<p>Against the backdrop of accelerated construction of domestic intelligent computing centers and rising demand for domestically produced computing power, MetaX achieved an annual operating revenue of 1.644 billion yuan, a year-on-year increase of 121.26%, continuing its rapid revenue growth.</p>
<p>Losses also improved simultaneously. According to the financial report released on Thursday, the company's net loss attributable to the parent in 2025 was 789 million yuan, a reduction of 43.97% compared to the previous year. The non-GAAP net loss was 830 million yuan, a reduction of 20.52% compared to the previous year.</p>
<p>In terms of business progress, the company emphasized that it has become one of the few GPU suppliers in China to achieve &#34;large-scale commercial application in thousand-card clusters.&#34; Cumulative GPU sales have exceeded 55,000 units. It is advancing the implementation of more industry scenarios and larger clusters through its MXMACA software stack, which is compatible with the CUDA ecosystem, and its self-developed MetaXLink high-speed interconnect capabilities.</p>
<h2>Revenue: 1.644 Billion Yuan, Doubling Growth, Q2 Contributed the Most</h2>
<p>In 2025, MetaX's revenue reached 1.644 billion yuan, a year-on-year increase of 121.26%. From a quarterly perspective, the company's revenue was concentrated in the second quarter:</p>
<ul>
<li>Q1: 320 million yuan</li>
<li>Q2: 595 million yuan</li>
<li>Q3: 321 million yuan</li>
<li>Q4: 408 million yuan</li>
</ul>
<p>The company's main business revolves around full-stack GPU chips and computing platforms. Revenue primarily comes from GPU boards, servers integrating its own boards, all-in-one machines/workstations, and intelligent computing cluster deliveries. It also provides a software stack (MXMACA) compatible with the mainstream CUDA ecosystem, meeting customer computing power needs in the form of &#34;hardware + software + system.&#34;</p>
<h2>Income Statement: Significant Loss Reduction, but Quarterly Fluctuations are Large and Profitability Quality Still Needs Observation</h2>
<p>The full-year net loss attributable to the parent was 789 million yuan, and the non-GAAP net loss was 830 million yuan, both narrowing compared to the previous year. Roughly calculating the net profit margin as net loss/revenue, it is still around -48%, indicating that the increase in scale is not yet sufficient to cover R&amp;D and industrialization costs.</p>
<p>It is noteworthy that profits fluctuated significantly between quarters:</p>
<ul>
<li>Q2 net profit attributable to the parent was +46.6189 million yuan, and even after excluding non-recurring items, it was still +16.5984 million yuan;</li>
<li>Q4 saw a net loss attributable to the parent of -444 million yuan and a non-GAAP net loss of -437 million yuan.</li>
</ul>
<p>In the phase where AI chip companies' delivery rhythms, R&amp;D investment recognition, expense accrual, and industrialization ramp-up are intertwined, quarterly fluctuations are not uncommon. However, this also suggests to investors that the company's profit recovery is not a linear process. Subsequent tracking of gross profit margin, expense ratios, customer acceptance, and payment collection rhythms, among other detailed indicators (to be combined with the full annual report), will be necessary.</p>
<h2>Products and Commercialization: Three Product Lines Cover Training/Inference and Graphics Rendering, Adopting a &#34;Solutions&#34; Approach</h2>
<p>MetaX is still in the typical AI chip R&amp;D phase of &#34;high investment, long cycle.&#34; The annual report shows that R&amp;D investment accounted for 62.49% of revenue. The cumulative unrecognized losses of the parent company are 1.549 billion yuan, and it plans no cash dividends for 2025. Net cash outflow from operating activities was 1.26 billion yuan, which improved year-on-year but still indicates pressure from cash consumption.</p>
<p>The company's product lines are more clearly divided by scenario:</p>
<ul>
<li>XiYUN C Series (C500/C600): Training/inference integrated, general-purpose computing, AI for Science;</li>
