Deepexi delivered impressive results in the second quarter. According to its 2026H1 interim report, Q2 revenue reached approximately CNY 230 million, with a net profit of about CNY 30.1 million, achieving a net profit margin of roughly 13.1%. For a vendor still investing heavily in enterprise large models, ontology, and agent platforms, this marks a critical profitability threshold: once revenue reaches a certain scale, Deepexi's existing business can cover R&D, sales, and administrative expenses while generating substantial profits. However, making money in a single quarter does not equate to establishing sustainable profitability. Deepexi's Q2 revenue accounted for approximately 81% of the first half, which is 4.27 times that of Q1. During the same period, the gross profit margin only increased by 1.5 percentage points year-on-year, and both adjusted profit and operating cash flow for the first half remained negative.
Putting these three sets of data together points to a more direct question: Does the high net profit margin in Q2 stem from the sustainable operational capabilities formed by the AI platform, or does it primarily result from the dilution of period expenses by revenue scale following the centralized acceptance of projects?
To determine whether the profits of an enterprise AI vendor can be sustained, three main factors must be evaluated: whether revenue growth can dilute expenses, whether individual customers continue to make additional purchases, and whether the platform can reduce project delivery costs. In summary, these are three operational leverage mechanisms: scale dilution, customer value, and platform reuse. Deepexi's Q2 single quarter demonstrated the scale dilution effect, but the other two leverage effects have not yet materialized.
01、The Profit Generated by Deepexi in Q2 Is Merely the Scale Dilution Leverage Effect
The most significant variable in Deepexi's Q2 profit turnaround was revenue, rather than gross profit margin or period expenses. Calculated based on the interim report data, Deepexi's Q1 revenue was approximately CNY 53.93 million, rising to about CNY 230 million in Q2; Q1 net loss was about CNY 62 million, while Q2 net profit was approximately CNY 30.1 million. Between the two quarters, revenue increased by about CNY 176 million, and profit improved by approximately CNY 92.1 million (as shown in Figure 1).
Figure 1: Deepexi's Q2 2026 Revenue Surges, Net Profit Turns from Negative to Positive
The improvement in gross profit margin provided only a small incremental contribution, which is insufficient to explain such a massive change in profit. In 2026H1, Deepexi's overall gross profit margin was 56.5%, a year-on-year increase of 1.5 percentage points; the gross profit margin of its core AI business, DeepexiOS, was 57.2%, up 2 percentage points year-on-year. This indicates that platform capabilities may have begun to improve direct costs, but the magnitude remains very limited. Nor did Deepexi trade profits for massive layoffs and reduced investments.
In the first half of the year, sales and marketing expenses increased by 33.2% year-on-year to CNY 65.68 million, and R&D expenses grew by 23.8% to CNY 72.13 million; the number of employees increased from 464 at the end of 2025 to 548. The sales expense ratio and R&D expense ratio decreased significantly, primarily because revenue growth far outpaced expense growth.
Therefore, the main mechanism behind Deepexi's Q2 profit is not complex: centralized project delivery and acceptance led to a surge in revenue. With the gross profit margin remaining basically stable and period expenses not increasing proportionally in the short term, the newly added gross profit crossed the break-even line. Ai Analysis believes that Deepexi achieved profitability in Q2 mainly due to concentrated revenue growth without a synchronous increase in expenses. This shows that Deepexi has crossed the single-quarter break-even line, but it does not yet prove that the AI platform has reduced the delivery cost of each project to form a stable and sustainable profitability model.
02、Whether the High Net Profit Margin Can Be Sustained Depends on the Scale of the Other Two Leverages
Dilution can bring profits on the premise of sustained and rapid revenue growth. However, 99.2% of Deepexi's first-half revenue was recognized at the point of project delivery or acceptance. The acceptance rhythm of project-based business fluctuates significantly, making it difficult to release revenue evenly across quarters. Therefore, relying solely on scale dilution makes it hard to sustain single-quarter profitability stably. For enterprise AI vendors to generate sustainable, scaled, and high-quality profits, they must ultimately establish the two levers of customer value and platform reuse.
Next, we will analyze and verify whether Deepexi possesses or demonstrates these two leverage effects.
