1. [para. 1][para. 2] China’s push to build a self-reliant AI hardware ecosystem is now yielding concrete gains for corporate bottom lines. In first-half 2026 earnings, the country’s three leading GPU developers — Cambricon, Hygon, and Moore Threads — demonstrated a transition from a state-subsidized promise to a real business success. Rapid enterprise AI deployment and U.S. export controls on advanced chips drove their growth, and each posted results that put them on a path toward sustained profitability.
2. [para. 3][para. 4][para. 5] Cambricon was the standout. It reported profit growth outpacing revenue growth for the first time — a key milestone showing it can offset heavy R&D overhead with high sales. Revenue and net profit attributable to shareholders both more than doubled year-on-year, reaching 6 billion yuan ($892 million) and 2.3 billion yuan, respectively. Cambricon also notched back-to-back quarterly net profits of over 1 billion yuan. Nearly all first-half revenue came from supplying cloud-based AI acceleration chips to major large-model developers like DeepSeek and Moonshot AI for training and inference tasks.
3. [para. 6][para. 7] Hygon booked the largest revenue among the trio, leveraging a hybrid business model of x86 architecture-based general-purpose CPUs and deep computing units (DCUs). It posted 1.8 billion yuan in net profit attributable to shareholders on revenue of 9.1 billion yuan, up 66.5% year-on-year. Its products’ high compatibility with legacy architecture makes it a preferred supplier for risk-averse enterprise clients in telecommunications, finance, and government. However, profit growth lagged revenue growth due to rising raw material costs and expanding sales and R&D expenses.
4. [para. 8][para. 9][para. 10] Moore Threads led in growth velocity. First-half revenue surged 147.4% year-on-year to 1.7 billion yuan, already exceeding its full-year 2025 revenue. As the only firm among the three developing multipurpose GPUs capable of handling both AI computing and graphics rendering, it narrowed its net loss attributable to shareholders to 11.6 million yuan from 270.9 million yuan a year earlier. Driven by large-scale computing cluster deliveries, Moore Threads is expected to hit a profitability inflection point, paving the way for a planned Hong Kong IPO it aims to complete within a year.
5. [para. 11][para. 12] The collective momentum highlights a fundamental shift in strategy. Rather than selling standalone chips, all three are increasingly delivering integrated turnkey solutions that combine hardware clusters with customized software. This system-level pivot aligns with booming domestic AI infrastructure spending. According to IDC, China’s acceleration server market is projected to expand from $37.7 billion in 2025 to $150.1 billion in 2030, representing a compound annual growth rate of more than 30%.
6. [para. 13][para. 14] Substantial risks remain. Inventory levels across all three companies ballooned during the first half, which should help make deliveries smoother, but a sudden slowdown in downstream demand or faster product iteration cycles could lead to severe write-downs that cut into profitability. Customer concentration is another lingering vulnerability, particularly for Cambricon, whose heavy reliance on hyperscale cloud clients makes its financial health highly sensitive to changes in their procurement strategies.
AI generated, for reference only
Sign in with Google
Sign in with Facebook
Sign in with 财新