The tech sector rally is not over yet, and the market leadership that drives the gains is about to shift.

In the past few months, the market has been debating a question: has the tech rally come to an end for good?

Bearish investors argue that institutional investors are cutting their positions, margin trading balances are flowing out, the Nasdaq is declining, the semiconductor sector has posted eight consecutive days of losses, and the ChiNext Index is showing a weak technical pattern.

Bullish investors point out that the trend of the AI industry remains unchanged, the State Council is still rolling out arrangements for the "Artificial Intelligence +" initiative, leading tech companies are conducting share repurchases, and broad-based ETFs are seeing rising trading volumes.

However, if this round of tech rally is not over, will the leading driver that has been propelling the market be replaced by a new one?

I believe this replacement will happen, and it is already underway.

What is the truth behind this round of leading gains?

People are used to labeling tech rallies with a single tag, such as computing power, applications, hardware, software, or thematic concepts, and pick one as the answer.

This round of tech rally in the A-share market has never been driven by a single factor.

It is a dual-engine machine, with one engine fueled by the income statement, and the other fueled by the valuation sheet.

Hardware is the part fueled by the income statement.

The development of optical modules from 400G to 800G, then to 1.6T and 3.2T follows a clear rhythm; high-end PCBs are upgraded alongside AI servers, with orders scheduled far into the future; for memory chips, Samsung, SK Hynix and Micron have shifted their production capacity to HBM, leaving gaps in traditional DRAM and NAND, which are filled by the two leading domestic memory makers, leading to explosive profit growth; servers, liquid cooling and semiconductor equipment all record solid shipment volumes and prices.

Clues from the three quarterly reports of securities firms suggest that the electronics, PCB and computing power hardware sectors are the most concentrated areas for positive pre-performance announcements.

The stock prices of these sectors are underpinned by actual profits.

Thematic concepts are the part fueled by the valuation sheet.

Humanoid robot complete machines, commercial space operations, mass production applications of solid-state batteries, short dramas and agents in AI applications are traded for their 0-to-1 option value. They feature high flexibility and huge room for imagination, but no profits are yet reflected in the income statement, and their performance relies on policy catalysts, technical milestones, founder narratives and trading sentiment.

This type of assets is extremely sensitive to liquidity: they surge when market sentiment is favorable, and slump as soon as the momentum fades.

As for software applications, there are plenty of stories, but it is still too early for them to take over the leading role. The State Council's statements of "driving technological innovation through practical application implementation", "coordination between computing power and power supply" and "high-quality data supply" all indicate that the application sector is still on the path of evolving from policy-driven development to a closed commercial loop.

The three first-order indicators, including call volume, payment rate and profit margin, are not yet strong enough to support software to fully take the lead.

Therefore, the structure of this round of rally is completely defined by hardware underpinning profits, thematic concepts underpinning valuations, and software queuing up behind.

This structure determines that the first step of the next round of shift will not be a big leap from "hardware to software", but a handover within the hardware sector first, shifting from the overall beta of total computing power to the alpha of computing power bottlenecks, or from complete machines and computing power leasing to power supporting facilities, liquid cooling, connectors, ceramic substrates, memory, PCBs and domestic semiconductor equipment.

All these links share one common point: the further AI develops, the more bottleneck constraints they form; the stronger the bottleneck constraints, the more solid their pricing power is.

The full takeover by software is the final step, not the first. Only by clarifying this point first can investors avoid being thrown off the market when the handover within the hardware sector takes place.

Penetration rate determines pricing logic

Treating a "track" as a pricing unit is the biggest misconception. Within the same track, different links are at different penetration rate stages, with vastly different pricing logics.

The shift of leading directions is often not a large cross-track move such as "switching from chips to robots", but a shift from complete machines to bottlenecks, or from thematic concepts to actual orders within the same industrial chain.

At the 0-to-1 penetration rate stage, pricing relies on option value, focusing on policies, technical milestones and founder narratives, and the income statement can be temporarily blank. Humanoid robot complete machines, commercial space operations, mass production applications of solid-state batteries and pure thematic AI applications are all at this stage.

Their stock prices act like long-dated options: when real interest rates rise, their valuations are compressed the most sharply; when liquidity tightens, they slump the most severely.

At the 1-to-10 penetration rate stage, pricing relies on growth, focusing on orders, production capacity, the slope of penetration rate, PS ratio and PEG ratio. AI computing power hardware, optical modules, PCBs, memory, liquid cooling, semiconductor equipment, robot components, upstream commercial space sectors, solid electrolytes, and overseas expansion of innovative drugs are all at this stage.

This stage is the most challenging, as valuation expansion and profit realization are in a race, and differentiation starts right here. Companies with orders move up, while those without orders drop, and completely opposite performance trends can emerge within the same track.

At the 10-to-100 penetration rate stage, pricing relies on cycles and value, focusing on ROE, cash flow, market share and price wars.

