This year's market correction has hit nearly every segment of the tech sector hard.
Robotics names began sliding early in the year. Unitree's listing in August gave the group a brief lift, but the rally quickly reversed, with Unitree itself falling from 1,100 to around 400 at its lowest point. Although there has been a rebound recently, valuations across the robotics space still do not look particularly elevated on the whole.
New energy has been the most painful area. Many investors tracking production scheduling data saw very strong numbers, yet the share prices of the corresponding companies kept falling from May onward. CATL held up until the end of August before entering a waterfall-style decline, with a drawdown of roughly 50%.
AI hardware appeared to be the year's brightest star, but it suffered a sharp selloff starting in early July, followed by an even more violent drop in late September, wiping out profits for many and leaving them badly bruised. Looking at the degree of deleveraging in margin positions on AI hardware names, the carnage has been extreme. The STAR 50 Index fell as much as 48.5% over four-plus months, and among 650 electronics and communications stocks, six in ten were cut in half from their highs for the year.
Last week, everyone saw the central bank's moves: on Monday, three government departments gathered at the central bank to discuss funding; on Wednesday, the relending rate for share buybacks and increased holdings was cut to 1%; and on Friday, the State Council executive meeting again raised the need to improve the capital market's stabilization mechanism. The management's stance is now clear. At this level, we should no longer panic — instead, we should consider buying the dip.
So the question is: among these tech directions, which one should we buy?
First, the fundamentals.
Among these directions, the one with the weakest fundamentals is humanoid robotics. The hype around robotics is certainly real — today, for example, an Apple team open-sourced the HALO humanoid robot control system, which was directly deployed on Unitree's G1 for validation and achieved a 91% success rate on individual tasks across six unfamiliar environments. But it still has no fundamentals. It is the future, but the real arrival of its fundamentals is still very far away.
Robot penetration will follow a certain sequence: first industrial robots, where the factory environment is fully controllable and which have been penetrating for more than a decade; then semi-controlled environments such as hotels and warehouses; and finally fully uncontrollable environments — which is exactly what humanoid robots must face. Having humanoid robots work in fully uncontrollable environments is the hardest tier. Humans are incredibly complex organisms shaped by millions of years of evolution, and it would already be remarkable if humanoid robots could achieve half of human functionality. Until humanoid robots reach that level of usability, they are just oversized toys and will struggle to achieve commercial scale. Musk has said time and again how many Tesla robots would be mass-produced in a given year, but it has never been delivered. That is the reason. So humanoid robotics can be believed in as the future, but that future is distant. Current pricing is pure concept. Similar is commercial spaceflight, which is even more purely conceptual.
Turning to new energy — forget solar, given the overcapacity, and see whether it can be resolved after October next year. Here, let us only discuss lithium batteries. In A-shares, lithium battery companies are either CATL's suppliers or CATL's competitors, so the core company is CATL. And CATL's just-released third-quarter report is not an easy read: revenue for the first three quarters rose 56%, but net profit rose only 32%; in the third quarter alone, net profit growth fell to 14.5% and declined 5.8% quarter-on-quarter; gross margin was 20.69%, a new low since the fourth quarter of 2022; and single-quarter operating cash flow was only 1.4 billion yuan, compared with 21.9 billion yuan a year earlier. Volumes are rising while prices and profits are falling: lithium carbonate-linked pricing is compressing selling prices, consumption tax and cuts to export tax rebates are raising costs, and the rising share of energy storage is diluting gross margin. With these four squeezes stacking together, the fourth quarter is only the first full quarter of consumption tax collection, so the pressure has not yet fully played out. What worries the market more is the competitive landscape — automakers are moving away from CATL, and September new energy vehicle sales already turned negative year-on-year. After Democrats took control of Congress, the environment for new energy may improve, and that direction is not wrong. But policy implementation takes time, and we need to observe for a while longer.
There is not much to say about AI hardware fundamentals. Last week, Hua Hong released its third-quarter report, which beat expectations. Comparing the three lines, the one with the strongest fundamentals is still AI hardware. The STAR Market third-quarter reports have just been fully disclosed: third-quarter revenue rose 44.76%, and net profit rose 455%. This is the only one of the three directions where earnings are being concentratedly realized. But AI hardware itself must be further divided: the domestic chain and the overseas chain.
Second, why we are more bullish on domestic semiconductors.
