The question is less whether the US ‘should’ and more whether the memory cycle makes selective dependence economically irresistible before it becomes politically unacceptable.
The YMTC parallel is worth noting. YMTC, sanctioned in 2022, just disclosed >10% global 3D NAND market share with Q1 revenue over 20B RMB, with reports that it's overwhelmed with orders. US sanctions accelerated domestic qualification (Xiaomi, Lenovo, Alibaba Cloud all switched early), which built the volume needed to reach cost-competitive scale. CXMT may follow the same arc: restrictions force domestic qualification, which builds scale, which eventually enables commodity pricing that undercuts the Big Three on lower tiers. The geopolitical lever ends up creating exactly the industrial policy outcome the US was trying to prevent.
I agree with your core judgment: at a moment when AI is absorbing more and more memory supply and commodity DRAM is moving into a tight-balance cycle, CXMT has already become a sufficiently meaningful marginal player in the global commodity DRAM market. From that perspective, rather than moving immediately to a total ban, the United States would be better off treating CXMT, within controlled limits, as a tool to ease cost pressure and strengthen buyer bargaining power.
Over the past decade, U.S. pressure on CXMT has indeed materially raised the company’s costs of R&D, capacity expansion, process ramp-up, and supply-chain organization, and it has also tried to confine CXMT to less advanced DRAM nodes and lower-end applications. But that pressure did not prevent CXMT from surviving, scaling output, and entering validation cycles with a broader range of end customers. Nor did it prevent China from building a far more complete support structure around domestic memory: local demand, local government capital, central state-backed funds, equipment and materials substitution, and downstream customer adoption.
That is because the United States is no longer dealing with a company that simply survives by relying on global capital markets and cross-border supply chains. It is dealing with a “system-level company” embedded inside China’s national semiconductor strategy, local fiscal support, policy-bank financing, downstream OEM relationships, and the broader push for domestic substitution. For a company like this, the objective function is not short-term profit maximization. Its real survival logic is strategic endurance: stay alive, keep the line running, get customers in the door, and gradually improve yield. For that reason, future U.S. attempts to keep squeezing CXMT will become much harder.
And the reason is not just political will. It is that the marginal effectiveness of additional pressure is declining. Why is it declining? Because sanctions are most effective at the earliest stage, before a Chinese company has formed a stable production line, before it has locked in domestic customers, and before it has built a substitution network in equipment and materials. Once a company has already survived its most fragile phase, and once the domestic market behind it is large enough, further escalation tends to produce three consequences. First, it continues to raise costs, but struggles to reverse the overall direction of development. Second, it further stimulates substitution in domestic equipment, materials, EDA, and components. Third, it pushes more global OEMs and branded device makers toward the practical choice of using Chinese memory in non-U.S. markets.
And this is no longer just about Apple. HP, Dell, Acer, and Asus have also been evaluating Chinese memory amid global supply tightness, at least for non-U.S. markets. Once CXMT enters the pool of credible alternatives, its strategic value rises by an entire tier.
That said, “it will be much harder for the U.S. to keep suppressing CXMT” does not mean “the U.S. has basically run out of tools.” Washington still holds several powerful levers. First, advanced equipment and services—the ability to continuously access and maintain more advanced etch, deposition, inspection, metrology, advanced packaging, and related software capabilities. Second, the HBM chain. Once BIS brought HBM into the focus of export controls in 2024, it became clear that Washington is not only worried about commodity DRAM; it is even more concerned that China could use the memory route to enter one of AI’s most critical bottlenecks. Third, allied coordination and service restrictions. The revised MATCH Act is part of an effort to align restrictions on Chinese chipmakers more closely with allies such as the Netherlands and Japan, while continuing to target equipment sales and servicing for facilities linked to CXMT, YMTC, and SMIC. The U.S. toolbox is not exhausted. It is simply becoming more expensive—diplomatically, industrially, and across global supply chains—to keep expanding the effect of those tools.
