I originally published this article on War on the Rocks on August 13, 2026. I am also publishing it here on my website:

Given the relentless demand for computing power, electronic components are in scarce supply. Prices for certain memory chips, known as DRAM, have surged by more than 50 percent in a single quarter this year, and have roughly quadrupled since last fall. Because DRAM supply is tight, Apple, Dell, and HP are currently evaluating memory from ChangXin Memory Technologies (CXMT), a company the Pentagon has designated as a Chinese military company. Apple, in particular, has sought assurances from the U.S. government that CXMT will not face future sanctions that would cut off its supply. Apple justifies its request by pledging to use CXMT chips for its products sold in China.

Even if every chip stays in China, the arrangement endangers national security. Such use and validation would build a state-backed champion into a global competitor in a market where capital is the attack vector. The Chinese government has successfully employed this strategy before. Recognizing this playbook, lawmakers published a bipartisan July 14 congressional letter demanding the Commerce Department add CXMT to its Entity List, an export-control designation that imposes licensing requirements on specified transactions involving listed entities. Given this history, the United States should not only refuse any assurances by these companies, but also place CXMT on the Entity List.

I am a data center regulatory attorney at Microsoft. The views expressed here are solely my own and do not reflect my employer’s position. Microsoft benefits from a larger and cheaper supply of memory, which my call to expand allied production would advance. However, in the near term, the position I take here would keep a low-cost supplier out of the Western supply chain. I argue for this position on national security grounds, rather than commercial ones.

Apple Has Run This Experiment Before

The case for procuring CXMT components is strongest if national security concerns are limited to the memory and whether it contains exploitable vulnerabilities or could expose user data. Apple, Dell, and HP have a conventional technical need, and CXMT offers another potential supplier in a constrained market. Apple also proposes using the chips in devices sold in China. On that framing, the arrangement appears to reduce the direct risk to Western user data, meaning federal intervention looks more like economic protectionism than security policy. However, such framing is far too narrow. Hardware components need not be compromised to create geopolitical leverage. The relevant risk is whether this procurement and validation help a state-backed supplier become a global competitor in a capital-intensive commodity market.

Western purchase of CXMT memory is a step backwards from the industry’s recognition of exposure to the Chinese government. In particular, Apple recognizes the risks created by its heavy dependence on China and has since expanded manufacturing in India and Vietnam partly to reduce that concentration. Yet sourcing DRAM from CXMT would introduce new exposure to China in a strategically sensitive component. The pursuit of CXMT suggests that short-term supply pressures are overriding the longer-term de-risking strategy, while advancing Beijing’s objective of making itself indispensable to critical technology supply chains through multiple structures.

The precedent for such structural dependencies is well-documented within Apple’s own operational history. In the book Apple in China, Patrick McGee describes how Apple was investing $55 billion a year in its Chinese supply chain by 2015, and by its own count, has trained 28 million workers there. The resulting asymmetry allows Beijing to regularly exert regulatory leverage over Apple’s operations. In 2018, Apple migrated its mainland iCloud data and encryption keys to a Chinese state-owned operator. In 2022, Apple limited open AirDrop sharing in China to a ten-minute window, restricting a feature that protesters had used to distribute information outside state-controlled channels. In 2023, Chinese government agencies restricted iPhone use. In 2024, Apple complied the same day Chinese regulators ordered Western messaging applications off its domestic app store. Apple is now spending to diversify assembly to India and Vietnam, but that shift has proved difficult. Amid reported directives from Chinese officials to curb equipment and technology transfers to India, Foxconn recalled hundreds of Chinese engineers from its Indian facilities.

A purchase agreement with CXMT would begin the same process, except now at the memory layer. Apple’s dependence on Chinese assembly did not arise because alternative factories were technically impossible. It developed over time. Apple built scale, expertise, infrastructure, and supplier networks in China, making alternatives progressively less competitive. Despite spending heavily to diversify assembly beyond China to decouple its assembly line from Beijing’s leverage, Apple’s proposed adoption of CXMT memory would initiate the same dependence one layer deeper in the hardware stack, trading short-term margin optimization for long-term structural vulnerability.

