Sugon H1 Net Profit Up 33%, Q2 Quarterly Surge of 226% QoQ
Nashnova编辑部
Sugon (中科曙光, 603019) posted RMB 9.71 billion in H1 attributable net profit, up 33% YoY, with Q2 alone surging 226% QoQ — the acceleration points to concentrated AI server deliveries rather than seasonal lift.
How much did Sugon earn, and where did it come from?
H1 revenue reached RMB 7.466 billion, up 27.62% YoY; attributable net profit hit RMB 9.71 billion, up 33.31%.
Non-recurring-adjusted net profit grew 48.45% — faster than headline profit. This means → the growth is driven by recurring core operations, not one-off gains like asset sales or subsidies.
IT equipment revenue was RMB 6.666 billion, or 89.3% of total revenue; gross margin rose to 27.23%, up 0.58 percentage points YoY.
Why did Q2 profit spike so sharply?
Q2 net profit came in at roughly RMB 744 million, versus just RMB 228 million in Q1 — a QoQ jump of about 226%.
In plain terms = one quarter earned more than triple the previous one; that far exceeds normal seasonality.
Analysts attribute the spike to concentrated AI server order deliveries and revenue recognition in Q2. This reflects Sugon's AI business entering a volume-shipment phase.
Why is Hygon the "profit engine" behind Sugon?
Hygon Information (海光信息), an associate company, posted H1 revenue of RMB 9.099 billion and net profit of RMB 1.798 billion, up 49.67% YoY.
Sugon booked RMB 504 million in equity-method investment income — nearly half of its own attributable profit. This means → Hygon's profitability directly sets Sugon's earnings ceiling.
The link goes beyond equity stakes: Hygon's DCU — deep computing unit, a domestically designed AI training and inference chip — is the core hardware inside Sugon's AI clusters. Sugon both sells Hygon's chips downstream and shares its profits upstream, binding the two across the supply chain and the income statement.
What do the "100,000-card cluster" and the strategic pivot signal?
In July 2026, Sugon announced the completion of "Sugon 8000 (Dengfeng)" — China's first AI supercluster claiming a fully domestic technology stack at the 100,000-card scale, a generational leap from the previous 10,000-card tier.
Senior VP Li Bin stated publicly that the strategic focus has shifted from "training-first" to "inference-service-first", with the performance yardstick moving from "card count" to "token-generation efficiency".
In plain terms = training is like building a factory once; inference is the factory shipping product every day. Inference serves end-users continuously, making revenue more recurring and customers stickier — a signal of the shift from selling hardware to selling services.
What do R&D spending and cash flow reveal?
H1 R&D spending reached RMB 1.094 billion, up 78.36% YoY, lifting R&D intensity to 14.65% of revenue. Convertible-bond proceeds are earmarked for AI compute clusters, training-inference integrated appliances, and domestically sourced storage.
Operating cash flow was negative RMB 187 million, but the deficit narrowed by RMB 1.195 billion YoY. The drag came from inventories rising RMB 3.259 billion YoY as the company pre-stocked for high-end compute orders.
This reflects a classic cycle: stockpile first, deliver next, recognize revenue later. A sharp inventory build often foreshadows revenue release in subsequent quarters — provided the orders actually land.
Can the full-year target be met?
On August 17, Kaiyuan Securities raised Sugon's 2026 full-year earnings forecast from RMB 2.639 billion to RMB 3.472 billion.
To hit that target, H2 net profit would need to reach roughly RMB 2.5 billion — more than 2.5 times the H1 figure.
This means → the full-year number hinges on two variables: the pace at which "Sugon 8000" contracts convert into recognized revenue, and the scale of Hygon's H2 profit contribution. A delay in either node could discount the annual result.
Content is for reference only, not financial advice.