The overseas chain follows the United States. Last week also saw a strong rebound after the central bank stepped in. Wednesday was normal, but on Thursday and Friday it became clear that at the current stage the domestic chain is better than the overseas chain, because things have changed in the United States. In the midterm elections, Democrats swept both the Senate and the House, and also won five governorships including Texas. Texas Governor-elect Inohosa, the first Democratic governor of Texas in more than three decades, made halting data center construction a core campaign promise. Last week she publicly stated: before protective measures are in place, data center-related projects must stop. How much does Texas matter? By Goldman Sachs' measure, Texas accounts for 26.7% of planned U.S. data center capacity, and the states Democrats won together account for 45%. Wells Fargo did the math: for every 10% reduction in Texas data centers, total U.S. AI capital expenditure falls by about 1.6%; if half the projects are canceled, the decline is about 8%. The market voted with its feet. On Thursday night last week, the Nasdaq plunged 5.52%, and the Philadelphia Semiconductor Index crashed 12.25% in a single day, the largest drop since March 2020, with Nvidia down 9% and AMD down 17%. The Nasdaq has now fallen for four straight weeks.
Beyond tracking U.S. volatility, the overseas chain itself has another worrying point: the competitive landscape. Whether optical modules, PCBs, or optical fiber, the capacity expansions attracted by the huge profits of the past two years are all on the way. Once this capex peak passes, no one can say clearly now who will survive and who will be eliminated.
The logic of the domestic chain is completely different. Domestic computing power will definitely continue to be built — water networks, new power grids, computing power networks, next-generation communications networks, urban underground pipeline networks, and logistics networks. These six networks are written explicitly in the 15th Five-Year Plan outline and special plans, and the National Development and Reform Commission estimates that related investment in 2026 alone will exceed 7 trillion yuan. Moreover, the 15th Five-Year Plan for the Information and Communications Industry specifically states: increase efforts to adapt domestic computing power chips. The more the overseas side tries to choke off supplies, the more the process of building computing power must push domestic semiconductor breakthroughs — this is the script the government most wants to see. So in terms of certainty, domestic semiconductors are greater.
In addition, there are trading factors at this level. What did the national team buy last week to support the market? Of the top ten ETF gainers across the market, eight were semiconductor-related; STAR 50 ETFs saw more than 20 billion yuan in net inflows in a single week.
Third, the bottom has likely appeared, and operations can retain some flexibility.
So whether from fundamentals or from the trading side, semiconductors have the greatest certainty at this level. The decline has been sufficient, the central bank has stepped back onto the table, and a phased bottom has likely already appeared. The worst case here is that the AI bubble truly bursts. But if it truly bursts, so what? Domestic semiconductors here can be compared to liquor in 2012. If it truly bursts, the industry's development is still far from its ceiling. With the index already down more than 40%, positions already trapped still have holding value. Just as with liquor back then, after it fell more than 40%, if you held on, the subsequent outcome was very good. AI in fact has not yet been falsified, and from both fundamentals and trading, the probability of a big rebound here is very high.
Of course, this is not a signal to go all in immediately, because overseas risks have not been lifted. First, the chart for U.S. AI hardware looks very ugly: the Philadelphia Semiconductor Index fell 12% on Thursday, rebounded on Friday night, but not strongly, and tonight the futures still look set to fall. Second, how the U.S. and Iran will proceed and where oil prices will go remain uncertain — last week oil prices fell 13.6% in a single week, with Brent crude posting its largest weekly dollar decline since 1988, yet U.S. forces were still striking Qeshm Island last Thursday, and Iran immediately lifted missile range restrictions. Fighting while talking remains the baseline scenario; at the same time, a stronger-than-expected nonfarm payrolls report pushed the probability of a December rate hike to 75%. Third, the Fed's December policy meeting is still ahead. Considering these risks, new positions here can be added slowly, not all at once. After all, if U.S. stocks have not stabilized, A-shares will find it difficult to embark on an independent trend rebound, let alone a reversal opportunity.
In the end, opportunities are created by declines, but bottoms are ground out. The right posture at this level is to hold cheap, good chips and slowly wait for the overseas stones to fall. If you feel confused about industrial investing; if you feel lost about how to find industrial leaders and potential companies; if you feel at a loss about sector volatility. Then please scan the code to join us, obtain more detailed industry chain analysis and investment strategies and, together with Gelonghui Research Institute, seize opportunities in the wave of investment and share in the dividends of the industrial revolution.
Comments