Capital Is the Attack Vector

In high-volume memory manufacturing, scale enables continued market viability. Memory suppliers must be able to endure inevitable periods of low demand. In 1995, ten DRAM suppliers held roughly 80 percent of the market. Since then, Hitachi, IBM, LG Electronics, Mitsubishi Electric, NEC, and Texas Instruments have left the market. Even government interventions could not change market dynamics. Germany extended Qimonda multiple bailout packages, yet it eventually liquidated. Japan infused ¥30 billion of public money into Elpida, and less than three years later, its collapse was the largest manufacturing bankruptcy of postwar Japan. DRAM prices fell below cost, and the suppliers unable to absorb the downturn collapsed. Over time, the number of suppliers consolidated to Samsung, SK Hynix, and Micron, which now hold more than 90 percent of the market.

State-backed enterprises upend these market dynamics by operating outside normal commercial constraints. In 2014, the Chinese government created a national integrated circuit plan, committing state funds to build a domestic chip industry. The following year, a Chinese state-backed group unsuccessfully attempted to buy Micron for $23 billion. Undeterred, the Chinese government stood up three memory ventures: one in flash memory, and two in DRAM, Fujian Jinhua and CXMT, then known as Innotron Memory. Private investors would not fund CXMT, so Hefei, the capital of Anhui province, did. In the decade that followed, CXMT absorbed losses that had bankrupted companies before it. Sustaining such a level of market endurance strongly supports a national security strategic purpose.

What Major Western Manufacturers Would Help Build

While testing is strictly a validation of technical quality, that quality is the direct result of state intervention. Apple, Dell, and HP’s interest in CXMT is driven by cost and supply chain leverage, but a low price is irrelevant if the hardware is unreliable. For example, by meeting Apple’s exacting standards, CXMT has proven that Beijing’s subsidies achieved its national security goals of funding a domestic challenger sophisticated enough to compete in this difficult market.

CXMT is the fourth-largest DRAM producer in the world, accounting for roughly 7.7 percent of market share in 2025. Despite its success in volume, securing the endorsement of a company like Apple is a different type of validation. As a Citi analyst note observed, Apple’s mere consideration of CXMT is “a strong validation of CXMT’s product reliability.” The same note said it changes “market perception of CXMT from a domestic substitution play to a credible global No. 4 DRAM maker.”

Even if Western manufacturers commit to using CXMT chips solely for its Chinese-market products, that commitment does not extend to the rest of the world nor is it limited to commoditized memory chips. While CXMT produces commoditized DRAM today, such potential partnerships will give it the runway to climb the value chain. As its prospectus indicates, CXMT plans to produce the high-bandwidth memory that AI requires. Should CXMT become a partner, it will be a proven supplier to the world and accelerate its program.

In addition, such a partnership would provide the Chinese government a path to its competitors’ margins, eroding the profit gap between manufacturing costs and selling prices. Vendors now compete against the Chinese government, structurally diluting the pricing power of the established oligopoly of Samsung, SK Hynix, and Micron.

Today, CXMT is one of the most profitable memory companies in the world. In the first quarter of 2026, it reported a 33 billion yuan profit on revenue that grew more than 700 percent. However, it is the current AI boom that makes today’s memory producers profitable. CXMT is the same company that lost $1.2 billion in 2024, a year when memory prices were rising. Memory companies are truly tested by the downturns that consolidate the memory market. In the next contraction, state support may allow CXMT to tolerate losses longer than commercial rivals. On the other hand, Micron, the only U.S.-headquartered DRAM producer faces ordinary financial constraints even as it has committed approximately $200 billion to domestic production, including production for defense and aerospace applications.

