XUNCE Technology (03317), which was previously "misjudged," presents a long-term investment opportunity during its current price correction. The announcement of DeepSeek-V4's official preview launch and simultaneous open-sourcing on April 24th caused significant volatility in the share prices of the so-called "Lobster Trio" - MiniMax, Zhipu, and XUNCE Technology - in the Hong Kong market. However, a deeper analysis reveals that the core logic behind XUNCE's stock fluctuations differs fundamentally from that of large model enterprises like MiniMax and Zhipu.
For large model companies, the primary anchor for valuation is their "model capability," a competitive asset. DeepSeek-V4 established a new industry benchmark with its significant performance advantages and disruptive cost structure, prompting a market-wide reassessment of the technological moats and future growth potential of existing players. This directly caused the volatility in large model company stocks. In contrast, XUNCE's stock movement resulted from both market disturbance and information misinterpretation.
Regarding market disturbance, the earlier narrative grouping the "Lobster Trio" under a single theme meant XUNCE was lumped together with large model companies during the peak of OpenClaw concept popularity, sharing a similar valuation logic. This led to short-term high correlation between its stock price and the large model sector. Consequently, when DeepSeek-V4's release triggered a re-rating of large model companies, market sentiment and capital flows naturally affected XUNCE as well, though this is merely a short-term trading impact.
Concerning information misinterpretation, some market participants incorrectly equated "significantly reduced computational load per token inference" with "future total token demand will decrease," leading to a simplistic but flawed linear extrapolation. DeepSeek-V4's new DSA sparse attention mechanism achieved a breakthrough compression at the token level, reducing the computational load for single-token inference in million-context scenarios to just 27% of the previous V3.2 model. Some investors misinterpreted this as a negative development, impacting the stock price of XUNCE, often called the "Number One Token Stock." This represents a fundamental logical error.
This view confuses the relationship between "unit efficiency" and "total demand," ignoring the historical pattern where technological revolutions often stimulate and unleash demand by lowering barriers to usage. For instance, improved internal combustion engine efficiency reduced fuel consumption per 100 kilometers, but it did not decrease total gasoline demand. Instead, due to increased car ownership and travel frequency, total gasoline consumption rose substantially, driving continued prosperity in the oil industry. The release of DeepSeek-V4 represents precisely such a paradigm shift where efficiency drives demand.
It is not merely a leader in technical specifications; it systematically breaks through core bottlenecks in current AI industry development across four dimensions: cost, autonomy/control, application, and ecosystem, and will significantly accelerate the explosion in token demand.
Cost breakthrough is the primary driver for demand release. DeepSeek-V4 pushes token processing costs to an extreme. Its Pro version is currently the lowest-cost option among large models, while its Flash version costs only 2 yuan per million output tokens, lower than OpenAI's GPT-5.4 Nano, and is priced at just 1% of some competitors. This transforms AI from an expensive "trial product" into an accessible "daily commodity," unleashing massive pent-up demand from enterprises and individuals previously inhibited by high costs.
Autonomy and control open the path for sustained cost reduction. DeepSeek-V4's deep compatibility with domestic chips like Huawei's Ascend and Cambricon lays a technical foundation for future cost decreases. DeepSeek has stated explicitly that once the Ascend 950 super-nodes are mass-marketed, the Pro version's price will be significantly reduced further, providing continuous momentum for demand growth.
Agent capabilities and multimodal integration are unlocking entirely new demand scenarios. The combination of DeepSeek-V4's powerful Agent capabilities and extremely low costs enables the large-scale deployment of persistent AI agents capable of long-term memory and executing complex tasks. IDC forecasts that the number of active AI Agents globally will exceed 2 billion by 2030. These agents, operating 24/7, will consume astronomical amounts of tokens. Simultaneously, processing multimodal data like images and audio will create new entry points for token demand.
The open-source ecosystem will spawn a massive number of applications, each a new consumption channel. By open-sourcing its flagship V4 model, DeepSeek broke the industry convention of "open-sourcing basic models while keeping flagship models proprietary." This strategy will significantly lower the global barrier to development, attracting developers to build a rich ecosystem. The ensuing flood of innovative applications will each become a new, persistent channel for token consumption.
It is foreseeable that DeepSeek-V4's "low-cost + open-source" combination will propel the already rapidly growing token market into a new phase of exponential expansion. According to the National Data Bureau, as of March 2026, China's daily token call volume exceeded 140 trillion, growing over a thousandfold in two years, with a recent three-month increase exceeding 40%. J.P. Morgan predicts that by 2030, domestic AI inference token consumption will grow to 39 quintillion, a 370-fold leap from 2025. Considering the massive global demand increment from token出海, the actual growth in domestic token consumption will be even more explosive, especially since Chinese models already account for 61% of the total token consumption among the top ten models on the OpenRouter platform.
This signifies that the efficiency revolution driven by technology will not reduce demand but will instead inaugurate a new era of vigorous token demand explosion. The value distribution landscape of the token economy will consequently be reshaped. As underlying models become efficient and cheap, the industry's competitive focus will shift from "competing on model parameters" to building a prosperous application ecosystem based on these models. Within this ecosystem, unique, real-time, high-quality data becomes the core of differentiated competition. Whoever possesses the vertical-specific token pipelines that refine raw data into tokens that large models can directly and efficiently utilize will control the ecosystem's lifeline.
XUNCE Technology is the essential "water seller" in this critical link. After a decade of deep cultivation, the company has refined raw data from high-barrier industries like finance and government affairs into vertical-specific tokens ready for direct use. Reports indicate its vertical token call prices reach $10-100 per million tokens, more than ten times higher than similar products from Anthropic, with prices continuing to rise as professional scenarios deepen. As an infrastructure provider of "data fuel," XUNCE's strategic positioning value will continue to compound alongside the expansion of the entire AI application ecosystem.
Explosive performance growth serves as the most direct validation. In April 2026, its Annual Recurring Revenue related to token calls surged by 300% quarter-over-quarter. This marks a transition where token call revenue has upgraded from a supplementary item to the core engine driving the company's revenue structure towards an exponential growth model "deeply tied to token call volume." The company aims to increase the proportion of token-related revenue to 20%-30% by 2026.
With the accelerated adoption of the tokenized payment model, XUNCE's growth logic is undergoing a fundamental reconstruction. Its revenue model evolves from the additive logic of "new customers × new modules" under traditional subscription models to a multiplicative model of "price per call × number of token calls × number of module applications." This implies that the value per customer will experience exponential amplification as enterprise AI applications deepen.
Therefore, XUNCE Technology not only benefits from the overall explosion in token demand but also, through its tokenized business model, embeds itself into the most certain and highest-growth potential key segment of future AI value distribution. This is the underlying logic behind the strong long-term optimism from numerous institutions. It directly supports the company's staggering 658% stock price increase within the year, propelling its market capitalization past the hundred-billion Hong Kong dollar milestone.
However, based on the company's strategic positioning as key AI industry infrastructure, as the entire AI sector accelerates its expansion, the current hundred-billion HKD market cap might merely represent a new starting point. Its true journey may be the long-term voyage towards the vast potential of a trillion-dollar US market capitalization.
Comments