PROFILE: Huawei, the Champion Forged Against Its Will — Born of Sanctions, King of Chinese AI

PROFILE: Huawei, the Champion Forged Against Its Will — Born of Sanctions, King of Chinese AI

Introduction: The Paradox of the Enemy Transformed Into a Giant

Washington Wanted to Crush Huawei — It Propelled It Instead

In 2019, the Trump administration placed Huawei Technologies on the U.S. Department of Commerce's Entity List, cutting the company off from Qualcomm, Intel, ASML equipment, and advanced manufacturing chips. The objective was explicit: weaken, or even destroy, China's technology champion. In 2020, Washington tightened the restrictions further. In 2022, export controls expanded to the entire advanced Chinese semiconductor industry. With each new wave, the same prediction: China can't do it alone. Huawei will collapse.

By June 2026, the reality is entirely different. Bloomberg reveals that China is preparing a plan worth 2 trillion yuan ($295 billion) over five years to build a national AI infrastructure network. The key condition: at least 80% of chips must come from domestic suppliers. And the domestic actor capable of delivering at the required scale is Huawei, with its Ascend 910B, 910C lineup and the new 950PR. Nvidia is de facto excluded. AMD likewise. The company that was supposed to be erased has become indispensable.

A National Champion by Default — but a Real Champion

We must be precise about what "champion by default" means. Huawei did not win the Chinese AI market because it was the best in a competitive market. It won because its American competitors were legally expelled. But "by default" does not mean "by accident." Huawei invested massively in its capabilities since 2019, knowing its survival depended on mastering the semiconductor value chain. The result: the Ascend 910B chip, produced on SMIC's 7nm process, achieves according to domestic Chinese benchmarks between 60 and 70% of Nvidia's H100 performance on transformer training workloads. A significant gap, but far narrower than two years ago. And the 950PR surpasses by some measures the only Nvidia chip still authorized in China — the H20.

The $295 Billion Plan — and Huawei's Central Role

The NDRC Plan's Architecture

The plan prepared by the National Development and Reform Commission (NDRC) is industrial before it is technological. It calls for building a network of interconnected data centers at national scale — regional hubs (Wulanchabu, Guizhou, Ningxia, Hangzhou) linked into a unified network by 2028. Operators: primarily China Mobile and China Telecom. Financing: sovereign debt and long-term special government bonds. When integrating the necessary power grid upgrades, the entire project could reach 5 trillion yuan, or $740 billion.

The 80% domestic sourcing rule is the plan's most powerful lever. It means that of $295 billion in infrastructure investment, at least $236 billion will go to Chinese suppliers. Huawei is the primary beneficiary, but not the only one: Biren Technology (BR100 and BR104 chips), Moore Threads (MTT S80), and Alibaba Cloud (Hanguang 800) all received in May 2026 official government approval for deployment in government and secure data infrastructure. Huawei projects $12 billion in AI revenue for 2026, growing at 60% year-over-year. Its primary client: ByteDance, with orders exceeding $5.6 billion for 2026 alone.

Jensen Huang Said It — Nvidia Has "Largely Ceded" the Chinese Market

Nvidia CEO Jensen Huang publicly declared in May 2026 that Nvidia had "largely ceded" the Chinese market for advanced AI accelerators. This declaration, made at an investor conference, constitutes an explicit acknowledgment of the structural change the NDRC plan represents. China previously accounted for between 20 and 25% of Nvidia's datacenter revenue. With the 80% domestic mandate, this market is formally closed — not merely because of American export controls, but because Chinese industrial policy has now institutionalized exclusion.

The only Nvidia chip still legally sellable in China for AI applications is the H20 — specifically downgraded to satisfy American export regulations. But benchmarks published by domestic Chinese researchers show that Huawei's Ascend 950PR surpasses the H20 on the main performance metrics. In other words: even where Nvidia is still permitted to sell, its product is no longer competitive against the domestic rival.

Huawei's Story: From a Shenzhen Startup to a Global Tech Champion

1987 — A Telephone Switch Reseller in a Shenzhen Warehouse

Ren Zhengfei founded Huawei Technologies in 1987 with capital of 21,000 yuan (less than $3,000 at the time). The company began by reselling Hong Kong telephone switches. In the 1990s, it developed its own telecom equipment and began penetrating China's rural markets, ignored by foreign multinationals. Its growth rested on a combination of aggressive pricing, strategic political relationships with the Chinese government, and an intensive R&D program funded by systematically reinvested commercial margins.

