The Architecture Gatekeeper.Original analysisNot investment advice

The Arm China story is not just about a rogue CEO. It is about control over the architecture channel. Arm owns one of the most important computing ecosystems in the world, but its access to China runs through an independent entity it does not directly control. In 2021, that looked like a dramatic corporate heist. In 2026, it looks like something more durable: the architecture gatekeeper problem.

In 2021, the Arm China story looked like corporate chaos.

A fired CEO who would not leave. A company chop that mattered more than the board vote. Security guards blocking access. A joint venture declaring its own path. A Chinese distributor presenting itself like an independent IP company.

The uploaded SemiAnalysis article called it the semiconductor heist of the century.¹ The wording was dramatic, but the underlying point was sharp. Arm had not necessarily lost its core IP. It had lost practical control over the entity that controlled access to China.

That distinction matters. Because in semiconductor IP, the product is not only the core design. It is also the licensing channel. And in China, that channel is Arm China.

The real asset was not only Arm’s IP. It was the right to stand between Arm and the Chinese customer.

---

1. The 2021 thesis was about the distribution layer

The uploaded SemiAnalysis piece was not a complaint about an isolated personnel dispute.¹ It was an argument about how power had quietly shifted from Arm Limited to the entity that controlled Arm’s China channel.

The article’s public skeleton is well known. SoftBank had acquired Arm in 2016. In 2018, Arm sold a 51% stake in its China joint venture to a consortium of Chinese investors for USD 775 million.¹ The joint venture, Arm China, received exclusive rights to distribute Arm IP in the People’s Republic of China.¹ Allen Wu, the executive running it, resisted removal after the board voted to oust him. The company seal — the chop — gave him operational authority that the board vote alone could not override. Security at Arm China’s offices reportedly blocked Arm representatives. Arm China began presenting itself as an independently operated Chinese-owned company.¹

The uploaded article reported that Arm China had claimed more than 20 billion cumulative shipments, more than 90 partners, and 29 partners in mass production using the IP, with a 400+ person China-based R&D team.¹ Arm China had also announced or rebranded a local XPU IP family, including NPU, SPU, ISP, and VPU blocks. The deeper claim of the article was not that Arm’s files had been stolen. It was that Arm China had become independent enough to set its own direction inside the channel that mattered most.¹

2. In IP, distribution is a control surface

It is easy to think of semiconductor IP as a file you license. It is not. It is a long bundle of contracts, support, and trust. When the licensing channel is concentrated in one entity, that entity becomes a control surface, whether the technology owner intends it to or not.

In IP, the channel can become a control surface. A distributor that controls customer access, contract terms, support, and pricing — and that does so under local law — can become powerful even without owning the underlying technology. That is the lens through which the Arm China story is best read.

In IP, distribution is a control surface. In China, that control surface is Arm China.

3. The governance drama changed, but the structure remained

The acute phase of the Arm China governance dispute was visible because Reuters and others covered it in detail. By 2022, Reuters reported that Arm China’s board had appointed two new co-CEOs and that Arm had said the leadership change complied with Chinese law.³ Reuters noted Wu had previously refused to step down after a 2020 board vote and that the dispute had complicated Arm’s ability to audit the unit’s financials.³

The leadership change matters, but it is not the whole story. The deeper structure — the bundle of contracts and equity that made Arm China the gatekeeper in the first place — was untouched by a change of CEOs. The reading that survives is simpler. The rogue-CEO drama was the symptom. The gatekeeper structure was the underlying condition.

The rogue-CEO drama was the symptom. The gatekeeper structure was the disease.

