The Token Factory layout has exceeded performance expectations, and the value reassessment of PHANCY (06682), positioned at the nerve center of the industry, is imminent.

PHANCY (06682), which has grown into a leading domestic third-party "Token factory," may have welcomed an opportunity for long-term strategic development. The key clues to verify this judgment are hidden in its interim financial report for the first half of 2026, released on August 20. According to the data, during the reporting period, the company generated revenue of 3.765 billion yuan (RMB, the same below), a year-on-year increase of 43.4%, with the growth rate improving approximately 8 percentage points from the 35.4% in the first quarter, indicating a renewed acceleration in revenue growth. The adjusted net profit attributable to shareholders was about 150 million yuan, turning a profit from a loss of 43.7 million yuan in the same period of 2025, and this figure is already 8.4 times the adjusted net profit attributable to shareholders for the full year of 2025. Clearly, after successfully crossing the break-even point in 2025, PHANCY has entered a new growth phase in 2026 characterized by "accelerating revenues and explosive profits." The underlying logic behind this strong momentum lies in the company's continued iterative elevation of its development strategy, which has accelerated its entrance into a harvest phase. In 2025, the company completed a strategic upgrade from "Fourth Paradigm" to "Paradigm Group," transitioning from a "single enterprise-level AI platform" to a "full-stack AI ecosystem," officially entering the AI 2.0 era. Building on this foundation, PHANCY will position its "Token factory" as the core engine driving commercial monetization in 2026. The combined effects of two strategic elevations have opened up a completely new situation for PHANCY. This interim financial report is undoubtedly a persuasive "touchstone," as PHANCY's "Token factory" layout has achieved substantial and exceeding expectations in terms of performance, becoming a core growth engine driving the company's fundamental breakout. It is believed that based on PHANCY's strong performance in its "Token factory" layout and the demonstrated strong growth potential, the companys true value is likely to undergo systematic reassessment in the capital market. Revenue growth has reached a new high since the listing, with a continuous high level of R&D investment during the transformation of profit quality. Looking at PHANCY's financial report, several details that might be easily overlooked by the market underscore the key "gold content" of this report. One such detail is the revenue growth rate of 43.4%, which marks the highest level of revenue growth for PHANCY in the first half since its listing in 2023. The core support for achieving this record revenue growth lies in the collaborative force of the company's three major business segments, particularly the API business, which performed exceptionally well during the reporting period. Since 2026, the scaling of AI applications represented by intelligent agents has driven an explosive surge in Token demand. Leveraging core technical capabilities such as HAMi vGPU and ModelHub, PHANCY has built a comprehensive capability system covering power scheduling, model adaptation, Token production, and delivery, enabling more and higher-quality Tokens to be produced under the same computing power conditions, thus firmly grasping the industry dividends. During the reporting period, PHANCY's Token call volume increased by approximately 620% year-on-year, driving the company's API business revenue to skyrocket by 860.8% year-on-year to approximately 460 million yuan, which increased the proportion of this business from 1.8% in the same period of 2025 to 12.3%. This leap in revenue proportion indicates that the API business has become a strong new growth curve for the company, also serving as a crucial support for PHANCY's significant revenue surpassing market expectations. As the foundational business of PHANCY, the AI Platform achieved rapid growth during the reporting period, continuing to play an important role as a "ballast stone." Benefiting from the explosive demand for domestic heterogeneous computing power, PHANCY has continuously secured high-value orders in computing power management, resulting in a year-on-year increase of approximately five times in the number of paying clients who prioritize RISE vGPU (HAMi commercial version), which drove the revenue of this business to grow 30% year-on-year to 3.088 billion yuan. At the same time, the increase in the proportion of high-value orders helped stabilize the gross profit margin, highlighting the high quality of growth. By June 30, 2026, this business had an order backlog of as much as 7 billion yuan, laying a solid foundation for sustained high growth in the future. The Agentic AI business, as PHANCY's key leverage for exploring high-value sectors, experienced steady growth during the reporting period, recording revenue of 214 million yuan, a year-on-year increase of approximately 4.9%. In this business, PHANCY maintained a focused strategy, prioritizing investment in industry scenarios with high strategic value and the potential to accumulate high-quality professional data, promoting it on a "results-based payment" model. Currently, PHANCY's Agentic AI business is focusing on the new energy sector and has made major efforts in the film and entertainment sector in the first half of 2026, rapidly completing scene implementation and data accumulation, making relevant reserves for the long-term core growth of