0713: Don’t panic

I had plenty of time to watch the market today. I kept an eye on it from the morning pre-market call auction all the way to the close. For some moves on the board, the tracking and analysis were shared in real time in the comments section of the morning news roundup post, so I won’t repeat it here. [Taoguba]
First, the matter of breaking below the yearly line: in theory, if it breaks below and can’t reclaim it within three sessions, it officially enters a technical bear market. When I look back at several instances from years ago, there were two times where it failed to reclaim within more than three days. In both cases, a very long bear market really did follow. The most recent one was the tariff war on April 7 last year—right after grinding through three sessions at the bottom, on the 4th day there was a gap-up and it reclaimed, and then it launched a rally that lasted up to 11 months. Up until the Shanghai Composite hit a ten-year high. So this time, I’ll still judge it by the three sessions: if it can’t reclaim the yearly line within three days, then get ready to hibernate. But I think it’s unlikely. If I expand the logic, it would be a long story. In short, at present there aren’t internal or external factors or an environment that supports entering a bear market. There will most likely be a rebound with some persistence either tomorrow or the day after tomorrow.

The reason for today’s big drop is the combined effect of both internal and external factors.
Internally, I commented on it before and during the session. A “big player” effect from ChangxIN’s listing caused some steady capital to guard against a siphoning effect that would generate market volatility, so they retreated early to avoid adjustments. Another is that before the July 15 mid-year report mandatory pre-disclosure period, some funds left to avoid risks brought by the 龙头 (“dragon head”) blowing up. For institutions with larger scale, their controls over risk and volatility are strict; the impact from the listings of SpaceX and Hailisuo is vivid. There were also some ghost stories. In the post, it mentioned “Rubin delay.” Someone may not believe that Jensen Huang clarified it, but based on actual actions in the supply chain, TSMC’s advanced packaging has already increased its Rubin schedule by 150%. Downstream customers are also reducing GB300 orders and switching to adding more Rubin; the ODM side feedback says current progress is normal, and delivery can be done within this month. These are clear positive signals. The ghost story about Xuchuang that was explained once last night got a positive market response today—it held the whole day and refused to turn green. And another ghost story about Shenghong being kicked out of the supply chain due to quality issues was also punctured by the investigation results: it wasn’t that Shenghong’s PCB board was unqualified; instead, when a downstream customer, Foxconn Industrial Internet, processed it at its Thailand plant, the jigs had a problem—now it’s already been resolved. It’s easy to manufacture a ghost story, but the repair process after the impact takes time.

In short, the main internal cause is quant funds smashing the market by “selling bad news” ghost stories. The core internal cause of this round of sharp drop is quant funds’ disorderly harvest, which completely rewrote the trading ecosystem of A-shares today. Look at the intraday chart of Xuelong Group, which topped the popularity chart after-hours tonight—those crazy “ghost blade” pattern visuals are something else. Then compare it with the move in commercial aerospace that exploded collectively last Friday against today’s走势—quant is really too rampant.

Externally, the Korean stocks media are definitely set up to take the blame. The index fell 10%, Hailisuo fell 15%. Compared with prior highs, the recent drawdown is close to a half. The trigger was a report issued by Korea Investment & Securities (KIS), predicting that Micron’s (Hailisuo’s) earnings would be 8% below market expectations. The market worried that after Hailisuo releases earnings, it would see funds realize gains and the stock would fall as it did last time with Samsung—so they chose to leave early. Over there in Korea, I guess it’s similar to China’s A-share: some self-media cut up and selectively quoted to manufacture anxiety. In KIS’s report, it also emphasized that this is merely a normalization adjustment after being incorporated into long-term contracts. Enterprise value will shift from looking at “single-quarter price breakout power” to looking at “the sustainability of high profitability,” so valuation logic is being reshaped. But the market only sees stampede under panic. Some analysts believe Hailisuo’s case shows it’s not that the fundamentals collapsed—it’s the market re-pricing a “longer, more stable earnings cycle.” The short-term pain caused by switching business model might actually be what “digs out the golden pit.”
If Korea’s situation is like this, then how is the A-share any different? Today, the top 1,000 stocks fell more than 7%, a hundred stocks hit their daily trading limits down—mud and sand washed down together. Those stocks that were wrongly sold off, but still have firm fundamentals and solid order logic, will definitely show a repair later on.

After the market closed today, there’s a big piece of news—everyone should have seen it: the 2026 World Artificial Intelligence Conference will be held on the 17th.

