CPU / DCU Design

Sitting by the window at a café, I ran the numbers from the past four quarters through my head and wanted to share my view on the next two quarters with you. Let's start with revenue. I'm assuming the pace of domestic substitution keeps accelerating—XinChuang procurement expands from Party and government agencies into key sectors like finance, telecom, and energy. With Hygon CPU's installed-base replacement plus incremental DCU contribution in AI compute scenarios, I'm modeling roughly 5.2 billion for 2026Q2, continuing the sequential climb from 2026Q1, primarily because Q2 has historically been the dense delivery and procurement landing window; then another step up to around 6.1 billion in 2026Q3, since internet and AI compute customers typically begin ramping up orders in Q3, and the new-generation DCU enters its mass-shipment window. This pace carries an embedded assumption: the x86 ecosystem lock-in effect continues to hold, customers don't migrate at scale to other domestic platforms, and there are no hard supply-chain disruptions. Now on profit. Gross margin rebounded from 53.4% in Q4 to 55.6% in Q1—I judge this as a one-off distortion from product mix and year-end settlement noise, with a moderate recovery ahead: 56% for Q2, 57% for Q3—driven by a rising share of high-end DCU, yield improvements from scaling up production, and continued concessions on packaging and storage procurement costs in negotiation. I'm modeling operating margin back to the 26%–27% central band, where scale effects dilute the three expense ratios; net margin stays at 18%–19%. Behind these two numbers sits another judgment: R&D spend continues to grow in absolute terms but won't double like revenue does, so expense ratios naturally come down. The biggest risk I have to flag separately: capacity at advanced-node foundries and HBM/high-bandwidth memory supply. If the U.S. tightens further, or if any link in the domestic substitution supply chain breaks, gross margin and shipment cadence both get pierced directly—this isn't a probabilistic event, it's a sword of Damocles hanging overhead. The second risk is customer concentration: procurement volatility from top customers will significantly amplify quarter-to-quarter noise in the numbers. So as you can see, the curve I'm drawing isn't aggressive. It rests on two assumptions holding simultaneously: sustained policy-driven demand, and no supply-chain surprises. If either assumption loosens, the curve gets redrawn.