The “Token Factory” layout has exceeded expectations, and the value of Paradigm Intelligence (06682) at the center of the card slot industry is about to be re-evaluated
Paradigm Intelligence (06682) has grown into a leading third-party “token factory” in China, or has ushered in a good long-term layout opportunity. The key clue to verify this judgment is hidden in its financial report for the first half of 2026 released on August 20.
According to the data, the company's revenue during the reporting period reached 3,765 billion yuan (RMB, same below), up 43.4% year on year, up about 8 percentage points from 35.4% in the first quarter, and revenue growth accelerated again; adjusted net profit to mother was about 150 million yuan, turning a year-on-year loss into profit (loss of 43.7 million yuan for the same period in 2025), and this figure was 8.4 times the adjusted net profit to mother for the full year of 2025.
Obviously, after successfully crossing the break-even point in 2025, Paradigm Intelligence entered a new stage of growth of “accelerated revenue and a huge explosion in profits” in 2026. The underlying logic behind this strong momentum is that after continuous upgrading and iteration of the Paradigm Intelligence Development Strategy, the company has already accelerated into the harvest period.
In 2025, the company completed a strategic upgrade from the “fourth paradigm” to the “paradigm group”, completely transitioned from a “single enterprise-level AI platform” to a “full-stack AI ecosystem”, and officially entered the AI 2.0 era. Based on this, Paradigm Intelligence established the “Token Factory” as the core engine for the company to drive commercial monetization in 2026. The superposition effect of the two strategic upgrades has opened up a new situation with Paradigm Intelligence.
This mid-term financial report is undoubtedly a very persuasive “litmus test”. The paradigmatic “Token Factory” layout has achieved substantial fulfillment beyond expectations on the performance side, and has become the core growth pole driving the company's fundamentals to explode.
The Zhitong Finance App believes that based on the “Token Factory” layout of Paradigm Intelligence, the strong implementation of the “Token Factory” layout on the performance side and the strong growth potential shown, the true value of Paradigm Intelligence is expected to usher in a systematic revaluation in the capital market.
The revenue growth rate reached a new high in the listing, and R&D investment continued to be high at a time of qualitative changes in profit
Looking at this financial report with a paradigm of intelligence, many details that are easily overlooked by the market are the key to highlighting the “gold content” of this financial report. One is the revenue growth rate of 43.4% on the revenue side, which has set the highest level of revenue growth in the first half of the year since Paradigm Smart was launched in 2023.
The core that supported the record revenue growth rate was the collaborative effort of the company's three major businesses, particularly the API business. The performance during the reporting period was particularly impressive. Since 2026, large-scale implementation of AI applications represented by smart devices has led to a blowout in token demand. Paradigm Intelligence relies on core technical capabilities such as Hami vGPU and ModelHub to build a full-link capability system covering computing power scheduling, model adaptation, token production and delivery, and can produce more and better quality tokens under the same computing power conditions, thus firmly grasping the industry dividends.
During the reporting period, Paradigm Intelligence's token call volume increased by about 620% year on year. This led to the company's API business revenue surging 860.8% year over year to about 464 million yuan. The business's share of total revenue increased from 1.8% in the same period in 2025 to 12.3%. The shift in revenue share shows that the API business has become a strong new growth curve for the company. This is also an important support for paradigm smart revenue surpassing market expectations.
As the basic platform of paradigm intelligence, the AI platform business achieved rapid growth during the reporting period and continued to play an important role as a “ballast stone.” Thanks to the explosion in domestic demand for heterogeneous computing power, high-value orders for paradigm intelligent computing power management continued to land, and the number of paying customers using RISE vGPU (HaMi Commercial Edition) as the primary requirement increased by about 5 times over the previous year, which led to a 30.0% year-on-year increase in revenue to 3.088 billion yuan.
At the same time, the increase in the share of high-value orders has led to a stabilization of gross margin, highlighting the high quality of growth. Furthermore, as of June 30, 2026, the business's active orders have reached 7 billion yuan, laying a solid foundation for continued high growth in the future.
As a key “gripper” for paradigmatic intelligence to explore high-value fields, the Agentic AI business achieved steady development during the reporting period, recording revenue of 214 million yuan, an increase of about 4.9% over the previous year. In this business, Paradigm Intelligence continues to focus on strategies, prioritizes investment in industry scenarios with high strategic value and high quality professional data, and implements promotion according to the “pay-for-results” model.