this business. Another significant highlight of this financial report is the qualitative leap in profit quality, as the company achieved a comprehensive profit for the first time in a genuine sense. Looking back at 2025, although the company had achieved a positive adjusted net profit attributable to shareholders, initially validating the operating profitability of its main business, its net profit was still in the red due to non-cash items like stock incentives and acquisition amortization. However, in the first half of 2026, this situation was completely rewritten: the net profit for the period reached 128 million yuan (compared to a loss of 73.89 million yuan in the same period of 2025), and both the net profit and adjusted net profit attributable to shareholders turned to profit. This means that the rapid growth of the main business has fully absorbed the impact of non-cash items on the financial statements, showing that PHANCY is not only "earning money operationally" but also "earning money on the books," achieving a historic leap from "adjusted profit to comprehensive profit." This is not only the most crucial turning point for a growth-oriented company, but it also marks the companys formal entry into a positive cycle of "self-financing." The comprehensive improvement in profit quality is primarily due to three factors: the scale effect brought by rapid growth in the main business; significant optimization of sales expenses; and a sharp year-on-year increase in other income net. Among these, the significant optimization of sales expenses is particularly noteworthy to investors. During the reporting period, PHANCY's sales and marketing expenses amounted to 28.337 million yuan, a staggering reduction of 85% compared to 189 million yuan in the same period of 2025, with the sales expense ratio declining to 0.75%. This is mainly due to the "Token factory" strategy driving a transformation of the business model from "heavy sales with manpower" to "product-driven + platform-acquired customers." With this switch in business model, PHANCY's low expense ratio is expected to become the norm. Furthermore, the company adheres to a long-term vision, building long-term competitive barriers through high-intensity R&D investment. The market is not short of companies that adjust R&D spending to release current profits, but PHANCY has not done so; the significant improvement in profit quality has been based on high-intensity R&D spending. During the reporting period, the company's R&D expenditure was 1.208 billion yuan, a year-on-year increase of 35.2%, accounting for 32.1% of total revenue. This high level of R&D expenditure proves that PHANCY has the capability to simultaneously accelerate the construction of competitive technological moats and ensure rapid profit release. This virtuous cycle of "profits feeding back into R&D, and R&D driving the next round of growth" is the healthiest growth model for a growth-oriented company and provides the strongest support for the company's long-term value. Strategically positioning itself at the operational center of the industry, PHANCY has built three major barriers reinforcing its long-term competitive moat. PHANCY's "Token factory" layout is not a single product but a capability system supported by two core technologiesHAMi vGPU and ModelHubcovering the entire process of power scheduling, model adaptation, Token production, and delivery, thereby forming a core hub that connects computing power vendors, model vendors, and enterprise users, thus establishing a closed value loop from heterogeneous computing, models to enterprise applications. The brilliance of this layout is that it avoids the most competitive battlefield; it does not compete upstream for computing power hardware nor in mid-stream for generic large model foundations, but instead positions itself precisely between "heterogeneous computing power" and "multi-model" where the value is maximized and extends naturally to the middle scheduling and adaptation layer, which is the center of industry operations. Based on this critical position, the company adopts a differentiated route downstream, avoiding low-value generic agents or purely project-based outsourcing. Instead, it focuses on high-strategic-value industry scenarios through Agentic AI that can accumulate high-quality professional data, realizing a closed loop from infrastructure to industry monetization through a "results-based payment" model. It is precisely because of this complete layout that the positioning differences between PHANCY and major model vendors and traditional cloud vendors become clear: the company chooses to be a neutral AI infrastructure service platform, neither binding to a single model nor merely selling computing power but offering a comprehensive service that includes computing power virtualization, model adaptation, intelligent routing, and security compliance, effectively solving the cost and efficiency pain points in the deployment of multiple models for enterprises. In fact, the exceeding expectations of PHANCY's "Token factory" strategy in terms of performance is merely the preliminary release of the value of this layout. The three key barriers it forms at the AI industry chain level are the core anchors of the company's moat and long-term value. The first barrier is the leading technology of "unified scheduling of heterogeneous computing power." When chips from NVIDIA, Ascend, Cambricon, Hygon, Biren, and Moore Threads coexist, the greatest pain point for enterprises is the different chip architectures, fragmented computing power, and utilization rates below 30%. PHANCY enters