Huawei under sanctions: a tech stack rebuilt

The ban on a supplier that had become strategic

Huawei was not sanctioned from a marginal position. When U.S. and European restrictions intensified, the company was already one of the world’s most competitive telecom suppliers, with a strong position in 5G. Dell’Oro Group said Huawei and ZTE had increased their shares of the global telecom equipment market in 2019, while Nokia and Cisco declined. That context matters: Huawei was not only seen as a cybersecurity risk, but also as a company able to shape a critical layer of the connected economy.

The U.S. case is specific. The United States dominates digital platforms, operating systems, cloud computing, advanced semiconductors and parts of enterprise software, but it no longer has a national champion comparable to Huawei, Ericsson or Nokia in mobile-network radio equipment. Against Huawei, Washington could not field a direct industrial equivalent. It therefore used blacklists, export restrictions, security diplomacy and pressure on allies. Open RAN is sometimes presented as a diversification answer, but it is a disaggregated and interoperable architecture, not the immediate return of a large integrated U.S. equipment maker.

The official narrative focused on cybersecurity. That framing should not be dismissed. 5G networks are critical infrastructure: they rely on software, updates, maintenance, radio equipment and network cores that can expose operators to supplier lock-in, disruption or interference. Reuters noted in 2019 that Washington had not publicly produced hard evidence showing Huawei equipment had been used for spying, but that U.S. officials emphasized a more structural risk: in 5G, an explicit backdoor is not required if a supplier retains decisive software access.

This distinction is essential. Saying that public accusations have not always been accompanied by open evidence does not mean the risk is imaginary. The U.K. Huawei Cyber Security Evaluation Centre, which assessed Huawei equipment in Britain, mainly pointed to software-engineering and cybersecurity weaknesses rather than public proof of organized espionage. The European Union defended a “high-risk supplier” approach in its 5G toolbox. That shifts the debate: the issue is not only finding a spy implant, but assessing an infrastructure relationship with a supplier subject to a foreign power.

Europe, however, cannot apply this reasoning only to China. The Snowden revelations, surveillance reports targeting Angela Merkel and other European officials through cooperation with Danish intelligence, and allegations involving Airbus and other European interests show that U.S. allies also conduct intelligence operations when strategic interests are at stake. The point is not to claim that China and the United States are identical: their political systems, legal frameworks and risks differ. The point is that foreign technological exposure remains strategic exposure, even when it comes from an ally.

Regional sources frame the topic differently. Reuters mainly covers the techno-industrial confrontation between Washington, Beijing, Brussels and operators. European and British institutions emphasize infrastructure security. Huawei presents its trajectory as a response through R&D and vertical integration. The South China Morning Post focuses more on product recovery and the national symbolism of smartphones. HPCwire and arXiv add technical perspective, but with different levels of verification.

These sources do not have the same blind spots. Huawei defends its figures and roadmap. Western institutions reason within a national-security framework. Reuters and AP have strong factual discipline, but often frame the issue through Washington-Beijing rivalry. The South China Morning Post, owned by Alibaba, remains useful on Chinese products, but should be read carefully on sovereignty issues. HPCwire and arXiv add technical granularity, without replacing independent audits or benchmarks.

The fuel of the comeback: capital, domestic demand and substitution

Huawei’s resilience is not only an engineering story. It required capital, sellable products, domestic demand and financial support channels. In its 2025 annual report, Huawei reported CNY880.9 billion in revenue, including CNY375.0 billion from ICT infrastructure, CNY344.5 billion from consumer business, CNY77.3 billion from Digital Power and CNY45.0 billion from intelligent automotive solutions. It also said China accounted for CNY616.2 billion in revenue, or about 70% of the total. These are company-reported figures, but they explain why Huawei was able to keep investing under sanctions.