This conference will see 300+ products make their global debut, with multiple listed companies bringing AI new products to the stage. Every year’s WAIC is a “side-by-side competition” for listed AI companies. From compute power chips to large models; from agent deployment to industry application solutions—leading companies all put their “back-pocket” new products onto the exhibition floor. The most eye-catching direction and the hottest investment mainline this year are domestic compute power, large models and terminal applications, as well as embodied intelligence. And regarding domestic compute power specifically, the main focus should be Huawei Atlas 950 SuperPoD ten-thousand-card-class super node real machine global first full offline exhibition—not an Ascend 950DT compute single card, but the full system rack for a ten-thousand-card liquid-cooled super node.

I believe this is also the main reason why the super node theme is broadly more resilient against declines today.

Positive sectors include:
High-speed connectors (the largest incremental piece; a super node essential requirement) -------- Lingqu 2.0 architecture comprehensively upgraded: 224G high-speed backplane / cables; the single-machine value is 5-8 times that of traditional servers. Usage in ten-thousand-card clusters surges:
Huafeng Technology 688629: Huawei’s first-tier primary supplier; the only mass-produced 224G connector in China; Hubble Investment holds shares; exclusive share of high-speed components for super node cabinets at 70%+; orders scheduled to 2027;
China Aviation Optical & Electronic and Aerospace Electric: liquid-cooled connectors and supporting high-speed cables second supply;
YIWA shares: high-speed copper cables and backplane supporting.

Optical interconnect / CPO/NPO (Lingqu’s all-optical core; optical module usage is 5-15 times that of a traditional cluster) -------- an 8192-card cluster needs massive 800G / 1.6T / 3.2T optical engine drives; near-packaged optical NPO is the 950 super node standard:
Foton Tech: Huawei gold supplier; core supply of 3.2T and 1.6T optical modules; exclusive matching of NPO near-packaged optical engine for Ascend super nodes;
Jingjie Xuchuang and Xin Yisheng: batch supply of 800G / 1.6T optical modules for Huawei Cloud and Ascend clusters;
Optoway Technology: supporting high-speed optical chips.

Ascend complete systems / servers (delivery carriers for ten-thousand-card clusters)

Gaoxin Development (Huakun Zhenyu): core partner for Ascend 950 super node complete systems; self-developed high-density liquid-cooled server racks; leading share in domestic training servers;
TuoWei Information: Huawei Ascend complete system contract manufacturer; core delivery party for Xiangjiang intelligent computing;
Foxconn Industrial Internet: ODM contract manufacturing of super node cabinets;
Unisplendour (H3C) and Inspur: liquid-cooled servers compatible with Ascend architecture; AI computing total-package.

Liquid cooling temperature control (950 fully liquid-cooled architecture; essential supporting requirement) -------- ten-thousand-card high-density cluster cooling relies completely on immersion / cold-plate liquid cooling; incremental surge:

InnoVans and Jialitu: liquid-cooled cabinets and refrigeration systems;
Gelan shares: immersion liquid cooling supporting.

Advanced packaging + HBM storage (Ascend 950 chip and HBM memory packaging essentials) -------- Ascend 950DT uses Chiplet and 2.5D advanced packaging; ten-thousand-card clusters support massive HBM memory:
Jentech Technology: core advanced packaging and testing for Ascend AI chips;
Tongfu Microelectronics: high-density GPU/NPU packaging;
Glorun Electronics and Deep Glitzy: AI storage and HBM module supporting;
Shen Nan Circuit and Xing Sen Technology: FC-BGA substrates.

All the recent declines are “sentiment kill,” “funds kill,” and “external kill.” This is absolutely not an industry-logic reversal or a deterioration of sector fundamentals. On the contrary, domestic AI compute power has positive catalysts that are being implemented intensively and persistently exceeding expectations. In the medium to long term, the certainty will remain the #1 across the entire market.

Other news during the day

About the mid-year report

This evening, fiber optics released their mid-year report:
Heng Tong Optoelectronics: expects net profit in the first half to increase year-on-year by 87%-121%. Q2 net profit expected to be 1.91 billion-2.463 billion, up 72%-122% quarter-on-quarter. The optical fiber segment is the main source of profit increase. Expectations of 2.2-2.5 billion also exist.
PCB and CCL performance is also pretty good:
Unisplendour (Shenzhen) Company: first-half net profit expected to increase year-on-year by 68%-78%. Q2 net profit expected to be 1.588 billion-1.758 billion, up 27.86%-41.55% quarter-on-quarter. The midpoint is slightly above expectations.
Shengyi Technology: expects first-half net profit to rise year-on-year by 117%-131%. Q2 net profit expected to be 2.46B-2.14 billion, up 67%-84% quarter-on-quarter. Above expectations. Shengyi Electronics: expects first-half net profit of 1.59B-1.76B, up 104%-114% year-on-year. Q2 net profit expected to be 637 million-692 million, up 43%-55% quarter-on-quarter.
Shen Nan Circuit: expects first-half net profit to rise year-on-year by 54%-69%. Q2 net profit expected to be 1.25 billion-1.45 billion, up 46%-70% quarter-on-quarter.
Jin’an Guoji: expects first-half net profit of 730 million-820 million, up 936%-1063% year-on-year. Q2 net profit expected to be 528 million-618 million, up 161%-206% quarter-on-quarter. Above expectations
HuaZheng New Materials: expects first-half net profit to rise year-on-year by 263%-380%. Q2 net profit expected to be 124 million-174 million, up 301%-462% quarter-on-quarter. The demand in the copper-clad laminate industry improves; actively expanding markets and timely capacity expansion to achieve a rise in both volume and price.

