This summary is sourced from the internet. If the organization believes it is inappropriate to disseminate, please leave a message in the backend of the 【Wenbagu Research Summary】 public account to delete the summary! We apologize for any inconvenience caused!
Pay attention to the accelerating trend of domestic ASICs❗️❗️
#Focus on the core changes of companies under the ASIC industry trendThe ASIC industry trend is confirmed, with a scarcity of forward-looking key companies, high domestic internet demand, and significant potential.
Industry chain research shows that the progress of self-developed chips is actively advancing, with a high increase in supply chain verification demand, accelerating orders, and key focus on core positioning companies!
#High prosperity in the ASIC track should be graspedBroadcom’s AI custom chip demand is strong, and the company previously stated that three major clients will bring a market of $60-90 billion in FY27, excluding the explosive demand for inference and contributions from new clients, with actual expectations for FY26/27 exceeding previous forecasts.
#Domestic 3D DRAM stacking cloud computing new high-performance solution3D DRAM is a definitive technological path, offering high bandwidth, low power consumption, and high flexibility, which is essential for domestic computing power.
Domestic manufacturers are laying out 3D solutions to replace HBM, applied in cloud computing high-performance fields, with performance exceeding expectations and accelerating volume production.
📈 It is recommended to pay attention to: 【Chipone Technology】, 【Aojie Technology】, 【Canxin Technology】, etc.
Rumors of increased orders for Chipone Technology have been verified, and the trend of AI ASICs continues to be confirmed! Furthermore, it is further confirmed that the upper management is resolute about domestic computing power.
In the coming week, we believe that the AI ASIC sector is likely to lead the market, followed by a shift to the GPU segment.
The latest changes are as follows:On Wednesday, the Financial Times reported that China has requested its tech companies to stop purchasing all AI chips from NVIDIA.
On Friday, we verified industry news that indeed, domestic CSPs are no longer allowed to place new orders for H20 this week.
The new domestic CSP computing power demand will turn to “increase orders” from domestic suppliers such as Cambrian, Haiguang Information, and Chipone (in fact, rumors of increased orders for Cambrian had already emerged on Wednesday). On Friday, news indicated that Chipone’s ZJ project mass production orders for next year would increase by 50% on top of the original basis.
From the perspective of Chipone Technology’s stock, a valuation of 100 billion will be the most solid bottom line, with a neutral view at 150 billion and an optimistic view at 200 billion.
After the mid-term report, the official disclosure of new signed orders solidifies the valuation belief, and Chipone Technology has now entered the “second phase”.
#We had previously hinted on Friday that it is advisable for leaders to get on board and stay steady, as more positive news is yet to be released, and more industry changes await exploration.
Barclays: Everyone is still underestimating, stop arguing about ASIC vs GPU.Our view: Just a few weeks ago, the latest craze in the AI field revolved around the $10 billion cooperation agreement between Broadcom (AVGO) and OpenAI (see our thoughts after the earnings report here).
Since the agreement was reached, industry discussions have shifted to “Is custom ASIC the future?”
Especially as NVIDIA (NVDA) reported slightly weak data in its computing business during the earnings report.
Today’s announcement indicates that we are in a golden period of computing demand explosion—although the announcements of custom chips are leading the way, general-purpose chips will still bear the majority of the workload for OpenAI (a non-public company, not within the research coverage) at least until the end of this decade.
According to NVIDIA’s Q2 FY2026 earnings report (full analysis here), a 1GW installation is equivalent to bringing NVIDIA over $35 billion in revenue, which means that deploying 10GW of NVIDIA systems in cooperation with OpenAI will generate over $350 billion in value.
NVIDIA’s scale is 3.5 times that of OpenAI’s ASIC project next year ($10 billion vs. $35 billion).
Clearly, in the current environment, both custom computing and general computing demands are peaking, but we believe the market’s recognition of NVIDIA is still insufficient—this deal is entirely incremental business, potentially adding an additional $35 billion to Wall Street’s expectations (including computing and networking businesses).
Follow our public account for updates on the latest institutional research.
Click “Read the original text” to jump to the 【Wenbagu Research】 PC version.