consumption expenses. With the self-developed Hanming software stack to achieve compatibility with CUDA and mainstream AI frameworks, the cost of code rewriting for customer migration has been drastically reduced. For enterprise-level customers, “no need to do it again” is itself a scarce reason to buy.

In terms of scenario selection, Dengjin's commercial implementation focuses on inference and cost-sensitive industry scenarios, rather than general large model training clusters. The scale of a single order in the smart city, enterprise private cloud, and industrial quality inspection markets is not necessarily impressive, but scattered demand, short decision-making chains, and heavy emphasis on localized services have just given domestic GPUs room for cost performance and personal service penetration. The company claims that it is the first GPU company in China to achieve large-scale commercial implementation. The Gao Lin and Nasi series have been mass-produced and introduced by leading companies in various industries such as the Internet and smart cities.

Of course, advantage does not equal victory. We also need to objectively look at its shortcomings at this stage of development — on the one hand, the overall market share is still not high. According to the 2025 revenue scale, the domestic market share is about 1.0%, which is still a long way to catch up; on the other hand, software ecosystem construction is a protracted battle. Compatibility is not equal to “complete equality”, and we still need to continue to invest in massive adaptation work to cover more niche industry models.

As can be seen from the above, Denglin Technology has broken through the misalignment of a heterogeneous architecture, deepened the blue ocean of reasoning power, and confirmed the commercial viability of the differentiated route with continued high revenue growth and a steady rise in gross profit margin. Even though the current market share still needs to be increased and ecological construction still needs to be cultivated, industry dividends continue to be released, and the wave of domestic substitution is unstoppable.

epilogue

The country's digital transformation has moved from single-point informatization to systematic intelligence in government enterprises, industry, and urban governance. Computing power is no longer just the business of a few companies, but the foundation of a safe and controlled industry. The market is not the location of only one or two leading companies; it requires a group of domestic computing power suppliers that can deliver stably, can be used in real scenarios, and are close to customers in terms of cost and service.

The value of Dengjin lies in the fact that it uses GPU+ in-chip heterogeneity and Hanming software stack to advance domestic GPUs from “technically feasible” to “commercially usable” using a migration experience and personal service with “no need to overturn”; it does not represent a financial improvement for a small company, but a realistic sample of domestic computing power implemented in multiple routes and multiple scenarios.

However, the listing is only a ticket to a new stage; it is not the climax of the story. Whether Denglin can actually come out depends on three things: whether it can continue to deliver hardware with stable performance and controlled yield; whether it can polish the software ecosystem from “compatible” to “easy to use” so that more models, ISVs, and integrators can adapt at low cost; and whether it can turn scattered projects into replicable industrial orders to form customer repurchases and ecological compound benefits.