But in my view, the point Aqib most underestimates about CXMT is also the point he most underestimates about China’s state-backed innovation system. The reason CXMT is harder to knock down today is not just that U.S. suppression failed to eliminate it. It is that China has already formed a closed-loop demand system for domestic memory substitution. Smartphones, PCs, servers, government and enterprise IT replacement, automotive electronics, industrial control, and cloud providers—all of these segments together provide a sufficiently large domestic space for trial-and-error, validation, and order buffering. On top of that, local governments, the Big Fund, and state-owned capital platforms can provide patient capital during loss-making phases, so the company does not need to behave like a Western listed firm that must prioritize ROIC in every single cycle.
In other words, CXMT did not grow through temporary trade arbitrage. It grew through state capacity, domestic market scale, and capital patience. As long as those three pillars remain in place, the probability that the United States can force it off the stage simply by launching one more round of restrictions is already quite low.
To be clear, CXMT’s global DRAM share is still only around 4%, and it remains far from the dominance of Samsung, SK hynix, and Micron. But that is not the most important point anymore. What matters just as much is that it is no longer at zero, no longer a laboratory project, and no longer something the market can dismiss as hypothetical. In a period of global memory tightness, it has become a new source of supply that international OEMs now have to evaluate seriously.
That is why the real dilemma facing the United States today is no longer simply whether to continue pressing CXMT. The deeper question is this: can Washington continue applying pressure at a reasonable cost while preserving the technology gap—without, in the process, accelerating China’s emergence of a more complete and self-sufficient semiconductor ecosystem?
If you still thinks the US operates on logic and reasoning, then you don’t understand where the US is at at all. Every event in the past few years have told you it’s about defaulting on the insurmountable debt through war provocation. The US will continue on this war path at the expense of civil war, war with allies, and chaos. It should be clear that no one is interested in bringing debt down from 130% of GDP. Why avoid or deny the inevitable when Congress and trump are acting like they don’t care about Armageddon? It’s because they.don’t.care. The smart countries need to dump treasuries and buy gold to sure up their currencies. It’s macro time, not micro.
The question is less whether the US ‘should’ and more whether the memory cycle makes selective dependence economically irresistible before it becomes politically unacceptable.
The YMTC parallel is worth noting. YMTC, sanctioned in 2022, just disclosed >10% global 3D NAND market share with Q1 revenue over 20B RMB, with reports that it's overwhelmed with orders. US sanctions accelerated domestic qualification (Xiaomi, Lenovo, Alibaba Cloud all switched early), which built the volume needed to reach cost-competitive scale. CXMT may follow the same arc: restrictions force domestic qualification, which builds scale, which eventually enables commodity pricing that undercuts the Big Three on lower tiers. The geopolitical lever ends up creating exactly the industrial policy outcome the US was trying to prevent.
I agree with your core judgment: at a moment when AI is absorbing more and more memory supply and commodity DRAM is moving into a tight-balance cycle, CXMT has already become a sufficiently meaningful marginal player in the global commodity DRAM market. From that perspective, rather than moving immediately to a total ban, the United States would be better off treating CXMT, within controlled limits, as a tool to ease cost pressure and strengthen buyer bargaining power.
Over the past decade, U.S. pressure on CXMT has indeed materially raised the company’s costs of R&D, capacity expansion, process ramp-up, and supply-chain organization, and it has also tried to confine CXMT to less advanced DRAM nodes and lower-end applications. But that pressure did not prevent CXMT from surviving, scaling output, and entering validation cycles with a broader range of end customers. Nor did it prevent China from building a far more complete support structure around domestic memory: local demand, local government capital, central state-backed funds, equipment and materials substitution, and downstream customer adoption.
That is because the United States is no longer dealing with a company that simply survives by relying on global capital markets and cross-border supply chains. It is dealing with a “system-level company” embedded inside China’s national semiconductor strategy, local fiscal support, policy-bank financing, downstream OEM relationships, and the broader push for domestic substitution. For a company like this, the objective function is not short-term profit maximization. Its real survival logic is strategic endurance: stay alive, keep the line running, get customers in the door, and gradually improve yield. For that reason, future U.S. attempts to keep squeezing CXMT will become much harder.