In the 2000s, Huawei massively internationalized — Africa, Southeast Asia, Eastern Europe, then Western Europe. In 2012, the company surpassed Ericsson to become the world's largest telecom equipment manufacturer. By 2018, it was on the verge of overtaking Apple in smartphones. Its 5G network was deployed in dozens of countries. That was when Washington decided to strike.

2019–2022: Sanctions as an Involuntary Accelerator

The 2019 restrictions did effectively cripple Huawei's smartphone division — its Android operating system was revoked, cutting off access to the Google Play Store and the entire Western app ecosystem. Smartphone sales plummeted catastrophically. But Huawei's leadership made a decisive strategic choice: pivot massively toward AI and infrastructure. Using revenues from B2B activities (5G networks, servers, cloud), the company funded a considerable R&D push on Ascend chips, the CANN compute platform (an alternative to Nvidia's CUDA), and the Atlas cluster lineup.

In August 2023, Huawei stunned the global industry by launching the Mate 60 Pro — a smartphone equipped with a Kirin 9000S chip manufactured by SMIC in 7nm. This launch, which came at the very moment U.S. Commerce Secretary Gina Raimondo was visiting Beijing, was interpreted as a deliberate demonstration: China can produce advanced chips despite sanctions. It's not at the level of Apple's 3nm or TSMC's 4nm — but it's there, it works, and it's improving.

The Ascend Chip — The Backbone of China's Sovereign AI

Ascend 910B, 910C, and 950PR — Rapid Progression

Huawei's Ascend lineup is the heart of its AI offensive. The Ascend 910B, produced on SMIC's 7nm process, was validated in 2023 as the first domestic Chinese AI accelerator capable of competing with Nvidia's entry-level offerings. The Ascend 910C improves floating-point compute performance and energy efficiency. The Ascend 950PR, whose specifications were partially published in 2025–2026, represents according to available data superior performance to Nvidia's H20 on inference tasks — the dominant use case in enterprise deployments.

In 2025, Huawei shipped approximately 812,000 Ascend chips — a considerable figure but still well below Nvidia's annual production capacity, which runs into the millions of datacenter units. The main bottleneck remains high-bandwidth memory (HBM), of which domestic Chinese production is limited. Experts estimate that domestic suppliers will cover approximately 76% of Chinese AI chip demand by 2030, even as that market grows toward $67 billion that year. The supply chain remains a limiting factor — but Huawei is making progress.

The CANN Software Stack — The Alternative to CUDA

In the AI industry, Nvidia's dominance rests not only on its chips — it rests on CUDA, its proprietary programming environment adopted by virtually all machine learning frameworks for over a decade. Hundreds of thousands of global developers have learned to code in CUDA. AI libraries, frameworks (PyTorch, TensorFlow), pre-trained models — all optimized for CUDA. This creates a compatibility gap that slows adoption of any alternative.

Huawei is developing CANN (Compute Architecture for Neural Networks) as a direct alternative. The company claims growing compatibility with major frameworks, and recent versions of PyTorch include native Ascend support via the MindSpore abstraction layer. ByteDance, Huawei's primary AI client, confirmed internal investments to adapt its development infrastructure to CANN. The gap remains real — but every billion dollars of ByteDance orders creates incentives to close it faster.

Huawei Beyond Chips — The Total Technology Empire

5G, Cloud, Autonomous Vehicles, Smartphones

AI and Ascend chips are Huawei's new front, but the company remains a diversified first-tier player across several industries. In 5G, despite exclusions imposed by the United States, the United Kingdom, Australia, and other Western allies, Huawei maintains infrastructure contracts in dozens of countries across Africa, Southeast Asia, the Middle East, and Latin America. These deployments represent both revenues and strategic positions in the critical networks of non-Western countries.

In the autonomous and connected vehicle sector, Huawei has developed a complete technology platform — Huawei HiCar, ADS 3.0 (advanced driver assistance system), and partnerships with Chinese manufacturers like AITO, Avatr, and Luxeed. In B2B cloud, Huawei Cloud is China's third-largest cloud provider, behind Alibaba and Tencent. In smartphones, the return of the Mate lineup with a domestic chip in 2023 marked a partial resurrection — sales remain well below pre-2019 levels, but the trend is positive.

The Governance Structure — the World's Most Opaque Tech Company

Huawei is a company apart in the global technology landscape: it is not publicly listed, it is officially owned 100% by its employees through an internal share mechanism, and it systematically refuses to disclose its ownership structure in detail. This opacity has for years fueled Western government suspicions about potential links with the People's Liberation Army and Chinese intelligence services. Reports from the U.S. Congress (2012), the British government (HCSEC), and other allied agencies have pointed to potential vulnerabilities in Huawei equipment without being able to establish formal proof of a deliberate backdoor.