4. Arm China is still the China gatekeeper

The cleanest evidence for the 2026 reading sits inside Arm’s own filings. Arm’s FY2025 Form 20-F describes Arm China as Arm’s primary access point to the PRC market for the foreseeable future, with substantially all PRC-related revenue earned through the Intellectual Property License Agreement (IPLA) with Arm China.² The 20-F describes Arm China as having certain exclusive rights to sublicense Arm standard IP offerings to PRC customers.²

The structure is the point. Arm owns the architecture and the IP. Arm China owns the doorway into the market that uses it most heavily. Whatever the 2021 narrative looked like, the 2026 structure makes the gatekeeper concept literal rather than rhetorical.²

Arm owns the architecture. Arm China owns the doorway.

5. Arm does not directly control Arm China

The next layer of the 20-F disclosure is the part that makes the gatekeeper framing structural rather than dramatic. Arm states that neither Arm nor SoftBank controls Arm China’s operations.² Arm transferred its remaining equity interest in Arm China to a SoftBank subsidiary in March 2022.² Arm now holds a 10% non-voting interest in Acetone Limited, which represents an approximately 4.8% indirect ownership interest in Arm China, with no direct management rights.²

The asymmetry is the point. The economic stake Arm holds in Arm China is small. The operational dependency on Arm China for access to the PRC market is large. The 2021 governance fight calmed down. The structural problem — Arm depending on a China gatekeeper it does not control — did not.²

The governance fight calmed down. The structural dependency did not.

6. The relationship is financially meaningful

Arm’s own filings quantify the size of the dependency. The FY2025 20-F discloses that revenues attributable to Arm China were approximately 17% of total revenue in FY2025, approximately 21% in FY2024, and approximately 24% in FY2023.² PRC revenues overall, including revenues earned indirectly through Arm China, accounted for approximately 19%, 22%, and 25% of Arm’s total revenue in FY2025, FY2024, and FY2023 respectively.²

The percentage is moving in a direction that looks comforting in isolation. The absolute exposure is still meaningful, and concentration in a single channel matters more than its share of the overall mix suggests. A 17% revenue line that flows through one gatekeeper is not the same risk as a 17% revenue line spread across thousands of customers.

Arm China is smaller as a percentage than before, but still too important to ignore.

7. The agreement lasts for decades

The most underappreciated detail in Arm’s 20-F is the duration of the licensing relationship. Arm discloses that the IPLA with Arm China has an initial term running through April 23, 2048, with automatic 10-year renewals thereafter.² Within that structure, Arm China can sublicense Arm’s standard IP offerings to PRC customers, and there are no material restrictions on the prices Arm China may set for sublicenses.²

That is a multi-decade structure. It is not a reseller agreement that can be re-papered next quarter. The 2021 dispute happened in the early years of an arrangement that, by Arm’s own filings, is intended to extend well into the second half of the century.

The China gatekeeper problem is embedded into the licensing model.

8. The risk is no longer just legal, it is strategic

Arm describes a specific bundle of operational risks tied to the Arm China relationship.² Arm depends on financial information provided by Arm China and has had past issues obtaining timely and accurate information. Arm has experienced late payments from Arm China. Arm is limited in its ability to monitor how Arm China protects Arm IP and customer data. Arm’s contractual protections are, in practice, only as strong as visibility and enforcement allow.²

The modern version of the 2021 concern is not simply “Has someone stolen the files?” It is a list of operational questions where the answers are harder to verify than in a normal arms-length supplier relationship.

None of these are unique to Arm. Any IP business with a concentrated regional distributor faces a version of the same surface. The point is that the language used in 2021 to describe the dispute can now be replaced with the language used in 2026 by Arm itself: a multi-decade IPLA, no direct management rights, limited visibility, primary PRC access point.² The shape did not change. It became the official record.

Arm can own the architecture and still lack full visibility into the China channel.