In the next chip market downturn, CXMT will dictate the pricing floor, as its expansion is directly capitalized by China’s national integrated circuit plan. Insulated by sovereign wealth, the company can indefinitely sustain below-cost pricing, threatening to restart the aggressive consolidation cycles that historically destroyed the West’s secondary memory ecosystems. This dynamic is a direct continuation of the DRAM price wars that drove out Qimonda and Elpida. Each cyclical contraction will structurally transfer market share to a state-capitalized entity.

Given that memory is foundational to many modern systems, with high-bandwidth memory serving as the bottleneck for AI, systemic dependence on CXMT would grant Beijing structural leverage over global computational infrastructure.

Perhaps Washington would rescue Micron from this hypothetical fate, as it rescued Intel. But as history has shown, that is no survival guarantee. Furthermore, a future bailout reveals the true cost of partnership with CXMT. When a dominant consumer tech firm optimizes its margins by sourcing from a strategic, state-backed company, it forces traditional market-driven vendors to operate at a loss. To keep those vendors viable, the U.S. government should repeatedly bridge the financial shortfall.

Adding CXMT to the Entity List

Although Apple has asked the U.S. government for assurances that CXMT will remain off the Entity List, the United States should instead add it. Its designation would impose licensing requirements on the export, reexport, or in-country transfer of the items specified in its entry. That could restrict the U.S.-controlled tools and technology on which its fabs depend, but it would not, by itself, impose a blanket prohibition on Western manufacturers purchasing CXMT chips. The potential effect can nevertheless be substantial. In 2018, after Commerce added Fujian Jinhua, another Chinese DRAM venture, to the list, the company ceased production within five months.

Commerce added Semiconductor Manufacturing International Corporation to the Entity List in December 2020, roughly five months after its July Shanghai offering. The company’s continued operation, however, also shows why designation is not existential. The goal in placing companies on the Entity List is not to put the company out of business. Rather, it is to deny it the Western technology and demand that could turn it into a source of supply-chain leverage for Beijing. Its effect depends on the scope of the listing, the license-review policy, enforcement, and allied coordination. Similarly, in 2022, Apple sought to source NAND flash memory from Yangtze Memory Technologies but suspended those plans after the United States added the company to the Entity List. Despite its addition to the Entity List, Yangtze Memory Technologies continues to exist and expand through a more localized supply chain rather than major Western manufacturer partnerships.

This designation may be seen as a provocation to the Chinese government. Indeed, while the United States previously removed CXMT from a draft Entity List in 2024, CXMT has since cleared the interagency review for designation, yet remains off the published list. While concerns of retaliation are understandable, restraint does not prevent future escalations. In 2023, the Chinese government banned its critical infrastructure operators from purchasing Micron products, in a move widely perceived as retaliation for U.S. export controls. Beijing’s retaliatory actions are dictated by its own strategic leverage and industrial objectives, independent of Washington’s regulatory posture.

Surging memory prices have forced downstream price hikes for Apple and other tech manufacturers, spreading supply constraints across other industries. However, framing this as a trade-off between immediate economic relief and long-term strategic risk ignores the stark operational reality that CXMT cannot offer immediate relief. Its baseline fabrication output is already entirely allocated to domestic Chinese demand.

Today, Samsung, SK Hynix, and Micron supply more than 90 percent of the global DRAM market. They are well positioned to supply what the market demands. However, new capacity takes years to build. By partnering with CXMT today, Western manufacturers provide it with the runway needed to gain a foothold in the market. At the next market downturn, the Chinese government’s support of CXMT will allow it to invest aggressively while its rivals are forced to reduce spending in this capital-intensive market. It is this asymmetric, market-insulted dynamic that will eventually erode their otherwise dominant market share.

