"This is another DeepSeek moment for semiconductors" — Bernstein's chief analyst Qingyuan Lin
On May 25, 2026, at the Shanghai IEEE International Circuits and Systems Symposium, Huawei's Tingbo He introduced the "τ law," sending the market into a frenzy. On May 26, the semiconductor sectors in both the A-shares and Hong Kong stocks experienced a split market — the divergence after the good news was fully priced in presents the best window for calm assessment.
1. Bernstein's qualitative analysis: another DeepSeek moment
On May 25, 2026, Bernstein's chief analyst Qingyuan Lin released a research report, making a highly impactful qualitative judgment: "another DeepSeek moment for China Semis." They maintain an Outperform rating for the Chinese semiconductor sector, primarily based on the fact that Huawei's τ law demonstrates China's ability to continue advancing performance upgrades without relying on EUV. This marks another signal-level breakthrough in semiconductor infrastructure, following the wave of local full-stack investments triggered by algorithmic breakthroughs in AI large models during the DeepSeek era.
The research report provided two sets of core technical validation data:
LogicFolding achieves a 1.5μm hybrid bonding pitch, whereas TSMC's SoIC is currently in large-scale production (HVM) at the 6μm level.
Cell-to-cell stacking boosts the density of the same node by 53.5% to 238 MTr/mm², meaning that Huawei has achieved a significant increase in equivalent density through architectural innovation at the same process node.
Bernstein simultaneously provided three major recommended targets and clearly defined the layers of beneficiaries: equipment chain > foundry > fabless. Specifically:
Target code logic positioning target price: SMIC 688981 / 0981.HK advanced logic + packaging manufacturing executor, DUV multiple exposure continuously benefits CNY120 / HKD70; Northern Huachuang 002371 advanced logic process equipment leader, exposure higher than AMECO CNY680; Tuojing Technology 688072 potential LogicFolding bonding equipment, viewed as core beneficiary CNY580.
Data source: Bernstein Research Report dated 2026.05.25
The research report also provides three major constraints that deserve high attention:
In the 3DIC packaging field, TSMC still leads—in terms of advanced packaging production scale, EDA toolchain, IP ecology, and yield control, TSMC has a deep accumulation.
Stacking multiple dies increases power density, making heat dissipation a key bottleneck—when two or even multiple layers of logic chips are vertically stacked, thermal density rises sharply, and if heat dissipation technology does not break through, the actual working frequency of the chips will be severely constrained.
Yield and cost are the barriers to scaling—3D stacking is essentially a 'multiplicative effect', where the yields of multiple layers multiplied will significantly lower the final yield, resulting in exponential cost increases.
The report also made key judgments: 'innovations from Huawei are the ones global players can copy', but global leading manufacturers also have the ability to learn from and even replicate this technological path. In the short term, Huawei still lacks EUV, and the gap will gradually narrow rather than disappear immediately.
Bernstein maintains an Outperform rating on AI fabless (Cambrian, Haiguang Info) but notes 'upside smaller'—the architectural dividends support performance extension but are directly confronted by competition from Huawei itself. In other words, the τ law holds up the performance ceiling, but those designing chips may be 'eaten' by Huawei, which is building its own ecosystem.
Two, market review: significant divergence has emerged
2.1 May 25: Explosive buying
After the release of the τ law on May 25, funds rushed into the semiconductor and communication sectors. SMIC in Hong Kong (0981.HK) saw a significant rise on May 25, closing at 79.85 HKD, with a sharp increase in turnover. In the A-share market, SMIC (688981) had a financing balance of 14.969 billion yuan that day, a week-on-week increase of 6.97%, with a net buying of 975 million yuan, continuing to grow for two days, ranking fifth among 405 companies in the electronics industry for net buying.
2.2 May 26: High opening followed by a low close, with intense tug-of-war between bulls and bears
On May 26, after the market opened, the semiconductor sector in Hong Kong and A-shares showed severe divergence:
In the Hong Kong market: SMIC (0981.HK) gapped up 14.97% to 91.80 HKD, once soaring 16.47% to 93 HKD during the day, then quickly retraced, closing at 84.40 HKD, with the rise narrowing to 5.7%, and an intraday volatility of 11.15%, with trading volume expanding to 372 million shares and a transaction amount of 32.58 billion HKD, with fierce fighting between buyers and sellers, net inflow of main funds of 806 million HKD.
