Chinese GPU Vendors: Revenue Surge, Quarterly Profit, and the Industry Logic Behind It

    A leading Chinese GPU maker posted 2025 revenue of CNY 1.506 billion (up 243%) and a first quarterly profit in Q1 2026, while full-year 2025 remained loss-making. This article unpacks the structural drivers, R&D intensity, competing go-to-market paths, and remaining gaps from quarterly to sustained profitability.

    Chinese GPU Vendors: Revenue Surge, Quarterly Profit, and the Industry Logic Behind It

    1. Dual signals: explosive revenue and a profitability inflection

    In 2025, a leading Chinese GPU vendor reported full-year revenue of CNY 1.506 billion, up 243.37% year on year. Growth continued in Q1 2026, with quarterly revenue of CNY 738 million (up 155.35%). Net profit attributable to the parent reached CNY 29.36 million that quarter—the company’s first quarterly profit since founding. Gross margin stayed in a tight band of 65.57%–67.35%, indicating a balance between pricing power and cost control.

    Full-year 2025 net profit attributable to the parent was still a loss of CNY 1.001 billion, though that was 38.16% narrower than the CNY 1.618 billion loss in 2024. The combination of quarterly profit and annual loss is typical of high-end compute chip companies in a strategic investment phase: heavy R&D to win future share, with staged losses used to build long-term barriers.

    2. Structural drivers of growth

    2.1 Compute localization accelerating

    Revenue jumped from CNY 438 million to CNY 1.506 billion as Chinese AI compute chips gained share. Public industry figures put Chinese vendors at about 41% of the overall market in 2025; one leading firm accounted for nearly half of Chinese chip shipments and roughly 20% of the total market. Large buyers such as major internet and cloud operators have added Chinese GPUs to core infrastructure shortlists, opening a commercial window that did not exist at this scale before.

    2.2 Product mix and customer validation

    Growth was not purely volume at the expense of price. Gross margin above 65% suggests products have reached a stage where performance, software compatibility, and support can command a premium. The lineup expanded from consumer graphics GPUs to general-purpose GPUs for AI training and inference, enabling entry into internet, finance, energy, and other high-value workloads. A 2025 A-share listing provided capital for capacity and customer expansion and narrowed the gap between brand trust and commercial qualification.

    2.3 Seasonality and delivery timing

    Part of the quarterly turnaround reflects industry delivery patterns. Large compute purchases often concentrate in the second half of the year; Q1 is frequently when existing contracts are recognized. CNY 738 million in a single quarter is a large share of the annual run-rate. Whether that profitability is durable depends on more even performance across four quarters.

    3. Narrowing losses versus R&D intensity

    3.1 What an 86.68% R&D-to-revenue ratio means

    In 2025 the vendor spent CNY 1.305 billion on R&D, or 86.68% of revenue—well above a typical industry range of 20%–30%. That reflects catch-up work on architecture, driver compatibility, and compiler toolchains. GPU value depends as much on a full software stack (ISA through application frameworks) as on peak FLOPS; later entrants cannot buy that stack off the shelf.

    As the revenue base grew, R&D intensity fell from earlier levels often above 200% toward ~86%. The company is moving from a pure-investment phase toward a period where spend and returns can move together. Annual profitability becomes more likely only if revenue growth stays ahead of R&D growth.

    3.2 Three paths that narrowed losses

    Losses evolved as: CNY 1.894 billion (2022) → 1.703 billion (2023) → 1.618 billion (2024) → 1.001 billion (2025). Narrowing did not come from simple cost cutting:

    • Scale diluted fixed costs (engineering headcount, EDA licenses, tape-outs).
    • Product maturity improved yields and reduced one-off customization, holding gross margin above 65%.
    • Relatively controlled sales and admin spend let operating leverage start to work in the right direction.

    4. Two different paths among Chinese compute-chip companies

    One path is proprietary architecture and a relatively closed stack, sold as system-level solutions with tight integration and deep ties to large government and enterprise accounts—strong on integration, weaker on openness and third-party developers. The other, represented here, is compatibility with mainstream compute ecosystems and cost-effective general-purpose GPUs: an “ecosystem adapter” that lowers migration cost and uses flexible commercial models to enter incremental demand.

    Neither path is universally superior; fit with resources matters. Unlisted firms need longer strategic patience and diversified funding to sustain high R&D. Listed firms must balance public-market pressure for near-term profit with long-cycle technical investment.

    5. From quarterly profit to sustained profitability

    5.1 Supply- and demand-side tailwinds

    Global AI compute demand remains in a high-growth phase. Models expanding from text into multimodal, inference, and edge use cases multiply demand beyond training. In China, intelligent computing centers and private industry-model deployments continue to expand the addressable market. Some forecasts put the domestic AI chip market at a compound annual growth rate above 40% over the next three years.

    5.2 Persistent challenges

    A profitable quarter does not guarantee a profitable year. Pressures include: high switching costs and inertia around incumbent GPU software ecosystems, requiring ongoing compatibility layers; uncertain access to advanced process capacity and HBM; and a growing number of domestic competitors, with risks of price competition and product similarity.

    5.3 Technical variables

    Next-generation GPU ceilings will depend on: interconnect and Chiplet packaging that scale compute without blowing up cost; a software stack that balances CUDA-class compatibility with independent innovation; and energy-efficiency gains via more specialized designs for inference and other targeted workloads.

    6. Conclusion: treating high growth plus losses as a transitional normal

    For Chinese GPU vendors, 2025–2026 revenue growth and a quarterly profit look like a preliminary result: products can be sold and the commercial loop can close at small scale. Moving from quarterly to full-year profit, and from edge use cases to core substitution, still requires crossing technical and ecosystem gaps.

    For technical decision-makers, these financials illustrate a broader question: how to judge the growth pace of a strategic compute industry. Capital patience, policy continuity, and an open approach to ecosystem building all matter. When the pattern of high growth plus losses gives way to scale that funds further R&D, Chinese GPUs will be closer to a more durable phase of development.