2.1 Customer Value Leverage
The customer value leverage is not simply about serving more customers, but rather integrating into the customers' more core production, operation, and decision-making processes, enabling individual customers to purchase more scenarios, sign larger orders, and continue to make additional purchases. Ultimately, this should be reflected in an increased average customer spend, additional purchases from existing customers, and extended contract durations. Deepexi has proven its ability to enter complex scenarios for large customers in manufacturing, energy, and public utilities, and there have been projects worth tens of millions of yuan and expansions within existing customers. However, public data has not yet shown this capability translating into a stable increase in average customer spend; instead, the average customer spend has declined as the customer base grows.
From 2024 to 2025, the number of FastAGI customers increased from 20 to 70, yet the average customer spend dropped from approximately CNY 4.52 million to CNY 3.64 million. The rapid growth in the number of customers indicates that market demand is taking off; the decline in average customer spend, however, suggests that revenue growth at the time came more from customer expansion rather than the continuous amplification of individual customer value. In 2026H1, Deepexi did not continue to disclose the number of customers, average customer spend, net revenue retention rate, and upsell revenue, making it impossible for external observers to judge whether there have been significant changes in the operation of large customers (as shown in Figure 2).
Figure 2: Verification Progress of Deepexi's Three Profit Formation Mechanisms
Historically, Deepexi has exhibited a combination of low customer retention rate and high net revenue retention rate. This indicates that enterprise AI is not necessarily a standard SaaS-style renewal business, but is more likely to rely on large customers, large orders, and long-term upsells: first building the data and AI foundation, and then expanding to more departments, business lines, and scenarios. However, a few large contracts can only prove the existence of capability; continuous increases in average customer spend and upsell data are required to prove that this is a stable mechanism.
2.2 Platform Reuse Leverage
The platform reuse leverage answers another question: when revenue increases, does the vendor still need to add a proportional amount of personnel, outsourcing, and third-party procurement resources? Deepexi has already built a relatively complete product system. FastData handles data governance and data engineering; Deepology accumulates enterprise ontology, industry knowledge, and skills; the Deepexi enterprise large model understands enterprise semantics and generates task plans; and DeepWorks organizes models, AI employees, and tools to execute tasks. Theoretically, these capabilities can accumulate project experience into reusable assets, reducing repetitive development and delivery work for subsequent projects.
In practice, Deepexi's per capita efficiency has indeed improved, and the revenue growth rate is also higher than the growth rate of related cost of sales. However, the number of employees still increased from 323 at the end of 2024 to 464 at the end of 2025, and further to 548 in the first half of 2026; during the same period, sales and R&D investments continued to increase, and the overall gross profit margin only improved by 1.5 percentage points (as shown in Figure 3).
Figure 3: Deepexi's Per Capita Revenue Increases Significantly, While Gross Profit Margin Improvement Is Limited
Based on existing data, it can only be concluded that revenue growth is faster than personnel and expense growth; it cannot yet indicate that the delivery resources required for new revenue are continuously declining. To judge whether a platform has truly formed reuse value, one cannot only look at internal labor efficiency. On-site deployment, outsourced implementation, third-party hardware and software, cloud resources, and computing power all belong to the complete delivery cost. If the investment in in-house personnel decreases but is made up for by external procurement, the cost has merely shifted its location.
Therefore, only when the unit delivery cost continues to decline and ultimately translates into an increased gross profit margin can the platform reuse leverage be considered truly established. In addition, book profits must be able to turn into collected cash. In 2026H1, Deepexi's net operating cash outflow was approximately CNY 91.03 million, and accounts receivable increased by 44.6% compared to the end of 2025 to approximately CNY 445 million.
Although profitability was achieved in Q2, it is not yet sufficient to prove that profits can be realized stably. From this perspective, for Deepexi's high net profit margin in Q2 to become a sustainable capability, at least three consecutive changes need to be observed: an increase in average customer spend and upsells from large customers, continuous improvement in gross profit margin alongside revenue scale, and a turnaround in operating cash flow to positive.