The incumbent producers are not blameless. Their record includes price-fixing in the 2000s, and their reluctance to overbuild contributed to supply lagging the AI boom. But they remain critical to solving the shortage. The United States should work with South Korea and other allies to accelerate new fabrication capacity using traditional levers, such as investment incentives, faster permitting, and long-term purchase commitments. The response to today’s shortage should expand allied supply rather than finance and qualify a state-backed competitor. By placing CXMT on the Entity List now, the United States can deny the Chinese government similar leverage before the world’s memory supply comes to depend on it. ↩︎

On July 16, 2025, the Trump administration loosened the sales ban on Nvidia artificial intelligence chips to China. This decision flies in the face of the growing bipartisan consensus in Washington that selling advanced AI chips to China poses an unacceptable national security risk to the United States, demonstrated by the letter that Senators Elizabeth Warren and Jim Banks recently sent to Nvidia. While such a consensus sounds prudent, it is dangerously short-sighted.

Export bans on high-end hardware may seem like a tool of strategic restraint. In reality, they are accelerating China’s technological independence, undermining U.S. influence, and weakening America’s ability to shape the global future of artificial intelligence.

The logic behind the bans is simple: deny China access to cutting-edge chips, and you delay its development of advanced AI systems used for surveillance, cyberwarfare, or military operations. But the real-world effect is often the opposite. China does not stop building. It builds without us, with more urgence and independence.

Take DeepSeek, a Chinese AI firm that achieved an efficiency breakthrough resulting from being constrained by less powerful hardware.^[1] Unable to access Nvidia’s top-tier GPUs, the company redesigned its architecture to run efficiently on more limited chips. By prioritizing computational efficiency and low-level GPU tuning, DeepSeek advanced despite the constraints. Hardware scarcity didn’t block innovation. It provoked it.

We’ve seen this before. In World War I, the United States relied on German optical glass, critical for rangefinders, binoculars, and periscopes. When war cut off the supply, the U.S. faced a strategic crisis. But denial didn’t paralyze American industry. It catalyzed it. Backed by wartime urgency, firms like Bausch & Lomb and institutions like the National Bureau of Standards created a domestic optics industry from scratch.^[2] The U.S. didn’t freeze in place. It caught up and then took the lead.

Every Chinese data center running Nvidia chips today is a node of American influence. As long as Chinese AI systems run on U.S. hardware, the U.S. retains leverage over performance, updates, supply chains, and, most importantly, alignment with global standards. That leverage vanishes the moment China builds a full-stack alternative. Export controls don’t preserve dominance. They motivate self-reliance.

This dynamic also applies to software. In early 2024, a sophisticated backdoor nearly compromised a core open-source utility called XZ Utils.^[3] A rogue contributor spent years building trust with project maintainers before inserting a subtle exploit during the release build process. The malicious code was not visible in the open repository and could have granted remote access to Linux systems worldwide. It was discovered only because one engineer noticed an unusual delay anomaly.

That discovery was a triumph of global transparency. But it was also a warning. Open-source software is not immune to manipulation just because its code is public. Governance matters. Who writes the code, who merges it, and who sets the defaults matters. If China dominates the development of open-source infrastructure, it will quietly shape what the world considers “normal.” Backdoors won’t need to be hidden if they’re built into the standard itself.

This is not theoretical. Much of the world’s AI stack, from model training frameworks to distributed computing libraries, depends on open-source tools. If the technical backbone of those tools becomes defined by institutions aligned with Chinese state priorities, the U.S. will lose more than influence. It will lose visibility. It will be using “open” software which is open only in name.

The real danger of ceding the hardware layer is that the software follows. And when both migrate abroad, so does the power to define what remains visible, what gets reviewed, and what becomes hidden in plain sight.

Meanwhile, the bans aren’t airtight. Chinese companies continue to acquire restricted chips via overseas intermediaries and black-market resellers. Enforcement is difficult, and circumvention is increasingly sophisticated. In the process, American firms lose access to a market they once led while China accelerates efforts to replace them.

Consider what we’re giving up. It is not just the economic value of a $4 trillion company like Nvidia selling globally. More importantly, it also includes the geopolitical leverage that comes from setting the technical standards others must follow. When the world builds on American platforms, America helps shape the rules of the road.