<li>XiSI N Series (N100/N260/N300): From traditional AI inference to generative AI inference;</li>
<li>XiCAI G Series (G100): Under research/planning for graphics rendering.</li>
</ul>
<p>In terms of business model, the company adopts a Fabless model (outsourcing wafer manufacturing, packaging, and board processing). The sales side operates through both direct sales and distribution. Compared to &#34;selling single cards,&#34; the annual report emphasizes delivering servers, all-in-one machines/workstations, and even intelligent computing clusters composed of multiple servers and network storage to customers, reflecting its choice of an &#34;engineering delivery + ecosystem adaptation&#34; strategy in the competitive domestic GPU industry.</p>
<p>The company states that its products have been deployed in over 10 intelligent computing clusters, covering national AI public computing platforms, operator intelligent computing platforms, and commercial intelligent computing centers. Regions include Beijing, Shanghai, Hangzhou, Changsha, Hong Kong, China, etc. It is also advancing scenario implementation in education and research, finance, transportation, energy, healthcare, major cultural and entertainment sectors, as well as in &#34;embodied AI&#34; and &#34;low-altitude economy&#34; directions.</p>
<p>MetaX's judgment on industry trends in the annual report focuses on three main lines:</p>
<ol>
<li>
<p>From &#34;Computing Power Sovereignty&#34; to &#34;Computing Power Equality&#34;: Represented by algorithm optimization and open-source diffusion brought by DeepSeek and others, short-term improvements in training efficiency may affect demand for single training instances. However, in the long run, under the &#34;Jevons Paradox,&#34; the explosion of AI applications may still lead to overall growth in computing power demand. Furthermore, the structure of computing power demand is shifting from &#34;heavy pre-training&#34; to a more balanced &#34;pre-training + post-training + inference.&#34;</p>
</li>
<li>
<p>Interconnection Determines Effective Computing Power in Distributed Parallelism: ScaleUp/ScaleOut parallelism is becoming mainstream. The bandwidth of multi-card interconnection directly affects cluster efficiency, and high-speed interconnection technology upgrades will continue.</p>
</li>
<li>
<p>Software Ecosystem Determines Usability and Landing Speed: In the era of large models, operators are frequently updated. The ability of chips to quickly adapt to frameworks, models, and toolchains is becoming a key foundation that affects the richness of scenarios.</p>
</li>
</ol>
<p>Regarding market size data, the company cites third-party research stating that China's AI accelerator chip market size was approximately 142.537 billion yuan in 2024, a year-on-year increase of 98.49%. It is projected to grow to 239.8 billion yuan in 2025. IDC data shows that the Chinese accelerated computing server market size was 22 billion US dollars in 2024, a year-on-year increase of 134%, and is expected to exceed 100 billion US dollars by 2029.</p>
<h2>Does Not Meet Cash Dividend Conditions</h2>
<p>The company notes that it &#34;was not profitable at the time of listing and has not yet achieved profitability.&#34; For 2025, it plans no cash dividends, no bonus shares, and no capitalization of reserves. This is because the parent company has accumulated unrecognized losses, and the distributable profit available in the parent company's statements is negative.</p>
<p>Regarding shareholder structure, as of the reporting period, there were 31,302 ordinary shareholders. Among the top ten shareholders, founder Chen Weiliang and his concerted action platform (Shanghai Jiaomai, Shanghai Xiji) hold significant stakes. Ge Weidong and his concerted action platform (Shanghai Chaos Investment) are also among the top ten, along with several equity investment funds and guidance funds. The audit opinion issued by Lixin is a standard unqualified opinion.</p>

688802

Wallstreetcn

MetaX

MetaX's 2025 Revenue Surges 1.2x Year-on-Year to 1.644 Billion Yuan, Net Loss Narrows Significantly to 790 Million Yuan, Cumulative GPU Sales Exceed 55,000 Units | CICC