this entry point through HAMi vGPU, which supports precise GPU segmentation and on-demand allocation, with power control accurate to 1% and memory allocation precise to the MB level, increasing industry average GPU utilization from less than 30% to a range of 70% to 90%. More critically, PHANCY's HAMi has been under development since 2018, with eight years of continuous technical accumulation, and has entered the CNCF Incubating stage. This level of engineering accumulation is difficult to replicate through short-term increases in R&D resources, meaning that even if latecomers wish to engage in heterogeneous scheduling, they will not catch up to this time window, giving PHANCY a significant first-mover and leading advantage. The second barrier is its position as a standard-setter for "model chip" bidirectional adaptation, which is also the most competitive aspect of PHANCY. The core bottleneck for the large-scale landing of AI lies in the adaptation issue of "multi-chip multi-model," and ModelHub has accumulated over 200,000 combinations of "model chip," certifying more than 100,000 models deeply adapted to nine major domestic chip platforms including Huawei Ascend, Moore Threads, Hygon, Cambricon, and Biren, covering over 95% of the operator library, with the company clearly indicating that it will break one million next year, reaching a scale comparable to that of the world's largest model community, Hugging Face. This means that PHANCY has positioned itself at the standard interface for the "dual approach of models and chips," with mainstream global open-source models needing to run on domestic computing power, and ModelHub currently being the most widely adapted community. This positioning in the "adaptation layer" has granted it a role analogous to "Android in the AI era." The third barrier is the business closed-loop export of "Token production and delivery + high-value scenario operational monetization." PHANCY's three major businesses form a three-dimensional matrix of "one Token factory foundation three forms of delivery three layers of customer needs three types of business models," utilizing shared technology to dilute costs, reinforcing barriers through data co-transformation, layer-by-layer customer segmentation to amplify LTV, and optimizing financial structures through commercial complementarity. This distinctly represents the core advantage over AI companies with single product lines. From a technological perspective, the three major businesses share the underlying capability of the "Token factory": AI Platform encapsulates privatization solutions, API is exposed on a pay-per-use basis, and Agentic AI incorporates industry know-how for a "results-based payment." Every optimization of the foundation benefits both the computing efficiency and model selection of all three lines, creating a leverage effect far exceeding that of a single product line: "one R&D effort, three forms of monetization." From a data perspective, the three lines of business route customer workload characteristics, Token call logs, and specialized data from vertical scenarios back to the same platform, continually feeding back into the ModelHub adaptive combinations and HAMi scheduling strategies, forming a positive cycle of "the more used the more accurately adapted the higher efficiency attracting more customers." The stronger the platform, the higher the competitiveness of the business lines. From the customer perspective, the three major areas cover progressive demands from AI Platform private deployment, API flexible invocation to Agentic AI embedding core business processes securing pricing power. This pathway of "selling products selling Tokens selling results" transforms PHANCY's cumulative total of over 1,000 signed clients from a static stock into a continuously releasing value "reservoir"every progressive layer of business is expected to push the customer from low-frequency product purchases to high-frequency Token consumption, even potentially solidifying into high-value long-term revenue shares. Hence, the lifecycle value per client will be progressively unlocked and continuously amplified. The acceleration of three major industry dividends creates a strong certainty for sustained high growth. With its strategic position in the high-value intermediate layer of "scheduling and adaptation" and the long-term moat constructed by the three significant barriers, PHANCY has become the most scarce "shovel seller" in the Token eranot personally training general large models for "gold mining," but instead providing computing power scheduling, model adaptation, and Token production delivery capabilities to all "mining operators." This "not mining, only selling shovels" model is exactly the ecological niche in the entire AI industry chain that withstands scrutiny. With the explosive growth of agents driving surging Token demand, accelerating domestic computing adaptation, and the transition of computing needs toward inference-driven multiple industry dividends, PHANCYs true value will increasingly become evident, and the companys continued high growth will possess strong certainty. Currently, the explosive growth of intelligent agents is fundamentally reshaping the Token demand curve. Anthropic's tests indicate that a single agent consumes about four times the number of Tokens needed to complete a typical task compared to a regular dialogue mode, while multi-agent collaborative systems could reach as high as 15 times, significantly increasing the intensity of Token consumption driven by agents. More critically, Gartner projects that by the end of 2026, 40% of global