By forcing China to build its own AI ecosystem, we heighten, not reduce, the risk of miscalculation, misalignment, and cyber escalation. The U.S. gains nothing by building a digital Iron Curtain around an adversary that is fully capable of building its own infrastructure and often moves faster under constraint.

If we want to shape the future of artificial intelligence, we need to be in the room – and in the rack. ↩︎

---

1. https://www.nytimes.com/2025/01/28/technology/china-deepseek-ai-silicon-valley.html ↩︎

2. https://gl-history.carnegiescience.edu/news/optical-glass-rivalry ↩︎

3. https://www.reuters.com/technology/cybersecurity/why-near-miss-cyberattack-put-us-officials-tech-industry-edge-2024-04-05/ ↩︎

Maxwell Strachan of Vice writes about Ryval, which is a startup that seeks to allow its users to fund lawsuits. Strachan writes:

A new tech startup plans to become “the stock market of litigation financing” by allowing everyday Americans to bet on civil lawsuits through the purchase (and trade) of associated crypto tokens. In doing so, the company hopes to provide funding to individuals who would otherwise not be able to pursue claims.

Ryval allows its users to purchase tokens to fund lawsuits. Strachan succinctly describes Ryval as a “crypto-infused and lawsuit-focused GoFundMe.”

Kyle Roche, one of the founders of Ryval, states that its goal is to “make access to justice more affordable.” Strachan contrasts Roche’s claim by proposing that Ryval’s website appears to instead prioritize the potential returns for investors.

Empowering everyday people to leverage the judicial system is a worthwhile cause. It is not uncommon for people with valid claims to walk away harmed because of a lack of knowledge or resources. For example, we all likely know someone who has had a valid warranty claim denied for something like an electronic device or motor vehicle. The idea of empowering David against the Goliaths who have continually avoided accountability due to resource imbalance is exciting, righteous, and popular.

That said, as Strachan points out, litigation financing is nothing new. A well-known instance involved Peter Thiel funding Hulk Hogan’s lawsuit against Gawker Media. In fact, Legalist is a litigation financing organization that received funding from the Thiel Foundation.

What makes Ryval’s proposal on litigation financing different is crypto. Ryval states the advantage from crypto is that it lowers the barriers to leverage the judicial system. In doing so, more people can participate in the judicial system by purchasing tokens. These tokens can be redeemed for a potential windfall at the end of litigation. Token owners may also buy and sell them during a case at a price of their choosing.

With this democratization, I can immediately think of two concerns. First, a noted criticism of litigation financing generally is that it may result in more cases, clogging up an already clogged judicial system. While a practical criticism, if the system becomes infused with cases from vulnerable populations with valid claims, this scenario appears worthwhile. Justice should be accessible to all, particularly those whose legitimate claims have been continually ignored because of their lack of resources. The other concern is particularly noteworthy. Unaccredited investors are at a significant disadvantage compared to accredited investors due to a one-year lockup period. If Ryval’s goal is to democratize access to the courts, invigorate interest in the judicial system, and hold the previously unaccountable accountable, that pursuit is meaningful and likely without controversy. However, should Ryval truly intend to leverage its platform to create a betting market as Strachan suggests, this one-year lockup period puts unaccredited investors at a significant disadvantage. A party’s chance at success can significantly change within a year. Summary judgment may even be granted before a year, ending the case entirely, and leaving unaccredited investors stuck holding the bag. This result may only highlight and exacerbate the accredited versus unaccredited investor divide, which is counter to Ryval’s goal.

Setting aside these above concerns and Ryval’s positioning of litigation financing as a betting ring, the concept itself is interesting. It’s like a litigation focused Goldfinch in that it shares a story, vets it, leverages crypto for crowdfunding, and distributes the funds to the parties. Perhaps the true takeaway is that Ryval continues the march towards a popular Web3 goal: the capability to financially organize around almost anything. Although an initiative towards the decentralization of the web, Web3 empowers individuals into collective action. I look forward to reading more about these initiatives and their outcomes. The web is exciting again. ↩︎

Mark Gurman wrote a great piece on Apple’s Chief Operating Officer, Jeff Williams.

