By | Ban Dai Jun
The 2026 interim report season reveals an extreme contrast in China's chip design industry:
Longsys saw a net profit increase of 74394%, GigaDevice grew by 1099%, and Cambricon achieved multi-billion-level profitability for two consecutive quarters. In the same field, StarPower's profits plummeted by 70%, CETC Chip turned from profit to loss, and HGTECH's net profit plunged by 800%.
If you analyze these companies as manufacturing, you would conclude 'high inventory risk, price hikes unsustainable.' If analyzed as financial institutions, you'd get 'balance sheet arbitrage.'
Both conclusions are wrong.
Because they overlook a more fundamental fact: The profits of the chip design industry in 2026H1 were not 'created'; they were 'institutionally transferred.'
But this article's purpose is not to expose darkness. It is to prove: It is precisely this much-criticized 'institutional transfer' mechanism that has accomplished something China's chip industry failed to do in the past twenty years—it has given chip design companies, for the first time, positive operating cash flow and an irreversible 'window for capability leap.'
I. The Distribution of Institutional Rents: Verified Hard Data
First, look at data from authoritative institutions.
TrendForce data shows that in Q1 2026, global contract prices for mainstream DRAM rose 90%–95% quarter-on-quarter, and NAND Flash prices rose 55%–60%, marking historic highs. The upward trend continued in Q2, with mainstream DRAM contract prices up 58%–63% and NAND Flash up 70%–75% quarter-on-quarter.
However, monitoring data from the Digital Economy Laboratory of the University of International Business and Economics shows that from January to March 2026, China's memory chip exports reached $19.14 billion, a year-on-year increase of 205.5%, while export volume grew only about 4%.
Volume up 4%, price up 205%—this is not demand expansion; it's a violent price redistribution under supply constraints.
Data from the China Passenger Car Association shows that from January to May 2026, the profit margin of China's auto industry was 3.4%, lower than the downstream industrial average of 6.1%, and has been declining over the past three years (4.3% in 2024, 4.1% in 2025, 3.2% in Q1 2026). He Xiaopeng, CEO of XPeng Motors, stated bluntly that most of the money saved by automakers through technological innovation was given back to partners in businesses like memory chips and lithium carbonate.
GigaDevice's 6.9 billion yuan profit and Longsys's 10.2 billion yuan profit are essentially an institutional transfer of profits from downstream sectors. They didn't 'create' value; the system (capacity squeeze + policy compliance) forced downstream sectors to 'cede' profits to upstream ones.
Cambricon is similar. On May 22, 2026, Li Chao, Deputy Director of the Policy Research Office of the National Development and Reform Commission, clearly stated that they would 'guide domestic large models to intensify efforts to adapt to domestic computing power chips.' In April 2026, State Council Decree No. 834, the 'Regulations on Promoting the Security and Stability of Industrial and Supply Chains,' took effect. Internet giants buy Cambricon chips not because they perform better than NVIDIA's, but because 'domestic substitution' is a political task.
Customers are not buying chips; they are buying 'compliance credentials.'
This creates a unique pricing mechanism: Chip price = Physical cost + Institutional rent. Institutional rent has no upper limit—it depends on the intensity of policy pressure, not on marginal improvements in technical performance.
The entire industry is split into two classes:
Class One: The Institutional Rent Recipients (Cambricon, GigaDevice, Longsys, Biwin Storage, Hygon)
Class Two: The Institutional Rent Payers (StarPower, CETC Chip, HGTECH, TOWED, Giantec)
This is not cyclical divergence; it is class differentiation. Unidirectional flow, zero-sum game, time mismatch.
II. The Fact Everyone Overlooks: Cash Flow Restructuring
Up to this point, the analytical framework aligns with the pessimistic view. But next, the same framework leads to a completely different conclusion.
What has been the biggest fatal flaw of China's chip design industry over the past twenty years?
It's not technological backwardness; it's the perpetual breakage of cash flow.