Currently, Paradigm Intelligence's Agentic AI business is deeply cultivated on the circuit with new energy power as the core, and in the first half of 2026, it is focusing on and actively laying out the film and television entertainment sector to quickly complete scene implementation and data accumulation, making corresponding reserves for the medium- to long-term core growth of the business.
Second, there has been a qualitative leap in profit quality, and full profit in the true sense of the word has been achieved for the first time. This is another key highlight of the current financial report. Looking back at 2025, although the company has achieved a correction in net profit due to adjustments and initial verification of the operating profitability of the main business, net book profit is still in a state of loss due to non-cash projects such as equity incentives, acquisitions and amortization.
By the first half of 2026, this situation was completely reversed: net profit for the period was recorded at 128 million yuan (loss of 73.894 million yuan for the same period in 2025), and adjusted net profit to mother both turned a loss into a profit. This means that the rapid growth of the main business has completely absorbed the reporting impact of non-cash projects. Paradigm Intelligence not only “makes money from operations”, but also “makes money on books,” completing a historic leap from “adjusted profit” to “full profit.” This is not only the most important inflection point for growing companies, but also marks that the company has officially entered a virtuous cycle of “self-hematopoiesis.”
However, there are three main reasons for the overall improvement in profit quality: scale effects brought about by rapid growth in main business; drastic optimization of sales expenses; and a sharp year-on-year increase in other net earnings. Among them, the drastic optimization of sales expenses is worthy of investors' attention.
During the reporting period, Paradigm Intelligence's sales and marketing expenses were 283.37 million yuan, a year-on-year decrease of 85% compared with 189 million yuan in the same period in 2025, and the sales expense ratio dropped to 0.75%. This is mainly due to the “Token Factory” strategy driving the financial implementation of the transformation of the business model from “reseller” to “product-driven+platform customer acquisition”. As business models change, the low cost rate of paradigm intelligence may be expected to become the norm.
Third, adhere to the long-term principle and build long-term competitive barriers with high-intensity research and development. There is no shortage of companies in the market that release current profits by adjusting R&D expenses, but this is not the case with paradigm intelligence. The drastic improvement in the company's profit quality is based on intensive R&D expenditure. During the reporting period, the company spent 1,208 billion yuan on R&D, a year-on-year increase of 35.2%, accounting for 32.1% of total revenue.
This high level of R&D expenditure proves that Paradigm Intelligence has the ability to “accelerate the construction of a technological moat” and “guarantee the rapid release of profits.” This positive cycle of “profit feeds back R&D and R&D drives the next round of growth” is the healthiest growth paradigm for growing companies, and also provides the most solid support for the company's long-term value.
The strategic card position is the center of operation of the industry, and the three major barriers build a long-term moat
The “Token Factory” layout of Paradigm Intelligence is not a single product, but rather a set of capabilities that rely on the two core technologies of Hami vGPU and ModelHub, covering the full link of computing power scheduling, model adaptation, and token production and delivery. It has become a core hub connecting computing power vendors, model manufacturers, and enterprise users, thus opening up a closed loop in the entire value chain from heterogeneous computing power and models to enterprise applications.
The subtlety of this layout is that it avoids the most popular battlefields. It neither uses upstream computing power hardware nor the midstream general-purpose large model base. Instead, precision cards are located in the middle scheduling and adaptation layer with the greatest value between “heterogeneous computing power” and “multiple models” and can be extended naturally. This is the center of the operation of the industry.
Based on this key card position, the company follows a differentiated route downstream. Instead of being a low-value general agent or pure project outsourcing, the company focuses on industry scenarios with high strategic value and can accumulate high-quality professional data through Agentic AI, and uses a “pay-for-results” model to achieve a closed loop from infrastructure to industry monetization.
Based on this complete layout, the differences between paradigm intelligence and the positioning of large model vendors and traditional cloud vendors are clear at a glance: the company chose to be a neutral AI infrastructure service platform, which neither binds to a single model nor simply sells computing power, but instead provides one-stop services from computing power virtualization, model adaptation, intelligent routing to security compliance, to help enterprises truly solve the cost and efficiency pain points in multi-model implementation.
In fact, the fulfillment of the Paradigm Intelligence “Token Factory” strategy beyond expectations on the performance side is only an initial release of this layout value. The three key barriers it forms at the AI industry chain level are the company's moat and the core anchor for long-term value.
The first barrier lies in the leading “unified scheduling of heterogeneous computing power” technology. When chips such as Nvidia, Shengteng, Cambrian, Haiguang, Wall Mount, and Mohr Threads coexist, the biggest pain points faced by enterprises are different chip architectures, fragmented computing power, and utilization rates of less than 30%.
