KNOWLEDGE ATLAS AI has delivered a financial report that has captured Wall Street's attention, with full-year revenue more than doubling year-over-year. However, the data point that truly shocked the market was a real-time figure disclosed post-earnings: as of March 31, 2026, the Annual Recurring Revenue (ARR) for the company's open platform API had surged to approximately 1.7 billion Chinese yuan (about $250 million USD). This represents an increase of over 2.4 times compared to the roughly 500 million yuan at the end of 2025, and a staggering 60-fold growth compared to 12 months prior. Both Morgan Stanley and J.P. Morgan view this as the most significant positive surprise in the earnings report. More convincingly, this growth was not achieved solely by sacrificing price for volume.
According to market analysis, J.P. Morgan specifically highlighted in its research report that the price per token on the KNOWLEDGE ATLAS API platform has risen 83% year-to-date, while demand continues to accelerate. This simultaneous increase in both volume and price is extremely rare amidst the intense price war currently raging in the domestic large model market. It directly confirms that KNOWLEDGE ATLAS has established substantive pricing power in high-value scenarios such as coding and AI agents.
From a profitability perspective, KNOWLEDGE ATLAS's loss structure is undergoing a qualitative change. In 2025, the company's research and development expenditure was approximately 3.2 billion yuan, roughly equivalent to its adjusted net loss. This implies that the gross profit generated by existing models is already sufficient to cover selling and administrative expenses. The core business has essentially reached a break-even point at the contribution level, meaning the entirety of the current losses essentially represents proactive investment into the iteration of the next-generation model.
J.P. Morgan believes that with the continuation of triple-digit revenue growth and the ongoing expansion of API gross margin (rising from just 3% in 2024 to 19% in 2025), the path to profitability is becoming increasingly clear. The firm anticipates the company will achieve profitability by 2029.
**ARR Surge: From 'Year-End Goal' to 'Already in Progress'** The most critical highlight of this earnings report is the ARR data disclosed by KNOWLEDGE ATLAS. As of March 31, 2026, the ARR for the open platform API had reached approximately $250 million USD, a 6.4-fold increase since the start of the year and a roughly 60-fold increase compared to 12 months ago. Management's stated year-end target is $1 billion USD. The current progress indicates this target is not a distant vision but is already on a fast track to being realized. Morgan Stanley categorized the better-than-expected ARR performance in its report as a core event that "strengthens the investment thesis," qualifying it as a "Major Surprise."
From a business structure perspective, revenue from cloud deployments in the second half of 2025 grew 431% year-over-year, far exceeding the 57% growth of private deployments. The proportion of cloud business to total revenue also rapidly increased from a previous single-digit percentage to 26%. This structural shift signifies that KNOWLEDGE ATLAS's business model is evolving from heavy-asset, low-repeat project-based delivery towards a light-asset, high-stickiness subscription-based API economy.
**Token Volume and Price Rise: Pricing Power is the Rarest Signal** In the context of the domestic large model sector generally being mired in a "price war," the fact that KNOWLEDGE ATLAS achieved an 83% year-to-date increase in token price while demand continued to grow is a phenomenon worthy of in-depth analysis. J.P. Morgan analyst Olivia Xu explicitly stated in the research report that the simultaneous upward movement in volume and price is the clearest signal of growth driven by "genuine model competitiveness" and "high-value workloads."
Specifically, for customers in scenarios like coding and AI agents, the payment logic has evolved from "pay-per-use" to "paying for task completion quality, throughput, and stability." This represents a fundamentally higher-level commercial relationship. When customers are willing to pay a premium for superior outcomes rather than merely seeking the lowest unit price, pricing power has quietly been established. From the model iteration path, KNOWLEDGE ATLAS's rapid progression from GLM-4.5/4.6/4.7 to GLM-5, coupled with sustained investment in production-grade coding, long-context reasoning, and multi-step execution stability, forms the technical foundation supporting this pricing power.
**Gross Margin Inflection Point: Cloud Business Crosses from Loss to Profit** In the second half of 2025, the gross margin for KNOWLEDGE ATLAS's cloud deployment business leapt significantly from -0.4% in the first half to 22.4%. This shift marks the formal crossing of the profitability threshold for the cloud business, entering a positive cycle driven by economies of scale.
Looking at the group as a whole, the gross margin for the open platform API has improved from 3% in 2024 to 19% in 2025, an increase of 16 percentage points. J.P. Morgan anticipates there is still significant room for this margin to improve as scale continues to expand and model inference efficiency is optimized. According to its forecast model, the group's overall gross margin will remain around 31% in 2026, further recover to 36% in 2027, and reach 37% in 2028.
Notably, the full-year 2025 gross profit was approximately 297 million yuan, while combined selling and administrative expenses for the same period were about 896 million yuan, and R&D expenditure was approximately 3.2 billion yuan. Excluding R&D investment, the gross profit already essentially covers non-R&D operating costs. This indicates that KNOWLEDGE ATLAS's core business model possesses self-sustaining capability; the current losses entirely reflect strategic R&D investment, not a flaw in the business model itself.
**Private Deployment: Upgrade Potential of Existing Assets** Alongside the explosive growth of the cloud API business, KNOWLEDGE ATLAS's established base of private deployments within China's regulated industries also forms a crucial pillar of J.P. Morgan's investment thesis. In the second half of 2025, private deployment revenue reached 372 million yuan, a 57% year-over-year increase, and its absolute scale remains more than double that of the cloud business.
J.P. Morgan believes this vast existing customer base holds unique strategic value. As the foundational models continue to iterate and upgrade, these deployed customers have a natural propensity for upgrades, which could evolve into a cyclical, predictable source of upgrade-driven revenue. The high sensitivity to data security in regulated industries such as finance, government, and energy also ensures that private deployment will remain an irreplaceable delivery model for the foreseeable future, constituting a structural moat for KNOWLEDGE ATLAS compared to purely cloud-based competitors.
**J.P. Morgan Significantly Raises Price Target** J.P. Morgan maintained its "Overweight" rating and raised its price target from HK$800 to HK$950. This valuation is based on a 30x multiple applied to the estimated 2030 earnings per share, discounted back to the end of 2026 using a 15% weighted average cost of capital (WACC). The 30x valuation premium primarily reflects the anticipated compound annual revenue growth rate exceeding 100% between 2026 and 2030.
According to the latest forecasts, KNOWLEDGE ATLAS's revenue is projected to reach 3.192 billion yuan in 2026 (a 341% increase), further grow to 7.257 billion yuan in 2027, leap to 19.725 billion yuan in 2028, and is expected to surpass 98.8 billion yuan by 2030. Adjusted net profit is forecast to turn positive in 2029, reaching 2.822 billion yuan, and further increase to 20.36 billion yuan in 2030.
Morgan Stanley also maintained its "Overweight" rating, with a price target of HK$560. Its valuation employs a Discounted Cash Flow (DCF) methodology, assuming a 15% WACC and a 3% perpetual growth rate, corresponding to a price-to-sales ratio of approximately 53x for 2027. While the two institutions differ in their methodological approaches, their judgment regarding KNOWLEDGE ATLAS's long-term value is highly aligned.
Comments