And the reason is not just political will. It is that the marginal effectiveness of additional pressure is declining. Why is it declining? Because sanctions are most effective at the earliest stage, before a Chinese company has formed a stable production line, before it has locked in domestic customers, and before it has built a substitution network in equipment and materials. Once a company has already survived its most fragile phase, and once the domestic market behind it is large enough, further escalation tends to produce three consequences. First, it continues to raise costs, but struggles to reverse the overall direction of development. Second, it further stimulates substitution in domestic equipment, materials, EDA, and components. Third, it pushes more global OEMs and branded device makers toward the practical choice of using Chinese memory in non-U.S. markets.
And this is no longer just about Apple. HP, Dell, Acer, and Asus have also been evaluating Chinese memory amid global supply tightness, at least for non-U.S. markets. Once CXMT enters the pool of credible alternatives, its strategic value rises by an entire tier.
That said, “it will be much harder for the U.S. to keep suppressing CXMT” does not mean “the U.S. has basically run out of tools.” Washington still holds several powerful levers. First, advanced equipment and services—the ability to continuously access and maintain more advanced etch, deposition, inspection, metrology, advanced packaging, and related software capabilities. Second, the HBM chain. Once BIS brought HBM into the focus of export controls in 2024, it became clear that Washington is not only worried about commodity DRAM; it is even more concerned that China could use the memory route to enter one of AI’s most critical bottlenecks. Third, allied coordination and service restrictions. The revised MATCH Act is part of an effort to align restrictions on Chinese chipmakers more closely with allies such as the Netherlands and Japan, while continuing to target equipment sales and servicing for facilities linked to CXMT, YMTC, and SMIC. The U.S. toolbox is not exhausted. It is simply becoming more expensive—diplomatically, industrially, and across global supply chains—to keep expanding the effect of those tools.
But in my view, the point Aqib most underestimates about CXMT is also the point he most underestimates about China’s state-backed innovation system. The reason CXMT is harder to knock down today is not just that U.S. suppression failed to eliminate it. It is that China has already formed a closed-loop demand system for domestic memory substitution. Smartphones, PCs, servers, government and enterprise IT replacement, automotive electronics, industrial control, and cloud providers—all of these segments together provide a sufficiently large domestic space for trial-and-error, validation, and order buffering. On top of that, local governments, the Big Fund, and state-owned capital platforms can provide patient capital during loss-making phases, so the company does not need to behave like a Western listed firm that must prioritize ROIC in every single cycle.
In other words, CXMT did not grow through temporary trade arbitrage. It grew through state capacity, domestic market scale, and capital patience. As long as those three pillars remain in place, the probability that the United States can force it off the stage simply by launching one more round of restrictions is already quite low.
To be clear, CXMT’s global DRAM share is still only around 4%, and it remains far from the dominance of Samsung, SK hynix, and Micron. But that is not the most important point anymore. What matters just as much is that it is no longer at zero, no longer a laboratory project, and no longer something the market can dismiss as hypothetical. In a period of global memory tightness, it has become a new source of supply that international OEMs now have to evaluate seriously.
That is why the real dilemma facing the United States today is no longer simply whether to continue pressing CXMT. The deeper question is this: can Washington continue applying pressure at a reasonable cost while preserving the technology gap—without, in the process, accelerating China’s emergence of a more complete and self-sufficient semiconductor ecosystem?
If you still thinks the US operates on logic and reasoning, then you don’t understand where the US is at at all. Every event in the past few years have told you it’s about defaulting on the insurmountable debt through war provocation. The US will continue on this war path at the expense of civil war, war with allies, and chaos. It should be clear that no one is interested in bringing debt down from 130% of GDP. Why avoid or deny the inevitable when Congress and trump are acting like they don’t care about Armageddon? It’s because they.don’t.care. The smart countries need to dump treasuries and buy gold to sure up their currencies. It’s macro time, not micro.