Always worrying about the next round of financing, never waiting for products to become profitable, forever forced into short-cycle R&D like 'product in 18 months, IPO in 3 years.' Without positive operating cash flow, there is no capital for long-cycle technological gambles; without long-cycle technological gambles, there's no catching up to NVIDIA, Qualcomm, Texas Instruments.
Although institutional rents distort short-term competition, they objectively accomplish an 'institutional cash flow restructuring'—systematically transferring the profit pool from downstream manufacturing to the upstream design segment through two valves: capacity scarcity and policy compliance.
This is not market-driven value creation, but it achieves something venture capital and subsidies failed to do in the past twenty years: giving chip design companies positive operating cash flow for the first time.
Look at a set of data drowned in the profit noise:
Cambricon, GigaDevice, Hygon, and Montage Technology all disclosed significant increases in R&D investment in their 2026 interim reports, with R&D directions shifting from 'mature process substitution' to 'automotive-grade/industrial-grade/AI computing power iteration.' The Siyuan 690 chip is already in mass production, with FP16 computing power exceeding 700 TFLOPS; GigaDevice's automotive-grade MCUs and industrial-grade NOR Flash are moving from concept to mass production.
This money doesn't come from venture capital or government subsidies; it comes from the cash flow restructuring enabled by institutional rents.
The key distinction: Whether restructuring is poison or antidote doesn't depend on the restructuring itself, but on how companies spend this money.
Spending it on hoarding capacity, buying equity, expanding government relations teams—that's the Dutch Disease, a slow suicide.
Investing it in 10-year-cycle technology—that's strategic depth, a capability leap after cash flow restructuring.
III. The Capability Leap Window: Why This Time Is Different
Some might say: 'There has been policy support and domestic substitution before. Why is this time different?'
Because previous transfusions were 'intermittent,' while this cash flow restructuring is 'structural and long-cycle.'
SK Hynix CEO Kwak Noh-Jung clearly stated: 'The memory chip shortage phenomenon may last beyond 2030.' TrendForce has revised its 2026 global memory market size forecast upward from $551.6 billion to $889.3 billion, a 61% increase.
This means: The window period for institutional rents is not one or two years, but five years or more. Long enough for a company to complete a full cycle from 'proof of concept' to 'product iteration' to 'ecosystem lock-in.'
More importantly, policies are shifting from 'procurement subsidies' to 'ecosystem binding.'
On May 22, 2026, the NDRC didn't simply say 'support domestic chips,' but explicitly stated it would 'guide domestic large models to intensify efforts to adapt to domestic computing power chips.' This is not a procurement subsidy; it's mandatory ecosystem binding—turning domestic chips from 'being procured' to 'being used,' from 'compliance credentials' to 'part of the technological ecosystem.'
In June 2026, the US further tightened AI chip export controls to China, even restricting sales of the H20. Nine domestic AI chips received National Security Level I certification. The greater the external pressure, the longer the window period for institutional rents, and the more time available for the capability leap.
This is not 'policy protecting backward capacity'; it's 'policy using non-market means to forcibly solve a problem the market couldn't solve'—the chicken-and-egg deadlock in the chip industry chain.
No customers, no revenue; no revenue, no R&D; no R&D, no products; no products, no customers.
Institutional rents break this deadlock: First, use forced procurement to create customers; then, use customer revenue to feed R&D; finally, use R&D results to lock in genuine market-driven customers.
IV. The 'Market-Driven Clearing' of the Payer Side: The Necessary Pain
StarPower's profits plummeted 70%, TOWED's revenue and profit both declined, HGTECH plunged 800%.
The market calls this a 'predicament.' But from another angle, it's 'market-driven clearing.'
The problem with China's power semiconductor, analog chip, and military communication chip industries over the past decade has been: Too much policy protection, too little market competition. Although StarPower's technology is among the top three globally, it has lived in the greenhouse of 'domestic substitution'—downstream automakers forced to procure, ineffective price transmission masked by subsidies, low R&D efficiency hidden by order growth.
The 2026H1 institutional rent distribution shattered that greenhouse. Payer-side companies without the protection of policy rents were thrown into real, fierce market competition for the first time—costs must be cut, prices must be transmitted, efficiency must be improved.