This uncertainty is itself a strategic data point. The exclusion of Huawei from Western 5G networks rests not on demonstrated proof of malice — it rests on a risk assessment in the context of geopolitical competition. Founder Ren Zhengfei has always denied any links with the Chinese government beyond the normal relationship between a company and its state. But in a world where China's 2017 National Intelligence Law requires all Chinese organizations to cooperate with intelligence services on request, the distinction between voluntary and compelled cooperation is legally difficult to maintain.

The Lesson for the West — What This Story Teaches About Technology Sanctions

Sanctions as an Accelerator of Adverse Technological Sovereignty

The story of Huawei between 2019 and 2026 offers a painful lesson for those who believe technology sanctions can definitively contain a determined adversary. Sanctions can delay, they can increase costs, they can force short-term quality compromises. But if the adversary has China's size, resources, and political will, they end up accelerating domestic development of what they were meant to deny. China in 2026 has a more advanced chip industry than it would have had without sanctions — because the sanctions eliminated any alternative to domestic investment.

The situation is different for smaller or less capitalized countries. Sanctions against North Korea or Iran have lasting effects precisely because these countries lack the industrial base, human capital, and financial resources to develop a sovereign technology supply chain. China is in a different category — and the technology containment strategy must be rethought accordingly.

What Nvidia Loses — and What Silicon Valley Cannot Recover

Nvidia's loss of the Chinese AI market is a structural change, not a cyclical one. Even if American export controls were lifted tomorrow, the 80% domestic sourcing rule from the NDRC plan would mean Nvidia could at best access 20% of the Chinese national AI network market. And in that residual market, the Ascend 950PR is now more performant than the only Nvidia product still authorized. The displacement dynamic is twofold: regulatory and competitive.

For Silicon Valley more broadly, the loss extends beyond Nvidia's direct revenues. China represented a testing ground, a large-scale deployment environment, and an accelerated learning environment for AI technologies. The data generated by hundreds of millions of Chinese users, the use cases developed in that context, the optimizations induced by performance constraints — all of this constitutes a learning base that American tech companies will not recover. The technological decoupling is real and bidirectional.

The Trajectory — Where Huawei Is Going in the Next Five Years

Huawei's Technology Bets for 2027–2030

Huawei is currently investing in several directions simultaneously. In semiconductors, the objective is to master the 5nm process in collaboration with SMIC — which involves working around the impossibility of accessing ASML's EUV lithography equipment by developing more complex but functional multi-patterning processes. In AI, the development of MindSpore 3.0 and the Atlas 900 architecture aims to provide a supercomputing infrastructure capable of competing with Meta and Microsoft Azure solutions for training large language models.

In telecom, Huawei is positioning its 5.5G platform (sometimes called 5G Advanced) as the next generation of network infrastructure, with offerings already presented in dozens of non-Western markets. In smart vehicles, its ADAS architecture is increasingly recognized as competitive with Mobileye and Tesla FSD offerings — on markets where Chinese manufacturers have access. This diversification is the ultimate resilience strategy: never again depend on a single market segment or a single buyer.

The Structural Limits That Even $295 Billion Cannot Erase

Despite its ascent, Huawei and the Chinese chip industry face serious structural limitations. Domestic high-bandwidth memory (HBM) production — the indispensable memory for modern AI chips — remains insufficient. SK Hynix and Samsung, the world's leading producers, can no longer sell to China under American export restrictions. CXMT, China's primary DRAM producer, is working to develop HBM, but the timelines are measured in years. Without sufficient domestic HBM, the $295 billion plan risks hitting a physical ceiling — you can't build as many Ascend chips as there is money to buy them if the memory is missing.

This ceiling is real, but not insurmountable over the long term. China is investing considerable sums in advanced memory research. And some observers suggest that the $295 billion plan is as much a political signal as a technical program — a way for Beijing to tell the rest of the world, and its own industry, that the path toward technological independence is irreversibly engaged.

Military Artificial Intelligence — Huawei's Next Frontier

Defense Contracts and the Reality of Dual Use

China's 2017 National Security Law requires every company operating in China to cooperate with national intelligence and defense if the state asks. For Huawei, whose network infrastructure is deployed in more than 170 countries, this legal provision creates a permanently half-open door. The Ascend systems running ByteDance's LLMs and Chinese universities are structurally the same as those which, according to several American think tank reports, are deployed in People's Liberation Army surveillance systems. The boundary between civilian AI and military AI is, at Huawei, the most porous of the Western democracies.