03、None of the Three Listed Samples Have Fully Completed the Loop
Deepexi is not an isolated case in pursuing a sustainable profitability model. The performances of Wenge Group and Hai Zhi Technology illustrate the common challenges faced by enterprise AI platform vendors from different perspectives. Public data from Wenge Group shows that improved delivery efficiency does not necessarily lead to a significant improvement in gross profit margin. Its average delivery cycle was shortened from 185 days in 2023 to 72 days in the first half of 2026, and labor costs in the cost of sales decreased by 28.5% year-on-year; however, in 2026H1, procurement of third-party hardware, software, and services increased by 260.7% year-on-year, and the proportion of procurement in the cost of sales rose from 25.1% to 62.5%.
Ultimately, the overall gross profit margin only increased from 54.6% to 55.0%. Therefore, the data disclosed by Wenge Group indicates that while platformization shortened the cycle and reduced in-house labor investment, some costs were shifted to external procurement, and unit economics did not improve synchronously. Hai Zhi Technology, on the other hand, presents a different kind of progress: platform reuse and upsells from existing customers can improve unit economics, but it is still a distance away from a complete profitability closed loop.
In 2026H1, Hai Zhi's revenue grew by 70.4%, and the overall gross profit margin increased from 38.5% to 44.7%; Atlas agent revenue grew by 135.6%, of which 69% came from existing graph customers, indicating that the platform foundation is extending to higher-value agent businesses. At the same time, Hai Zhi still reported an adjusted loss of approximately CNY 19.6 million, with a net operating cash outflow of about CNY 108 million, and accounts receivable continued to grow.
The three samples present three different types of progress: Deepexi achieved a high single-quarter net profit margin after revenue scale dilution; Wenge Group significantly shortened its delivery cycle; and Hai Zhi improved its gross profit margin and formed a certain degree of cross-product upselling. However, none of them has simultaneously proven that scaled revenue, customer upsells, reduced unit delivery costs, and cash returns have formed a stable closed loop. This does not lead to the conclusion that enterprise-level AI platform vendors cannot be profitable; a more accurate conclusion is: market demand has taken off, but the sustainable profitability of enterprise AI remains to be verified.
Palantir's latest financial performance illustrates the potential business outcomes after the customer value leverage matures: AI is deeply integrated into core businesses, large customers continue to make additional purchases, and revenue growth is accompanied by high gross profit margins, high net profit margins, and strong cash flow. However, it is difficult for Chinese vendors to simply replicate this result. Domestic enterprise AI budgets are usually scattered across different departments and annual projects, with private deployment, fixed total price, and payment upon acceptance still being quite common; customers demand open interfaces and multi-vendor compatibility, while cloud vendors, large model companies, software vendors, and system integrators are competing for the same pool of budgets.
Even if enterprise AI vendors enter the top-tier customers, they may not necessarily secure long-term contracts, continuous upsells, and sufficiently strong pricing power. Another path is to maintain a mid-to-large customer base, expand revenue with more customers and projects, and then improve the gross profit margin through platform reuse. However, Wenge Group and Hai Zhi Technology show that this path is equally not easy. An increase in customers will bring sales, solution design, deployment, acceptance, and service activities, and the improvement in internal per capita efficiency may be offset by outsourcing and procurement costs. Revenue can be scaled, but profits may remain thin for a long time.
04、The Next Round of Elimination in Enterprise AI Depends on Business Outcomes
The business revenues of Deepexi's DeepexiOS, Wenge Group's AI decision-making platform, and Hai Zhi's Atlas agents are all growing rapidly, indicating that demand has emerged, but competition will also intensify accordingly. Cloud vendors, large model companies, enterprise software vendors, data intelligence vendors, and system integrators are competing for the same pool of enterprise budgets, business scenarios, and top-tier customers.
In the next stage, enterprise AI vendors will continue to adopt different technical routes, and ontology, knowledge graphs, enterprise large models, and agent platforms may coexist for a long time. The choice of technical route will not directly determine the competitive outcome; what truly matters is whether these technologies can drive customer upsells, reduce project delivery costs, and translate into sustainable profits.
Combining the three listed vendor samples, Ai Analysis believes that the next round of elimination for enterprise AI vendors is not about one technical route eliminating another, but rather the elimination based on business outcomes. Vendors that can transform technology into customer value leverage or platform reuse leverage will have the opportunity to stay at the table; vendors that can only rely on revenue scale to dilute expenses in the long run will be the first to face pressure.