MetaX's revenue surges, But still faces losses

On March 5, MetaX released its performance forecast for the first quarter of 2026. The company expects to achieve revenue between 400 million and 600 million yuan, representing a year-on-year increase of 24.84% to 87.26%. However, the net profit attributable to the parent company is projected to remain at a loss, estimated between 90.76 million and 181.51 million yuan. This opening report for the year continues the trend seen in 2025, where annual revenue reached 1.644 billion yuan (up 121.26% YoY), while net loss stood at 781 million yuan.

The coexistence of high revenue growth and persistent losses has become the most defining financial characteristic of leading domestic GPU manufacturers like MetaX. AI chip R&D is notoriously a high-investment sector; even industry leaders must endure long-term capital outlays before reaching a profitability breakthrough. For instance, Cambricon did not see its first quarterly profit until Q4 2024, when its quarterly revenue reached 989 million yuan.

"MetaX S" Launch Completes Four-Tier Product Matrix

The core driver of MetaX’s steady performance growth lies in continuous breakthroughs in product strength and the refinement of its full-matrix layout. In January 2026, the company released the MetaX S-series (Xi Suo), a GPU specifically optimized for AI-for-Science (AI4S). This product precisely fills a gap in the domestic GPU market for scientific computing and serves as a strategic move to capture the high-end computing market. According to company disclosures, the S206 entered mass production in January, with subsequent models S301 and S302 slated for release in Q2 2026.

With this, MetaX has established four major product lines: the C-series (Xi Yun) for integrated training and inference, the N-series (Xi Si) for AI inference, the G-series (Xi Cai) for graphics rendering, and the S-series (Xi Suo) for scientific intelligence. This makes MetaX one of the few domestic GPU firms covering all computing scenarios. Financial trends suggest the company has entered a positive trajectory of "high revenue growth + narrowing losses," as economies of scale begin to manifest and R&D/marketing costs are progressively amortized. As multiple new products hit mass production in 2026, shipment volumes are expected to rise further, strengthening revenue momentum and accelerating the pace of loss reduction.

The "Hard Bones" Behind the Prosperity

While the domestic GPU industry is enjoying a period of growth, it must also confront three core challenges: ecosystem, capacity, and competition. The journey of MetaX and its peers serves as a microcosm of the industry's struggle to break through these barriers.

Regarding ecosystem barriers, MetaX’s self-developed MXMACA software platform has achieved deep compatibility with the CUDA ecosystem, reaching a direct adaptation rate of 92.94% for CUDA projects. It has also completed full-link adaptation for mainstream AI frameworks and operating systems, significantly lowering the migration threshold for developers. However, objectively speaking, complex projects still require minor compilation adjustments. Real-world costs remain for downstream enterprises moving from adaptation to large-scale deployment, and building true "ecosystem stickiness" requires long-term cultivation.

Capacity constraints have become a collective headache. With global advanced packaging (such as CoWoS) capacity remaining tight, domestic GPU firms face immense pressure on mass-market delivery. Leading companies are attempting to secure capacity through advance bookings with foundries and OSAT (Outsourced Semiconductor Assembly and Test) providers, while optimizing packaging schemes via proprietary interconnect technologies. Nevertheless, total supply remains constrained by the broader industry environment, and delivery stability has yet to be fully validated.

Finally, the competitive landscape is intensifying. In the high-end computing sector, a significant gap remains between domestic GPUs and international giants in terms of core architecture, raw performance, and ecosystem maturity. Commercializing and refining high-end products will take time. Conversely, the mid-to-low-end market is becoming crowded with homogenous products, where core technical moats have yet to be established. Perhaps most challenging is that major downstream clients—such as ByteDance and Alibaba—are increasingly developing their own in-house AI chips, further squeezing the market space for independent domestic GPU vendors.