In the A-share market: SMIC (688981) closed at 149.18 yuan, plummeting 4.37%, with an intraday volatility of 5.73% and a turnover rate as high as 9%, with a transaction amount of 26.787 billion yuan, forming a typical pattern of high opening followed by a low close with significant volume drop. Northern Huachuang (002371) closed at 670.25 yuan, down 4.00%, with a transaction amount of 9.988 billion yuan and a turnover rate of 2.05%. Tuojing Technology (688072) was an outlier, closing at 750.00 yuan, up 4.02%, with a turnover rate of 4.39% and a transaction amount of 8.998 billion yuan, contrasting sharply with SMIC and Northern Huachuang.
In the AI fabless sector: Cambrian opened down 1.67%, closing down about 0.82% at 1280.28 yuan; Haiguang Info opened down 2.03%, closing down approximately 2.4%. The sentiment for second-tier fabless companies is significantly weaker than that for equipment and foundry segments, consistent with Bernstein's 'upside smaller' judgment.
2.3 Market interpretation
The market characteristics on May 26 were very clear: after good news was announced the next day, the market experienced significant divergence, with bulls and bears fiercely clashing. Most of those who jumped in early to chase prices got trapped by intraday pullbacks. This is a classic 'Buy the rumor, sell the news' scenario—expectations from the significant rise on May 25 had largely been priced in, and the tug-of-war on May 26 was essentially a game between profit-takers and believers in the good news.
If this kind of turmoil continues, it is highly likely to end with a significant correction in the short term—only with a large number of profit-takers exiting can the selling pressure decrease, allowing the sector index to stabilize.
Three, what is the τ law? From 'smaller' to 'faster'
3.1 The dilemma of Moore's Law
The core of the traditional Moore's Law is 'geometric miniaturization': the number of transistors doubles every 18-24 months, performance doubles, costs decrease, and the core method is to make transistors smaller and smaller (the process moves from micron level to nanometer level). But the problem arises: after 7nm, the miniaturization benefits sharply diminish, and the design budget for the 2nm node has already exceeded $1 billion. Lithography machines are nearing their physical limits, 'being able to make things smaller until they can no longer be shrunk'.
3.2 Core logic of the τ law
The τ law proposed by Huawei revolves around 'time miniaturization' replacing 'geometric miniaturization'. τ is the Greek letter representing time constant (system response speed), with a smaller τ value meaning a faster system response. The goal is no longer to continue to shrink transistor sizes but to compress signal transmission delays across four levels: devices, circuits, chips, and systems, enhancing overall system efficiency.
The key measure is 'Logic Folding': folding logic units originally laid out on a two-dimensional plane into a 3D stacked structure, using 'elevators' for vertical transmission instead of long-distance horizontal 'walking', which significantly shortens signal paths. At the same time, in conjunction with Huawei's unified bus (reducing remote access latency from tens of microseconds to 150 nanoseconds), HiONE optoelectronic synergy, and other system-level optimizations, τ values are continuously compressed.
3.3 'Breaking the deadlock' and 'copying homework'
The most exciting aspect of the τ law is that it bypasses the EUV blockade, proving that even without the most advanced lithography machines, continuous improvements in chip performance can still be achieved through system-level innovation and architectural restructuring. This is a plan for 'overtaking on a curve'.
But looking at it calmly, the τ law is more of Huawei's refinement and theorization of practices from the past six years—He Tingbo disclosed that based on this law, Huawei has designed and mass-produced 381 models of chips over the past six years. Chip design and system architecture innovation, especially optimization methodologies at the system level, can also be imitated and copied by global giants. This is Huawei's innovation forced by sanctions, which does not inherently constitute a moat against Chinese semiconductor peers (especially fabless)—Huawei's architectural innovation will ultimately be reflected in Huawei's own chips, while fabless must face traditional competitors as well as Huawei's 'in-house chip production'.