Mature process manufacturing, ordinary consumer electronics, generic drugs and some software services are all at this stage.

The valuation center moves down, and the market focuses on shareholder returns rather than growth stories.

What is the current situation? Most AI hardware sectors are in the mid-to-late phase of the 1-to-10 stage: optical modules, PCBs, memory and servers have already realized profits, but their valuations are not cheap; liquid cooling, connectors, ceramic substrates, domestic memory and semiconductor equipment are in the middle phase of the 1-to-10 stage, and order verification determines their differentiation; humanoid robots, commercial space and solid-state batteries are at the late phase of the 0-to-1 stage to the early phase of the 1-to-1 stage, with strong thematic attributes; the overseas expansion part of innovative drugs and CXO sectors are in the 1-to-10 stage, while generic drugs are in the 10-to-100 stage.

The tech rally is far from over. The next round of leading directions will not simply switch tracks, and I believe it will most likely shift to different links. The market will shift from the total volume story of the computing power arms race to scarcity pricing of computing power bottlenecks; from the vision of humanoid robot complete machines to order realization of the Tesla supply chain and core components; from the concept of solid-state batteries to the layout of electrolytes, equipment and dry process technologies.

Links that enter the income statement first will get pricing power first. The track remains the same, but the leading driver changes.

In addition, the penetration rate curve is not uniform.

At the 1-to-10 stage, the biggest risk is that the slope of the penetration rate suddenly flattens. For example, if computing power leasing prices stop rising, or the profit margins of CSPs are squeezed, the growth pricing of this link will instantly turn into cyclical pricing, and the valuation will drop even more sharply than that of 0-to-1 thematic assets.

On the contrary, if a 0-to-1 thematic asset suddenly gets a large order, a major policy support or a key large client, its penetration rate slope will steepen, and it can also seize the leading position in a short period of time.

Therefore, when looking at the penetration rate, we should not only focus on the stage, but also on the changing direction of its slope.

When the slope rises, pricing power rises; when the slope flattens, pricing power falls.

Which assets are the most vulnerable when liquidity tightens, and which benefit the most from better-than-expected earnings?

However, within the same track, different links have different fates. So what kind of links get hurt first, and what kind of links benefit first?

This depends on whether the macro conditions are favorable. Industrial trends determine the general direction, liquidity determines the rhythm, and earnings realization determines who gets pricing power first.

Once liquidity tightens, which type of tech assets are the most vulnerable? And once earnings exceed expectations, which type of links benefit the most?

By clarifying these two scenarios, the sequence of the handover of leading drivers will become clear, as it determines the rhythm and intensity of the shift.

If liquidity tightens, the most vulnerable targets are not the whole "tech sector", but three types of assets.

The first type: 0-to-1 long-term cash flow themes, including humanoid robot complete machines, commercial space operations, mass production applications of solid-state batteries, and pure thematic AI applications. They act like long-dated options, and when real interest rates rise, their valuations are compressed most directly.

The second type: high-position growth stocks with high financing proportion, high trading concentration and heavy institutional holdings. CITIC Securities points out that this round of correction is triggered by heavily held institutional stocks and transmitted to sectors with high financing proportion; Zhongtai Securities directly warns of the critical point of negative liquidity feedback; data from Kaiyuan Securities shows that the concentration of trading volume of the top 5% stocks has dropped to 41.01%, and the vulnerability is improving but not eliminated.

The third type: capital-heavy computing power leasing and AI infrastructure with high capital expenditure. Once computing power leasing prices and CSP profit margins fall short of expectations, the pressure on the balance sheet will quickly become prominent.

Hardware that has already realized profits will also see valuation declines, but underpinned by the income statement, the magnitude of decline and the recovery speed are different from those of pure thematic assets.

This is the real difference between "profitable" and "profitless" assets when liquidity tightens.

On the contrary, if earnings exceed expectations, the links that benefit the most are those with supply bottlenecks, rising prices and expanding volume, and domestic substitution logic. For the memory sector, the key indicator is memory prices; for high-end PCBs and optical modules, it is the shipment volume of AI servers; for liquid cooling, connectors and ceramic substrates, it is the penetration of new architectures; for semiconductor equipment and domestic memory, it is the substitution rhythm driven by export control policies; for power supporting facilities and grid energy storage, it is the implementation of "computing-power and power-supply coordination"; for overseas expansion of innovative drugs and CXO sectors, it is policy price protection and clinical data catalysts.

All these links share the common feature that the evidence of upward first-order indicators appears on them first.

The market will give bonuses to the first group of companies that get verified, and send bills to the last group that get verified.

In addition, when liquidity tightens, the market usually first sells off "overvalued" assets, then "unprofitable pseudo-growth" assets, and finally "genuine profitable assets".

Overvalued assets refer to those with excessively high valuations, pseudo-growth assets refer to those with no profits, and genuine assets refer to those with profits but are wrongly sold off.