enterprise applications will embed task-oriented AI agents, compared to less than 5% in 2025. The simultaneous nonlinear expansion of both the number of agents and Token consumption intensity suggests that the explosive growth in Token demand will have a sustained, structural DRIVE. Leveraging the full-link cost advantages built around HAMi vGPU and ModelHub, PHANCY is poised to occupy the most advantageous position to realize its Token demand surge. The company expects full-year API revenue to reach 20 times that of the entire year of 2025. The full-link capability enables higher output and superior unit economics under the same computing power conditions, propelling the API business profits into an upward cycle, with ample long-term framework orders, as this business accelerates towards a scaled explosion. The accelerated penetration of intelligent agents opens up growth space for Agentic AI. The real industry data that PHANCY has accumulated in high-value scenarios is laying the foundation for large-scale replication across more industries from energy and film to manufacturing, finance, and retailthis is precisely the key step for the "shovel seller" to extend from infrastructure to industry monetization and secure long-term pricing power. With the acceleration of large-scale replication, the Agentic AI business will enter a growth acceleration channel. In terms of domestic computing adaptation, the share of domestic AI chips surged to 41% in 2025, supporting over 60% of intelligent computing demand in the first half of 2026, triggering "Day 0" level adaptation for multiple domestic chips upon the release of leading models. The tipping point for large-scale commercialization has arrived. However, the "combinatorial explosion" from adapting more than a dozen GPUs to vast models also brings acute pain points: the average utilization rate of GPUs in intelligent computing centers is less than 30%, with inter-chip scheduling losses as high as 30%, and nearly 99% of open-source models on Hugging Face cannot run directly on domestic cards. Here, PHANCY's HAMi vGPU and ModelHub XC, built over eight years, precisely fit into this essential pathway, establishing PHANCY as the "standard interface" that all model vendors and computing power vendors cannot bypass in the industrial leap from "usable" to "user-friendly" domestic computing power. Each new model release and every new chip mass production continuously add new combinatorial assets to the adaptation library of ModelHub XC; each completion of a domestic intelligent computing center requires HAMi to ensure the computing power is "fully utilized, managed well, and running steadily." As the domestic computing ecosystem becomes more prosperous, its intermediate position becomes irreplaceable, further deepening the moat. Moreover, the structure of computing power demand is irrevocably switching from "training-led" to "inference-led." Deloitte estimates that the global share of inference loads will rise from about one-third of AI computing power in 2023 to about two-thirds by 2026, and Nvidia predicts that the potential market size of AI inference chips will reach $1 trillion by 2027, making inference-driven computing power demand an inevitable trend. For PHANCY, this transition precisely positions it in the most comfortable value range. Inference is about "who can optimize heterogeneous computing power scheduling and minimize unit Token costs," which is the main arena for HAMi vGPU, allowing every bit of computing power procurement to be transformed into real Token output. As inference becomes the absolute leader in computing consumption, PHANCYs value as a "shovel seller" owing to "improved heterogeneous scheduling efficiency" will be indispensable, and the substantial backlog of 7 billion yuan in orders serves as the best proof of this logic. In summary: As 2026 begins, the capital markets pricing logic for AI companies is undergoing a profound reconfigurationfrom a complete shift from "narrative faith" to "performance verification." Over the past two years, the market generously covered grand AI visions and substantial capital expenditures; however, investors are now scrutinizing every bit of output quality with a magnifying glass, and commercial monetization capability has become the core metric for differentiating AI company valuations. In this shift of aesthetics, capital no longer lingers for stories, but rather paves the way for deliverables. PHANCY, precisely positioned at the peak of this valuation structure shift, stands to benefit significantly. While pure large model companies are questioned about return cycles due to hefty training costs, and pure cloud vendors face cash flow challenges due to heavy asset investments, PHANCY's fourfold combination of "upgraded revenue structure + verified comprehensive profitability + key strategic positioning + three barriers" demonstrates its real monetization capability in the essential aspects of "connectivity and scheduling." Such intermediary infrastructure companies that "stand up to accounting" are positioned in the most scarce ecological niche within the entire AI industry chain, easiest to obtain valuation premiums. As Tokens become the value anchor of the smart era, PHANCY serves as the "invisible hub" connecting computing power, models, and enterprise applicationsits true value is progressively being reassessed by the market as the industry dividends are unleashed. With strong certainty for sustained high growth, PHANCY is already a core beneficiary in the inevitable trend of AI transitioning from "technological validation" to "scale commercialization."