9. Arm China’s local IP push matters more in the AI era

The uploaded 2021 article also reported that Arm China was developing its own IP under the XPU banner, with NPU, SPU, ISP, and VPU blocks, and that it positioned this as an in-house IP roadmap rather than a pure Arm-Limited reseller story.¹ Arm China has continued to discuss self-developed product families publicly. Naming and scope should be treated with caution, but the broad direction Arm China itself describes has been consistent.⁶

The strategic point is simple. The distributor did not remain only a distributor. It became an IP platform. In the AI era, a gatekeeper that also publishes its own NPU and CPU IP is not just a channel risk. It is an architectural risk, because every additional local design weakens the case that the channel is purely a conduit for Arm Limited.

The local distributor is trying to own more of the local architecture stack.

10. Why export controls make the channel more strategic

The geopolitical layer makes the gatekeeper question heavier. Arm’s 20-F describes that U.S. and U.K. export-control regimes can require licenses for certain high-end processors going to PRC customers, that Arm’s Neoverse V-series processors meet or exceed performance thresholds under those regimes, and that Arm has addressed some demand with other CPU cores that do not exceed those thresholds.²

The China channel is no longer just commercial. It is geopolitical. As more high-end compute becomes restricted, the value of the entity that controls the licensing path into China — both for what it can deliver and for what it has to refuse — increases.

The more restricted advanced compute becomes, the more strategic the China licensing channel becomes.

11. RISC-V is the long-term architecture pressure

RISC-V is not an immediate Arm replacement. It is a sovereignty option.

12. Arm is stronger now, which makes the gatekeeper problem bigger

The strange feature of the 2026 Arm China story is that it gets more important precisely because Arm Limited’s position is stronger. Arm’s public materials describe a multi-segment ecosystem: smartphones, Edge AI, Physical AI for robotics and automotive, and Cloud AI through Neoverse and the data-centre roadmap.⁴ The narrative is no longer simply “phones run on Arm.”

When Arm was mostly a smartphone-IP story, Arm China was a sales-line problem. In the AI era, with Arm IP touching cloud, edge, automotive, embedded systems, and infrastructure, the same channel becomes more consequential. The same gatekeeper now stands between Arm and a much bigger surface of Chinese demand.

Arm’s success makes Arm China’s gatekeeper role more valuable.

13. The actual 2026 thesis

The correct claim here is not “Arm China stole Arm.” That language is too crude and was not the substance of even the 2021 article, which was careful to separate the question of confirmed IP theft from the question of practical control.¹

The correct claim is more structural. Arm China revealed a deeper semiconductor IP risk: losing control over distribution can become almost as serious as losing control over technology. The 2021 article was right to focus on the governance shock, the company chop, the independent branding, and Arm China’s own IP ambitions. The 2026 update is even more structural. Arm’s own filings show that Arm China remains independent, that Arm does not control it, that the IPLA runs deep into the future, and that Arm still depends on Arm China for access to the PRC market.² The heist framing aged into a gatekeeper problem.

“Arm China is not Arm’s death sentence. It is Arm’s China-control tax.”

14. What could break the thesis

The thesis here is that the Arm China gatekeeper structure is durable and that the AI era makes it more important. There are honest reasons that reading could be wrong, or could become wrong on a timescale that matters.

Arm can remain globally dominant and still pay a China-control tax.

15. What could break the bear case

There are equally honest reasons the bear case here could be too dark, and the gatekeeper structure could remain manageable for years.

16. What to watch

The honest position is that the gatekeeper structure is durable but the outcome over the next few years is open. These are the signals worth tracking.

17. The architecture gatekeeper

The Arm China story is not just about a rogue CEO.

It is about control over the architecture channel. Arm owns one of the most important computing ecosystems in the world. But its access to China runs through an independent entity it does not directly control.² In 2021, that looked like a dramatic corporate heist.¹ In 2026, it looks like something more durable: the architecture gatekeeper problem.

If there is a lesson for the rest of the semiconductor IP world, it is this. The technology is hard. The licence channel is also hard. When the channel is concentrated, governed locally, and extends to 2048, the channel becomes its own form of architecture — one that does not always sit on the same side of the table as the IP owner.²

18. Glossary

---