Recent reports from the Central Intelligence Agency (CIA) and the U.S. Congressional Research Service document how Ascend 910B and 910C architectures are used in autonomous targeting systems examining satellite imagery. The computing power needed for military large vision models is comparable to that of commercial LLMs. This means every Ascend performance gain in the commercial space is a proportional gain in military applications. Washington inadvertently created, in trying to stifle a commercial champion, a Chinese military AI champion.

Geographic Diversification as a Shield Against Future Sanctions

One of the lessons Huawei drew from the 2019–2022 sanctions is the need to diversify its markets to reduce dependence on any single jurisdiction. Its 2024–2028 strategy targets Sub-Saharan Africa, South America, and Southeast Asia for the deployment of its 5G infrastructure and data centers. In these regions, Huawei offers unbeatable prices, decentralized technical support, and reduced-rate financing through Chinese state banks. Europe and the United States cannot compete on price alone.

This geographic expansion is not merely commercial — it is geopolitical. Every 5G Huawei network deployed in an African or South American country is infrastructure that falls outside NATO security standards, captures data in developing economies, and creates lasting technological dependencies. The Digital Silk Road is not a marketing slogan — it is an infrastructure strategy that transforms emerging markets into Chinese technological spheres of influence.

Conclusion: A Champion Born in Constraint

What Huawei's Story Teaches About Resilience

Huawei's trajectory between 2019 and 2026 is one of the most remarkable stories of industrial resilience in recent history. A company cut off from its most critical suppliers, targeted by the sanctions of the world's most powerful government, declared doomed by virtually every analyst — and seven years later it is the indispensable provider for a national $295 billion plan. This resilience is not solely due to Beijing's industrial policy. It is due to real company decisions: massive R&D investment, the pivot to B2B infrastructure when smartphones failed, the development of the CANN software stack when CUDA became inaccessible.

These decisions were made under constraint — but they were made. And their results are real. China today has an advanced AI chip manufacturing capability that did not exist in 2019. It owes this partly to its own engineers. It also owes it, paradoxically, to American strategic decisions that closed off all other options. Huawei is the champion forged against its will. And the technological world of the future will have to reckon with it.

What This Means for the Coming Decades

The NDRC's $295 billion plan is not merely an investment in data centers. It is the institutionalization of a sovereign technological ecosystem of which Huawei is the core. If this plan is executed — even partially — China will have by 2030 a national AI infrastructure powered 80% by its own chips, operated by its own state-owned companies, developed on its own software frameworks. This is a structural transformation that goes far beyond a chip market Nvidia lost. It is the emergence of a second global technology pole — inaccessible, incompatible, and increasingly competitive with the Western ecosystem. Huawei is its symbol. But it is the entire system that must be watched.

By Maxime Marquette, columnist

Columnist's transparency note

My Sources and Limits

This profile rests on verified open sources: the Bloomberg report of June 9, 2026, analyses from Tom's Hardware, Ashare Insights, and abhs.in on the NDRC plan, Jensen Huang's public statements, and Huawei production data published in industry reports. I am not a semiconductor specialist. I do not claim access to Huawei's internal technical specifications or to the classified assessments of allied governments on the security risks associated with the company.

What I Don't Know

I do not know Huawei's real ownership structure or the extent of its links with the Chinese government beyond what is public. The Ascend chip performance figures I cited are based on benchmarks published by Chinese institutions — they may be more optimistic than reality. The NDRC plan of $295 billion is described as "under discussion" — it has not yet been formally adopted, and its implementation modalities remain uncertain.

Sources

Primary sources

Bloomberg — China: $295 billion plan for national AI infrastructure, 80% domestic — June 9, 2026

ABHS — Analysis of NDRC plan: Huawei, Biren, Moore Threads, Nvidia excluded — June 23, 2026

Tom's Hardware — China AI Plan $295B, 80% domestic chips, Nvidia and AMD excluded — June 10, 2026

Secondary sources

Ashare Insights — Analysis of Chinese AI CapEx shock: Sino-US structural decoupling — June 18, 2026

Bloomberg Newsletter — China AI Plan: challenge to American leadership, compute "smart grid" — June 12, 2026

SimmTester — Technical analysis: Ascend 950PR vs H20, domestic HBM, supply bottleneck — June 11, 2026