Four, industry voices: differentiation consensus under highly consistent expectations
4.1 Institutions are generally bullish
Over the past two trading days, more than a dozen institutions released research reports overnight, with a highly consistent judgment:
CITIC Securities: The τ law will bring profound changes across four levels: transistors, circuits, chips, and systems, optimizing system topology to bridge the gap in short-term process nodes. China's semiconductors are expected to accelerate development through a change in path.
Huatai Securities: The τ law is essentially an 'evolutionary version' of the current globally popular system synergy optimization methodology in the semiconductor industry, but Huawei has theorized it into a predictable, iterative evolutionary framework.
Zheshang Securities: The τ law is expected to drive technological updates across the industry chain, recommending focus on foundries, advanced packaging and testing, and equipment sectors. Logic folding and 3D folding will systematically boost the demand for copper plating equipment, CMP equipment, hybrid bonding equipment, and consumables.
Guosheng Securities: The τ law is the first new principle proposed by China in the global semiconductor field to guide industrial development, marking China's leap from 'follower' to 'leader', with the potential to reshape the global competitive landscape.
Open Source Securities: The τ law will catalyze three major sub-sectors: optical communication, liquid cooling, and domestic computing power. Liquid cooling will shift from 'optional' to 'mandatory'—NVIDIA's GB300 TDP has reached 1400W, and Huawei's Ascend 950PR processor power consumption is 600W, with power consumption expected to continue rising under 3D stacking; the 80kW limit of air cooling single cabinets is hard to support.
4.2 But there are hidden differentiations within the consensus
Behind the collective optimism, there are three warning signals that should not be ignored:
First, 'consensus' itself is a risk signal. Overnight, more than a dozen institutions all issued bullish reports, which historically in the A-share market is often not a bottom signal but an over-exuberance signal. The historical pattern is: when everyone thinks 'this is a long-term opportunity,' it is highly likely that short-term they will face severe corrections. The surge in margin trading balances is a typical precursor to 'smart money taking the initiative, and latecomers picking up the pieces.'
Secondly, the question of 'Can I chase this?' has been asked too many times. From the end of trading on May 25 to before the market opened on May 26, social media and investment forums were flooded with questions about whether it was still possible to buy. This collective anxiety usually signals that good news has been adequately priced in. Analyzing from a game theory perspective, the information in the research report had already been reflected in the market, and when the news became public, smart money was entering the cash-out phase.
Thirdly, divergence has already begun. On May 26, the semiconductor sector collectively opened lower (SMIC A-shares opened down more than 3%), and after a high-level volatile surge, gradually retreated. If it was merely emotion fermentation, the opening should have continued to be strong. The low opening and volatility indicate that bulls and bears have entered substantive game.
Five, distinguishing the genuine from the false: who is 'riding the wave' and who truly has α?
5.1 Fourfold screening framework
In the face of the sector's momentum driven by the τ law, it is essential to conduct a fourfold screening of targets: 'positioning, technology, performance, valuation'—at least three criteria should be met to warrant attention; otherwise, avoid them at all costs.
① Positioning: Equipment chain (Northern Huachuang, Tuojing Technology, Zhongwei Company) > Foundry (SMIC, Huahong Company) > Fabless (Cambrian, Haiguang Info, etc.). Bernstein's α stratification judgment has been made clear, and combined with Zheshang Securities' point that 'multi-layer vertical heterogeneous architecture places higher demands on equipment precision, which will systematically drive the demand for TSV etching equipment and CMP equipment', the equipment segment is the most direct carrier for the commercial implementation of the τ law, with the strongest rigid demand.
② Technology: Companies with actual technical reserves in areas related to logic folding and 3D stacking differ vastly from those merely 'riding the labels'. Tuojing Technology has a complete product line in 3D integration, including PECVD, ALD, SACVD, with a total of 2140 patent applications (including PCT) and 362 granted invention patents, resulting in high hardware barriers. Cambrian and Haiguang Info, while benefiting from the performance ceiling opened by the τ law, face competitive pressure from Huawei’s self-developed AI chips, and the architectural dividends may be 'eaten' by Huawei.
③ Performance: Changxin Technology's Q1 2026 revenue was 50.8 billion yuan, a year-on-year increase of 719%, with a net profit of 33 billion yuan, a year-on-year increase of 1268%, corresponding to the explosive growth in AI and advanced packaging demand, indicating that the advanced testing and storage track already has real industrial prosperity support, rather than being purely concept-driven.