Paradigm Intelligence enters this portal through Hami vGPU, supports fine GPU segmentation and on-demand allocation. Computing power control is accurate to 1%, and video memory allocation is accurate to the MB level, increasing the industry average GPU utilization rate of less than 30% to the 70%-90% range.
More importantly, Hami of Paradigm Intelligence began R&D in 2018, and has continued to accumulate technology for 8 years and has entered the CNCF incubation stage. This engineering accumulation is difficult to replicate by increasing R&D resources in the short term. This means that latecomers can't keep up with this time window even if they want to do heterogeneous scheduling. Paradigm Intelligence has a significant first-mover and leading edge.
The second barrier is being the standard-setter for “model x chip” bidirectional adaptation. This is also the most competitive part of Paradigm Intelligence. The core bottleneck in large-scale AI implementation is the “multi-chip x multi-model” adaptation problem, and ModelHub (Xinchuang Mold Box) has accumulated more than 200,000 “model x chip” adaptation combinations, and the certified models have surpassed 100,000 models, and are deeply compatible with 9 mainstream domestic chip platforms such as Huawei Ascend, Moore Thread, Haiguang, Cambrian, and Wall Zao. The operator library coverage rate is over 95%, and breaking through the company's clear guidelines will reach the level of HuggingFace, the world's largest model community.
This means that the paradigm smart card is in a standard interface position where the “model and chip go both ways”. The world's mainstream open source models run on domestic computing power. ModelHub is currently the most widely adapted community. This kind of “adaptation layer” card slot gives it an ecosystem similar to “Android in the AI era.”
The third barrier is the commercial closed-loop exit of “token production and delivery+high-value scenario operation and monetization”. The three major businesses of Paradigm Intelligence form a three-dimensional matrix of “one token factory base × three types of delivery forms × three types of business models”, using technology to share dilution costs, strengthen barriers through data sharing, hierarchical customer expansion of LTV, and commercial complementarity to optimize financial structures. This is the core advantage that distinguishes AI companies with a single product line.
At the technical level, the three major businesses share the underlying capabilities of the “Token Factory”: AI Platform packaging and privatization solutions, API pay-as-you-go billing exposed to the outside world, and Agentic AI superimposes industry know-how to achieve “pay-for-results”. Every time the base is optimized, the computing power efficiency of the three lines benefits simultaneously with model selection. “One R&D, three monetization” allows R&D output to have a leverage effect far exceeding that of a single product line.
At the data level, the three business lines stream customer workload characteristics, token call logs, and vertical scene professional data back to the same platform to continuously feed back ModelHub's adaptation combination and HAMI scheduling strategy, forming a positive cycle of “more use → more accurate → more efficient → attract more customers”. The stronger the platform, the simultaneous increase in business line competitiveness.
At the customer level, the three major segments cover progressive requirements from AI platform privatization and flexible API calls to agentic AI embedded in core business processes to lock pricing rights. This path of “selling products → selling tokens → selling results” allows Paradigm Intelligence to accumulate more than 1,000 contracted customers from a static inventory to a “reservoir” that continuously releases value — the progression of each layer of business is expected to drive customers from low-frequency product procurement to high-frequency token consumption, and even eventually or can be settled into high-value long-term results. The lifetime value of a single customer is unlocked and continuously expanded.
The dividends of the three major industries are being released at an accelerated pace, and there is strong certainty that they will continue to grow at a high level
With a strategic card position in the middle of a high-value industry called “scheduling and adaptation,” and a long-term moat built with three major barriers, paradigm intelligence has become the most scarce “shovel seller” in the Token era — instead of personally training a general-purpose model “gold digger,” it provides computing power scheduling, model adaptation, and token production and delivery capabilities for all “gold diggers”.
This “no gold mining, just selling shovels” model is the most tested ecological position in the entire AI industry chain. Furthermore, as the explosion of smart devices drives a surge in demand for tokens, the accelerated implementation of domestic computing power adaptation, and the shift of computing power demand to the centralized release of dividends in multiple industries dominated by reasoning, the true value of paradigm intelligence will become more prominent, and the company's continued high growth is highly certain.
Currently, the smart device explosion is fundamentally reshaping the token demand curve. According to Anthropic's actual measurements, the token consumption for a single agent to complete a typical task is about 4 times that of normal conversation mode, and up to 15 times that of multi-agent collaboration systems. Agents have driven a significant increase in token consumption intensity.