The rhythm of the shift usually follows this sequence.

Therefore, when seeing the overall decline of the tech sector, we cannot generalize it. We need to distinguish which layer of assets is being sold off: when overvalued assets are sold off, profitable assets also fall along, which is a mispricing; when pseudo-growth assets are sold off, pure thematic assets drop the most sharply, and profitable assets start to differentiate; when genuine assets are sold off, it is in fact a window for profitable links to be revalued.

It is precisely because of this that the market easily misinterprets the "clearance of overvalued and pseudo-growth assets" as the "end of the overall tech main line". Therefore, we must separate the definition of "end" next: what is coming to an end, is it the universal rally of all low-quality thematic assets, or the long-term trend of the tech industry? These two are completely different things.

The universal thematic rally is over, but the tech main line is not

The signals that the universal thematic rally is over have been fully displayed.

The Wind All A Index has corrected by more than 20% from its high in mid-May, and its valuation has returned to around 20 times; the CSI 300 Index has fallen for 10 consecutive weeks, exceeding the 9 consecutive weeks of decline at the end of 2011; the trading volume concentration of the top 5% stocks has dropped from a high level to 41.01%; institutional investors are cutting positions, and margin trading balances are flowing out; securities firms' strategy recommendations have shifted from "offense" to "balance", "defense" and "waiting for the right trading window".

All these signals indicate that the stage of indiscriminate valuation expansion, the stage of indiscriminate institutional group holdings, and the stage where all stocks rise regardless of quality have all come to an end.

But what is needed for the end of the tech main line? The first-order indicators need to be falsified.

Memory prices turn down, computing power card leasing prices collapse, CSP profit margins deteriorate, AI capital expenditure is revised down, domestic substitution orders are falsified, and policies shift. None of these things have happened yet.

The State Council is still rolling out arrangements for "Artificial Intelligence +", requiring "deepening and expansion", "coordination between computing power and power supply" and "scenario cultivation"; two government departments plan to revise the export catalog to add AI, biomedicine and photovoltaic technologies to the list, which adds scarcity premium for strategic technologies; leading tech companies are conducting intensive share repurchases, including Shengyi Technology, Hygon Information, Sugon, Montage Technology and Gcore, and industrial capital is re-pricing the cash flow discount rate with real money; the positive pre-performance announcements for the three quarterly reports are highly concentrated in the electronics, PCB and computing power hardware sectors.

These signals indicate that the industrial trend of the tech main line is still there, but the pricing logic has changed.

Therefore, replacing the binary debate of "whether the rally is over or not" with the concept of "direction shift" will make the logic much clearer.

What has ended is the universal thematic rally, not the tech main line; what has ended is valuation expansion, not the industrial trend; what has ended is the pricing logic based on institutional group holdings, not the pricing of tech assets.

The next round of leading directions will change, but will not go beyond the tech sector, and will shift to links with upward first-order indicators, 1-to-10 penetration rate, verifiable cash flow, supply bottlenecks and strategic scarcity.

Furthermore, the end of the universal thematic rally is actually a good thing for the tech main line.

During the universal rally, good companies and bad companies rise together, and pricing power is in the hands of market sentiment; during the differentiation stage, profitable companies and unprofitable companies go separate ways, and pricing power returns to the income statement.

The first act of the tech bull market relies on liquidity and stories; the second act relies on profits and cash flow.

The first act has ended, and the second act has just begun.

As for the leading directions of the second act, I think they can be broken down into four specific shifts.

The first shift: from total computing power to computing power bottlenecks.

Computing power does not only refer to GPU. Power supply, liquid cooling, connectors, ceramic substrates, memory, PCBs and domestic equipment are the real bottlenecks in the computing power inflation chain. The State Council mentions "coordination between computing power and power supply", Caitong Securities points out the liquid cooling, connectors and ceramic substrates in the Rubin chain, Zhongtai Securities points out domestic memory and semiconductor equipment, and China Galaxy Securities points out communication equipment, PCBs, memory and servers.

These clues indicate that the next round of hardware rally will shift from the "computing power concept" to the rent-collecting links of "computing power public utility". Whoever is stuck in the bottleneck will get the pricing power.

The second shift: from universal thematic rally to scenario stratification.

The State Council emphasizes traction through practical application implementation, connection between supply and demand, pilot test verification and scenario cultivation; the export catalog adds AI, biomedicine and photovoltaic technologies. This means that AI, biomedicine and photovoltaics have changed from "thematic concepts" to "strategic scarce assets", and obtained technical sovereignty premium.

But for the application sector to take over the leading role, the first-order indicators of call volume, payment rate and profit margin must be seen, and the popularity of short dramas and agents alone is not enough.

The essence of scenario stratification is that the market starts to price separately for "scenarios that can be implemented" and "stories that can only be told".

The third shift: from single computing power focus to rotation across six tracks.

Chips and semiconductors, computing power and algorithms, humanoid