Williams also has often relied in meetings on a pocket-size notebook, and colleagues say they make sure to follow up on any part of the conversation they see him write down.

This tidbit is fascinating to me. How come Williams uses a paper notebook? This is the Chief Operating Officer at Apple, a company whose iPhone is, frankly, the most ubiquitous pocket-size notebook around.

Admittedly, I, too, prefer a paper notebook to write down thoughts, even though I really want to just use my phone for notes. Phones make more sense. It’s always with me, my writing is in the cloud and not stuck in one spot, and software makes note taking more organized. There’s no readily apparent downside. Yet I still carry a notebook and pen with me where ever I go.

I find Williams’s habit much more interesting when transposed against Scott Forstall’s note taking habits (paywall):

Forstall takes detailed notes without pen, paper, or laptop. “He listens to you and he starts typing on his iPhone,” says Matt Murphy, a partner at Kleiner Perkins and the manager of a fund at the firm that invests in iOS developers. “You’re thinking he’s not listening and sending a text message, then you realize he’s taking notes.”

Perhaps that’s part of why I eschew using my phone to take notes. I feel people think, actively or subconsciously, I’m ignoring them. Their immediate perception is a bias I have to overcome. When I take notes using a notebook, people seem to immediately think I’m actively listening to them, taking notes

There’s also the pleasant, tactile, and near instantaneous feel of using a notebook. You pull it out of your pocket and start scribbling away. For some reason, taking notes on a phone feels less accessible in comparison. Truly, I don’t believe this feeling is quantifiably objective. It might be easier and more efficient to use my phone. But yet, given the choice, I nearly always opt to use a paper notebook.

Based on my personal interactions, it seems Forstall’s phone preference is the minority. Here’s hoping using phones to take notes eventually becomes more prevalent and socially acceptable. I’ll do my part to keep using Cultured Code’s fantastic Things 3 more often as part of my workflow. Maybe I’ll preface pulling out my phone with a “Let me write this down.” ↩︎

Apple introduced Sign in with Apple, which is its own single sign-on method. It allows users to use a universal login ID across different applications and websites. This approach is very much indicative of today’s Apple using its massive market presence to positively influence privacy culture.

Creating an account for each individual app and website can be painful. Accounts create friction and get in the way of people immediately using the product. Accounts are, however, necessary. How else would you access your personalized social media, purchase products, or access your online finances?

People will generally take the path of least resistance. Developers want to reduce friction. Thus, single sign-on became popular. As noted in my examples above, if a company can tie social media, purchase information, financial information, and other personalized data together, that’s very desirable to advertisers. Advertisers continue to amass as much data as they can.

Thankfully, with Sign In with Apple, Apple is using its unique, market leading position to push back. Sign In with Apple allows clear choice of what information you choose to share with the app or website. It also prevents cross app or website tracking with anonymized email addresses, should you so choose.

What I find fascinating is Apple requiring its Sign In to be included for apps that support other third-party sign-in options. That requirement demonstrates Apple’s willingness to throw its weight around for privacy protections. Some may deem it as overreaching but frankly, amassing as much of my data as possible is overreaching.

I imagine advertisers are unhappy with Apple’s decision. I, however, am thrilled a market leader is pushing back against the insatiable data beast. Here’s hoping Apple’s approach empowers users to care about data privacy. ↩︎

Phil Spencer, Head of Xbox:

We’ve not always lived up to our aspiration of keeping gamers at the center of everything we do when it comes to the experience they’ve had on Windows.

The way Microsoft has pivoted within the past few years is incredibly impressive. While other companies seek to isolate themselves and their customers into a locked ecosystem, Microsoft’s open approach is both welcome and refreshing. Who in 2010 would have predicted Google would be the one locking things down while Microsoft opens things up? ↩︎