More importantly, Gartner predicts that 40% of global enterprise applications will embed task-based AI agents by the end of 2026, compared to less than 5% in 2025. The number of agents and the intensity of token consumption expand nonlinearly at the same time, making the blowout in token demand a continuous, structural driving force.
With the full-link cost advantage built by Hami vGPU and ModelHub, Paradigm Intelligence occupies the most favorable position in the blowout in token demand, and the company expects the annual API revenue to reach 20 times that of the whole year of 2025. Full link capability enables the same computing power to achieve higher output and better unit economy, driving the profit margin of the API business into an upward cycle. Long-term framework orders are sufficient, and the business has accelerated into a large-scale explosion stage.
The accelerated penetration of intelligent bodies has opened up space for Agentic AI to grow. The real industry data accumulated by Paradigm Intelligence in high-value scenarios is laying the foundation for large-scale replication from energy, film and television to manufacturing, finance, retail, etc. — this is a key step for “sellers” to extend from infrastructure to industry monetization and lock in long-term pricing power. As large-scale replication accelerates, the Agentic AI business will enter a growth acceleration channel.
At the level of domestic computing power adaptation, the share of domestic AI chips jumped to 41% in 2025, supporting over 60% of intelligent computing demand in the first half of 2026. The release of the head model triggered “Day 0” level adaptation of many domestic chips, and the inflection point for large-scale commercial use has reached. However, the “combo explosion” where more than ten GPUs were adapted one by one also brought sharp pain points: the average utilization rate of GPUs in intelligent computing centers was less than 30%, cross-chip scheduling losses were as high as 30%, and nearly 99% of the HuggingFace open source model could not run directly on domestic cards.
However, Paradigm Smart Hami vGPU+ModelHub XC's two engineering barriers built up over eight years just happened to be stuck in this much-needed path. Paradigm Intelligence has become a “standard interface” that no model manufacturer or computing power manufacturer can bypass as domestic computing power moves from “usable” to “easy to use”. Every time a new model is released and every new chip is mass-produced, a new portfolio asset is added to the ModelHub XC adapter library; and the construction of every domestic intelligent computing center requires HaMI to “fully utilize, manage, and run steadily” computing power. The more prosperous the domestic computing power ecosystem, the more irreplaceable its middle tier card slots are, and the deeper the moat.
Furthermore, the computing power demand structure is being irreversibly switched from “training led” to “inference led.” Deloitte expects the global inference load to rise from about 1/3 of AI computing power in 2023 to about 2/3 in 2026. Nvidia also predicts that the potential market size of AI inference chips will reach 1 trillion US dollars in 2027, and the dominance of computing power inference demand has become an inevitable trend.
For paradigm intelligence, this shift just pushes it into the most comfortable value range. The reasoning comparison is “who can optimally schedule heterogeneous computing power and keep the cost per token to the minimum”. This is the home of Hami vGPU, so that every bit of computing power purchased can be converted into actual token output. When reasoning becomes the absolute main force in computing power consumption, the seller value of “heterogeneous scheduling and efficiency improvement” of paradigm intelligence will be indispensable. The high on-hand order of 7 billion yuan is the best proof of this logic.
Summarize:
Entering 2026, the capital market's pricing logic for AI companies is undergoing a profound restructure—from a “belief narrative” to a “performance test.” Over the past two years, the market has generously paid for ambitious AI visions and huge capital expenses; now, capital has begun to use a magnifying glass to examine the quality of each output, and commercial cashability has become the core yardstick for distinguishing the valuation hierarchy of AI companies. In this aesthetic transition, capital is no longer limited to the story; it only makes way for fulfillment.
Paradigm intelligence stands precisely on the most beneficial side of this round of valuation system switching. When Chunda Model Company was questioned due to huge training costs, and pure cloud vendors faced cash flow tests due to heavy asset investment, Paradigm Intelligence proved its true ability to monetize the immediate aspects of “connection and scheduling” with the quadruple combination of “revenue structure upgrade+comprehensive profit verification+key strategic card positions+triple barriers”.
This kind of middle-tier infrastructure company that “can stand up to settlement” is the most scarce ecosystem in the entire AI industry chain and the easiest to obtain a valuation premium. When tokens became a value anchor in the intelligent era, paradigm intelligence was an “invisible hub” connecting computing power, models, and enterprise applications — its true value is gradually being re-evaluated by the market as industry dividends are released. Paradigm intelligence with continuous high growth and strong certainty is already the core beneficiary target of AI in the inevitable trend of moving from “technology verification” to “large-scale commercialization.”