Original source：Morgan Stanley Asia LimitedFirst published on Unhidea：Original source date：
Key takeaway
Morgan Stanley remains positive on growth in Hygon Information’s CPU and GPU businesses, but a higher GPU revenue share and increased minority interests prevented the revenue forecast increase from translating into higher earnings attributable to the parent company.
Why were earnings forecasts cut as revenue expectations rose?
Morgan Stanley updated its model based on the 2026 third-quarter results and cut its earnings forecasts for 2026, 2027 and 2028 by 14%, 14% and 12%, respectively. It lowered its price target from 430 yuan to 380 yuan and maintained its Overweight rating.
The bank raised its GPU average selling price assumptions because of rising high-bandwidth memory (HBM) prices, and consequently increased its revenue forecasts for 2027 and 2028. However, the higher GPU revenue share lowered its overall gross margin assumptions.
CPU business growth was stronger than the bank had previously expected, also increasing minority interests in the model and further reducing its forecasts for earnings attributable to shareholders of the listed company.
How does the shift in growth drivers affect gross margin?
Morgan Stanley expects GPU/DCU revenue to grow by 83.7%, 107.4% and 48.8% in 2026, 2027 and 2028, respectively; CPU revenue is expected to grow by 32.8%, 15.0% and 28.6% over the same period. Faster GPU/DCU growth means its share of revenue will rise.
The bank’s model projects overall gross margin to fall from 57.8% in 2025 to 56.7% in 2026, then to 52.2% in 2027 and 49.7% in 2028. Revenue growth and a decline in overall gross margin can therefore occur at the same time.
What growth conditions underpin the maintained Overweight rating?
Morgan Stanley is optimistic about expanding demand for AI computing power in China and the domestic substitution of CPUs and AI GPUs. It also believes the development of agentic AI applications will increase the role of CPUs in next-generation AI system architectures.
The bank’s base case assumes a compound annual revenue growth rate of 48% from 2025 to 2028, supported by generative AI infrastructure buildout, purchases of cloud training and inference chips, and an increased share of high-performance AI chips.
In the downside scenario, if domestic AI investment and purchases by major customers are slower than expected, market share is lost, and price competition, wafer manufacturing costs and capacity utilization weigh on earnings, the compound annual revenue growth rate from 2025 to 2028 could fall below 30%, and gross margins in 2026 and 2027 could fall below 40%.