More crucially is the downstream feedback effect. After automaker profits were compressed to the extreme of 3.4%, they began extending olive branches to upstream chip companies—not just procurement, but equity binding, joint R&D, co-construction of capacity. Similar to the Toyota-Denso, Volkswagen-Infineon models.
StarPower's predicament objectively constitutes bargaining chips for upgrading from 'supplier' to 'ecosystem partner.' Of course, the cost of this 'market-driven clearing' is real: capacity contraction, order volatility, supply chain pressure. Industrial restructuring is never painless, but the pain itself is a signal—it proves the old model has reached its end, and a new model is being forced into existence.
V. The True Light: Independent Growth 'Beyond Institutional Rents'
In all the preceding analysis, one type of company was overlooked: Their growth does not entirely rely on institutional rents but is based on genuine technological barriers and market-driven demand.
Montage Technology is typical. In the first half, net profit attributable to shareholders grew 63.9%–81.2%, and non-GAAP net profit grew 14.5%–32.9%. Although the non-GAAP growth rate is lower than the GAAP figure, it is positive, real, stemming from market demand for DDR5 interface chips—AI server memory upgrades are part of a real technological iteration cycle, not policy-mandated procurement.
Fudan Microelectronics is similar. Non-GAAP net profit grew 92%–147%. Demand for automotive-grade MCUs and FPGAs is independent of the consumer electronics cycle; domestic substitution in automotive electronics is a market-driven choice (automakers actively diversifying for supply chain security), not policy coercion.
The existence of these companies demonstrates the boundaries of the 'institutional rents' framework: Rents are indeed corroding some companies, but they are also shielding another group of companies, allowing them to complete the independent growth of market-driven capabilities. When rents recede, the former will be exposed, while the latter will become protagonists of the new cycle.
This is the true light—not the reflection of rents, but the faint glimmer of independent technological growth.
VI. The Second Half of 2026: Three Strangulation Mechanisms and Three Leap Signals
Based on the above analysis, the chip design industry faces three strangulation mechanisms in the second half of 2026, alongside the emergence of three leap signals:
Strangulation and leap are two sides of the same coin. Strangulation weeds out companies that 'only know how to collect rents,' while leap rewards those that 'transform rents into capabilities.'
VII. Conclusion: Planting Flowers in the Abyss
The extreme divergence in China's chip design industry in 2026H1 is indeed the result of 'institutional rent distribution.' Profits were not 'created'; they were 'institutionally transferred'—from downstream payers to upstream recipients. This transfer is unidirectional, zero-sum, and time-mismatched.
But the core judgment of this article is: Institutional rents are not the end point; they represent a 'capability leap window' following 'cash flow restructuring.'
The biggest dilemma for China's chip industry over the past twenty years has been the 'short-cycle R&D trap caused by cash flow breakage.' Although institutional rents distort short-term competition, they objectively break the chicken-and-egg deadlock—first using forced procurement to create customers, then using customer revenue to feed R&D, finally using R&D results to lock in genuine market-driven customers.
SK Hynix says memory shortages may last until 2030, US export controls are tightening, domestic substitution policies are upgrading— this window period is long enough for companies with real capability to complete the leap, transforming from 'rent collectors' into 'technology definers.'
The strangulation mechanisms of the second half (FIFO reversal, demand backlash, capacity bottlenecks) will still activate, and the cast of the divergence will change. But the new selection criteria will no longer be 'who collects the most rent,' but rather 'who has transformed rent into irreplaceable technological barriers.'
The essence of investing in chip design is not betting on the persistence of institutional rents, but identifying companies that are transforming 'cash flow restructuring' into 'independent viability.'
The moment the feast is most sumptuous is often also when the signal to leave is clearest. But this time, it's not just the old paradigm leaving—on the ruins of the old paradigm, new species with genuine market-driven viability are quietly sprouting under the cover of rents.
There is light at the bottom of the abyss. That light is not the reflection of rents; it's the faint glimmer of independent technological growth.