The same report gives CNY192.3 billion in R&D spending in 2025, equal to 21.8% of revenue, CNY1.382 trillion invested in R&D over ten years, 114,000 R&D employees and 165,000 active granted patents worldwide. These figures do not automatically prove technological leadership, but they show an unusual capacity for sustained effort. Huawei did not simply wait for restrictions to ease: it replaced components, qualified local suppliers, redesigned boards, strengthened software systems and shifted part of the competition toward system integration.

Chinese backing took several forms. The recovery should not be reduced to a public bailout, because Huawei retained cash-generating businesses. But the environment clearly mattered: banks, operator orders, public-sector demand, patriotic consumption, local suppliers, semiconductor funds and industrial policy. It mainly worked as an execution environment: credit, outlets, political alignment, local supply chains and temporary acceptance of imperfect solutions. China gave Huawei a domestic market large enough to test, correct, industrialize and sell strategic alternatives.

In smartphones, this strategy becomes visible. Reuters describes the Pura 80 line as another step in Huawei’s return to China’s premium segment, with four models and prices ranging from CNY6,499 to CNY9,999 for the Ultra. The range emphasizes XMAGE, macro telephoto lenses and AI functions able to identify objects or places. But the high price drew criticism from Chinese users, and the absence or complexity of Google services remains a major limitation outside China.

The Mate XT illustrates another dimension: Huawei uses some products as engineering showcases. AP described its global launch in Kuala Lumpur as a symbolic victory in the context of U.S. restrictions, while also noting its limits: a €3,499 price tag, durability questions, supply constraints and app limitations. Such a device does not prove global smartphone dominance. It shows instead that Huawei can still take visible hardware risks at a time when Apple remains more cautious on foldable formats.

HarmonyOS Next is more strategic than spectacular. Reuters reported that HarmonyOS overtook iOS in China in the first quarter of 2024 to become the country’s second-best-selling mobile operating system behind Android, according to Counterpoint, and that it had not yet launched on smartphones outside China. The same article reports a Huawei statement claiming more than 900 million devices running HarmonyOS and 2.4 million developers in the ecosystem. These are declarative figures, but they indicate that Huawei is trying less to replace Android globally than to build Chinese critical mass.

In AI, Huawei is trying to shift the debate. A direct comparison between one Ascend chip and one Nvidia chip is not enough. Reuters reported that Huawei plans an Atlas 950 in the fourth quarter of 2026 supporting 8,192 Ascend 950DT chips in 160 cabinets and 1,000 square meters, followed by an Atlas 960 in the fourth quarter of 2027 with up to 15,488 chips. These figures are company announcements reported by Reuters: real performance, yields, costs, availability and software stability will need confirmation by customers and independent benchmarks.

The arXiv paper on CloudMatrix384 describes a system combining 384 Ascend 910C chips and 192 Kunpeng CPUs around a high-bandwidth unified bus. The authors report 6,688 tokens per second per NPU for prefill and 1,943 for decode in their evaluated setup. These results illustrate the “system against chip” logic: compensating part of the process-node gap through interconnect, parallelization and software optimization. They do not replace independent industry benchmarks.

High-performance computing adds another layer, if handled precisely. HPCwire reported that LineShine, a CPU-only exascale-class supercomputer announced in Shenzhen, would be built in two phases, with the first phase reportedly based on 100 Huawei Kunpeng servers totaling 12,800 cores. That does not allow the article to call LineShine a Huawei supercomputer. It only supports the claim that Kunpeng appears as a building block in the first phase of a Chinese national project. Details on the second phase, timing and effective performance remain uncertain.

Products that may matter, and what they do not prove yet

The products to watch do not all belong to the same timeline. Pura phones, the Mate XT and future foldables matter as showcases for design, computational photography and differentiation against Apple or Samsung. Their potential outside China will depend less on specifications than on software trust, apps, service networks, distributors and acceptability for users accustomed to Android or iOS. The human question is simple: does formal innovation really improve everyday use, or does it mainly demonstrate industrial capability?