④ Valuation: Needs to be viewed calmly. SMIC's A-share dynamic P/E ratio is 219.55 times, static P/E ratio is 237.15 times, and the current financing balance ratio is relatively high, posing significant risks for short-term chasing. Tuojing Technology's 52-week increase has reached 410.24%, with over four times the increase itself indicating that a significant portion of the potential good news has already been priced in. Northern Huachuang's 52-week increase is 112.15%, and its valuation is also at historical highs.
5.2 Layering of beneficiaries
In the long run, beneficiaries should be companies with systemic innovation capabilities, not just speculators who 'hoard labels':
On the equipment side (Northern Huachuang, Tuojing Technology, etc.): direct positioning with the most rigid demand, Bernstein's exposure judgment has confirmed this.
Foundry/testing (SMIC, Changdian Technology): Once the τ law is widely commercialized, demand for advanced packaging and wafer foundry will surge, but attention should be paid to the fact that the current stock prices already contain relatively optimistic expectations.
Fabless: The τ law has broadened the performance ceiling for domestic chips, but the competitive landscape will become increasingly fierce—Huawei is both the proponent of the standards and the biggest beneficiary, and fabless companies must find differentiated survival space, or else the 'upside space will be limited.'
As for those trying to ride the hype with weak technical reserves, once the sector corrects, they are likely to be the first ones 'caught swimming naked'.
Six, operational thoughts: sell on rebounds, no adding positions unless there’s a significant drop
The strong rise on May 25 and the sharp fluctuations on May 26 clearly outline the current market state: consensus formation, but the consensus has largely been reflected in the price.
For the time being, the semiconductor and communication sectors are likely to exhibit a pattern of 'ups and downs, overall sideways movement':
Maintain a core position (large leading stocks);
For the remaining positions, engage in swing trading, reduce positions on rebounds, and do not add positions unless there’s a significant drop;
To trade on rebounds, be 'fast, accurate, and ruthless'—the τ law is a long-term logic, but the sector's short-term fluctuations could be very intense.
Regarding AI fabless (like Cambrian and Haiguang Info), Bernstein's judgment of 'upside smaller' is worth paying attention to—valuation expansion is temporarily under pressure, not a direction that funds prioritize attacking.
Seven, long-term perspective: paradigm watershed at the cycle level
Setting aside market sentiment, the essence of the τ law is that it shifts China's semiconductor evolution path from 'process catching up' to 'system innovation', adding narrative dimensions and strategic depth.
Bernstein likened it to 'another DeepSeek moment'—analogous to the beginning of 2025 when DeepSeek triggered a wave of domestic AI investment frenzy through algorithm efficiency breakthroughs, the τ law may similarly trigger a reassessment of China's semiconductor full-stack capabilities.
But two risks need to be noted: heat dissipation and yield. The thermal density issue of 3D stacking could limit the actual working frequency of chips, making them more suitable for AI inference rather than high-frequency CPUs; the 'multiplicative effect' of multi-layer yield means that the yield of a single layer must be close to 100%, otherwise the costs after scaling will be hard to bear—this is the core obstacle for the τ law to go from laboratory to industrialization.
'Tao' refers to the tao of nourishing one's strength, not 'emptying one's pockets'. The stock market performance of A/H shares on May 26 confirms this rule: the short-term gains from speculative hype are hard to penetrate through the long-term fog of industrial evolution. Those that can truly transcend cycles are the enterprises that can transform the τ law into actual technology and performance—the equipment sector is the most certain beneficiary, followed by the foundry sector, while fabless must seek differentiated space under Huawei's shadow.
'Replacing space miniaturization with time miniaturization and covering single-point breakthroughs with system optimization'—this is the core thought of the τ law, and also the cognitive framework that investors should maintain when examining this round of China's semiconductor market. The Chinese semiconductor industry undoubtedly stands at a paradigm watershed at the cycle level, but those truly capable of going far are only those with solid positioning and deep technological accumulation. α will emerge after β recedes, and what is needed most at this moment is not impulse but calm.