Test and packaging:
Huatian Technology: expects first-half net profit to rise year-on-year by 231%-275%. Q2 net profit expected to be 663 million-763 million, up 664%-779% quarter-on-quarter. Non-recurring items excluding 1.886 billion - 2.686 billion. The demand in the integrated circuit market rises, driving increased production scale and business revenue.
Bojie Shares: expects first-half net profit to rise year-on-year by 643%-816% for AI server-related testing solutions and MLCC equipment businesses; scale and returns improve in sync. Q2 net profit expected to be 84 million-119 million, up 27%-80% quarter-on-quarter. Above expectations
Liyang Microelectronics: expects first-half net profit of 85 million to turn loss into profit year-on-year; 12-inch silicon wafer production and sales volume grow significantly. Q2 net profit expected to be 78 million, above expectations
Beijing Junzheng: expects first-half net profit to rise year-on-year by 431%-531%. Q2 net profit expected to be 760 million-963 million, up 138%-202% quarter-on-quarter. Strong demand for storage chips; DRAM product prices rise by a large margin.

Other important news:
1、The State Council approved the 《Expansion of Consumption “Fifteen-Five” Plan》: by 2030, the residents’ consumption rate will increase significantly; the total retail sales of consumer goods by society will reach around 60 trillion. Continue to strengthen the development trend of a stable and positive capital market; increase residents’ income from property via multiple channels in urban and rural areas.
Commentary: A routine five-year consumption plan would only talk about subsidies, cultural tourism, home appliances, and domestic demand. Directly linking “stabilizing the stock market” with “increasing resident income and boosting consumption” into a national-level top-tier document is the biggest upside surprise signal this time. On the weekend, I shared a key takeaway from an interview with Morgan Stanley China’s chief economist. You really have to admire how these top-tier professionals conduct in-depth research and forward-looking analysis on the economy and policies. His expectations for this month’s important meeting content are almost implemented one by one.
2、The 《National Health “Fifteen-Five” Plan》 was released: supports innovation drugs and medical device development and applications along the whole chain; supports clinical use of innovative drugs.
3、The Ministry of Civil Affairs and 14 other departments released the 《Three-Year Action Plan (2026-2028) for Expanding Capacity and Improving Quality in Rehabilitation Assistive Devices Industry》: focuses on breakthrough technologies such as brain-computer interface, rehabilitation and care robots, and bionic prosthetics, and promotes the expansion and quality improvement of my country’s rehabilitation assistive device industry.
4、Commercial spaceflight:
Long March 5 Yao-14 carrier rocket arrives at Wenchang and will carry out the Chang’e 7 mission;
SpaceX’s Starship will conduct its 13th test flight on Thursday; 20 V3 satellites will be deployed for the first time.
5、ByteDance responds: there are no plans to do intelligent driving business. Previously, it was rumored that ByteDance is exploring entry into the autonomous driving field and that the Seed world model team would be responsible.
6、META announces it will add $40 billion in investment in a data center in Louisiana and expand that data center to 5GW of computing capacity.

Today’s trading: For today’s operations, I can say I absorbed the lesson from Friday. I noticed early warning signs were off and immediately cleared the position, reducing losses.
During the session, there were two “buy-the-dip” attempts. The first one buying Huahai Qingke was a bit too eager, but the trend was still resilient—when the broader market fell like this, it still repeatedly turned red.
In the afternoon, I bought Eisen shares.
One equipment, one material—I plan to allocate 30% of the position to “buy the dip” in domestic AI compute power tomorrow.

This afternoon, the babysitter called to say my dad was in a coma. She urgently called 120 and sent him to the hospital. I was busy until evening before I got home. Fortunately, his current condition is stable. There may be a temporary pause in updates afterward, so please forgive me, friends.