The Chinese market nonetheless shows that Huawei has a solid outlet. Reuters reported that during China’s 618 shopping festival in 2026, smartphone sales fell 13% year on year, while Huawei grew 19% and led the market with a 21% share. This data should be interpreted carefully: it covers a specific promotional period, not annual global dominance. It does confirm, however, that Huawei can turn its technology narrative into sales in a demanding domestic market.

Ascend, CloudMatrix and Atlas are the most structural products, but mainly for China in the short term. They can matter in Chinese data centers, local clouds and possibly in some non-Western markets where cost, availability and sovereignty are weighed differently. They do not yet prove that Huawei can replace Nvidia. The real question is therefore less “can Huawei beat Nvidia?” than “can Huawei provide enough compute, sufficiently available and integrated, to reduce China’s exposure?”

The “post-Moore” strategy reinforces this shift. Reuters reported that Huawei is promoting a system-performance logic, sometimes presented as Tau Scaling or “Her’s Law”, to compensate for limits on access to advanced lithography machines. The idea is to gain through chip integration, interconnect and reduced data movement. Uncertainty remains high: Reuters notes that heat, real efficiency and timing questions are still open. This point should therefore be treated as a technical ambition, not as established proof.

Kunpeng and TaiShan are less visible but important. General-purpose server CPUs can support clouds, public administrations, compute centers, public infrastructure and HPC projects. Their value is not only raw performance: it is supply continuity, integration with local software and the ability to build infrastructure that Beijing considers controllable. That logic may interest some countries outside China, but it will face the same questions: auditability, security, compatibility, financing and long-term supplier relationship.

The hierarchy outside China is uneven. The most exportable vectors outside the West appear to be 5G-A, private industrial networks, Digital Power, solar, storage and selected operator services, especially in Asia, the Middle East, Africa and Latin America. Huawei also claims more than 60 million 5G-A users by the end of 2025, a declarative figure that mainly confirms the strategic importance given to this intermediate generation. Premium smartphones and foldables remain possible but constrained by software and distribution. Ascend, Atlas, CloudMatrix and HarmonyOS Next are mainly strategic for China in the short term. Kunpeng servers and Huawei cloud can interest non-Western markets, but they will be difficult to impose in European or U.S. critical infrastructure.

Smart vehicles complete the stack. Huawei provides software cockpits, assisted-driving systems and components for connected vehicles, without being a conventional carmaker. Its intelligent automotive business reached CNY45.0 billion in 2025 according to its annual report, up 72.1%. That is a significant increase, but it still depends on partners, regulation, liability in accidents and strong competition from BYD, Tesla, Xiaomi, Xpeng and other Chinese carmakers.

Huawei remains constrained: local suppliers still catching up, exposure to the Chinese market, a decisive political environment and uncertain international acceptance. But the company has changed nature. From a contested global supplier, it has become an integrated Chinese industrial stack. Its strength does not come from one product. It comes from its ability to connect telecoms, chips, software, servers, cloud, vehicles and energy into a coherent strategy.

For Europe, the main lesson is not to choose Huawei against the United States, or the United States against Huawei. It is to recognize that digital sovereignty cannot be selective. Critical infrastructure dependent on a foreign supplier must be assessed for what it technically, legally and politically enables. Technology is not merely a purchase. It is a long-term relationship of maintenance, data, financing, updates and power.

FAQ

Was Huawei banned only for cybersecurity reasons?

No. Cybersecurity is the official motive and a real risk. But the industrial context also matters: Huawei was already very strong in 5G, while the United States no longer had a comparable national radio-equipment supplier.

Can Huawei really compete with Nvidia in AI?

Huawei can reduce China’s exposure to Nvidia if Ascend, CloudMatrix and Atlas systems deploy at scale. But performance, software maturity, supply and real cost still need independent confirmation.

Can Huawei products matter outside China?

Yes, especially in some non-Western markets and in areas such as private 5G, energy, servers or premium camera phones. In Europe and the United States, political and software distrust will remain a major obstacle.