Home » Uncategorized » Morgan Stanley Slashes SMIC Huawei AI GPU Revenue Forecast Over Poor Yields

Morgan Stanley Slashes SMIC Huawei AI GPU Revenue Forecast Over Poor Yields

by ytools
3 comments 0 views

Morgan Stanley has cut its forecast for Huawei’s AI chip revenues produced at Semiconductor Manufacturing International Corporation (SMIC) by more than half, citing severe yield problems at China’s most advanced chip foundry. This revision not only reflects technical limitations but also highlights the geopolitical and industrial pressures shaping the global semiconductor industry.

SMIC has been at the center of China’s push for semiconductor self-sufficiency, particularly after the United States imposed export restrictions that cut the firm off from the latest extreme ultraviolet (EUV) lithography machines produced by ASML.
Morgan Stanley Slashes SMIC Huawei AI GPU Revenue Forecast Over Poor Yields
Without EUV, SMIC is forced to rely on deep ultraviolet (DUV) systems and multi-patterning techniques, which significantly complicate manufacturing and reduce efficiency. This constraint has left SMIC struggling to achieve yields competitive with Taiwan Semiconductor Manufacturing Company (TSMC) or Samsung.

According to Morgan Stanley’s report, SMIC will likely manage to produce only about 18,000 AI chip wafers monthly by 2027, far below what analysts previously expected. In the near term, the bank projects that by the end of this year SMIC’s usable yields for Huawei’s AI chips will hover around just 30 percent. In semiconductor terms, yield is the proportion of chips per wafer that meet the required quality standards. Low yields mean higher costs and fewer working chips reaching the market, undercutting both profitability and competitiveness.

At present, SMIC is expected to output roughly 7,000 wafers per month, focused on Huawei’s Ascend 910B processors, before transitioning to the more advanced Ascend 910C in 2026. The 910C, composed of two 910B dies, doubles the complexity of manufacturing and packaging. On a 12-inch wafer, SMIC can fit 78 dies of the 910B or 39 dies of the 910C. Pricing reflects this difficulty: Morgan Stanley estimates each 910B will cost Huawei RMB 50,000 this year, while the 910C will come in at RMB 110,000 due to added die and packaging expenses.

The financial projections in the report show a steep decline from earlier expectations. SMIC’s revenues from Huawei’s AI GPU wafers are now projected at RMB 58.5 million in 2025, RMB 94 million in 2026, and RMB 136 million in 2027. Earlier estimates were far more optimistic, at RMB 146 million, RMB 212 million, and RMB 286.5 million, respectively. The gap underscores how crucial yield improvements are: without EUV, SMIC must rely on incremental process enhancements to push its yields from 30 percent today toward 70 percent by 2027. Whether that trajectory is realistic remains an open question.

This struggle places Huawei and SMIC in a difficult spot. On the one hand, China’s tech champions are determined to compete with NVIDIA, which dominates the global AI chip market with its cutting-edge GPUs. On the other, the technical limitations of China’s domestic fabs mean costs remain prohibitively high. A single Huawei AI processor costing upwards of RMB 100,000 illustrates how geopolitical restrictions ripple down to the price tag of actual hardware.

Critics argue that Morgan Stanley’s report is not just about economics but also about market signaling. Some observers believe Western financial institutions downplay Chinese progress to protect established players like NVIDIA. Others point out that, regardless of intent, the numbers show that China’s chip industry still faces a steep uphill battle. SMIC may be China’s answer to Intel, but without access to EUV technology, its growth potential is capped by physics and process complexity.

Whether yields improve as projected will determine how far Huawei can scale its AI ambitions. If SMIC manages to hit 70 percent yields by 2027, its competitiveness will strengthen, but until then, Huawei’s reliance on high-cost chips could limit adoption outside government-backed projects. In the meantime, NVIDIA remains well-positioned to dominate, though the very existence of a Chinese alternative ensures the global chip race remains intense.

You may also like

3 comments

DevDude007 January 3, 2026 - 7:50 pm

expect same treatment when alibaba or broadcom start eating shares

Reply
Speculator3000 January 5, 2026 - 10:50 pm

banks always do this gatekeeping crap when nvidia feels heat

Reply
Fanat1k January 25, 2026 - 12:50 am

huawei chips super expensive rn but china will keep pushing

Reply

Leave a Comment