China's AI Chip Renaissance: The Quarter That Changed Everything

*Cambricon's profit surge of 185%. Hygon crossing ¥40 billion in quarterly revenue. Moore Threads flipping from red to black. And a single figure — 41% — that signals the end of Nvidia's monopoly in China's data centers. The Q1 2026 earnings season wasn't just good. It was historic.*

*China's domestic AI chip industry has reached an inflection point — and the numbers prove it.*

Executive Summary: The Quarter Nobody Expected

For years, China's AI chip industry was defined by what it couldn't do. Couldn't access the most advanced manufacturing. Couldn't match Nvidia's software ecosystem. Couldn't turn a profit. The narrative was clear: domestic silicon was a political necessity, not a commercial winner.

The Q1 2026 earnings reports destroyed that narrative.

In a single quarter, China's leading AI chip companies collectively demonstrated that domestic silicon has crossed the threshold from "promising prototype" to "profit-generating product." The financial metrics are startling:

*Sources: Company Q1 2026 filings, exchange data as of April 30, 2026*

Three companies turned a profit. Two saw revenue growth exceed 150%. And for the first time since the AI boom began, a domestic AI chip maker — Cambricon — posted quarterly operating cash flow of ¥834 million, compared to negative ¥1.4 billion in Q1 2025.

This isn't subsidy-driven growth. These are companies selling real products to real customers — cloud providers, banks, telecom operators, automakers — who have a choice and are increasingly choosing domestic.

Chapter 1: Cambricon — From China's "Nvidia Hope" to a Real Business

The ¥10 Billion Profit Nobody Predicted

When Cambricon went public on Shanghai's STAR Market in 2020, it was valued at ¥25.7 billion — a pure-play AI chip bet in a market hungry for semiconductor independence. For the next five years, the company burned cash, posted losses, and watched its stock price yo-yo on government procurement rumors rather than product traction.

Q1 2026 changed the story completely.

*Source: Cambricon Q1 2026 filing*

The cash flow figure deserves particular attention. Positive operating cash flow means Cambricon's customers are paying on time, in volume, and sustainably. The company noted that its products have passed "rigorous customer environment validation" across cloud computing, energy, education, finance, telecom, medical, and internet sectors.

The Product Behind the Numbers

Cambricon's MLU (MLU: Machine Learning Unit) series — particularly the MLU590 — has become the chip of choice for inference workloads in Chinese data centers. The company claims its latest accelerator approaches 70% of Nvidia A100 performance in FP16 at roughly 60% of the price. More importantly, its software stack has matured to the point where developers can migrate models from CUDA with minimal friction.

The MLU690, the next-generation accelerator, is expected to launch in Q3 2026 and will reportedly target training workloads — a segment currently dominated by Nvidia and Huawei Ascend.

Technical Note: Cambricon's software ecosystem, Cambricon Neuware, has expanded to support over 400 AI models including DeepSeek, Qwen, and LLaMA derivatives. The company's strategy mirrors Nvidia's playbook: hardware performance matters, but software lock-in wins markets.

Chapter 2: Hygon Information — The Quiet Giant With a Dual Advantage

CPU + DCU: A Strategy Nobody Else Can Copy

While Cambricon grabbed headlines with its profit surge, Hygon Information posted the highest absolute revenue in Q1: ¥40.34 billion, up 68.1% year-over-year. But Hygon's story isn't just about AI chips — it's about a unique dual-product strategy that no other Chinese semiconductor company can replicate.

*CPU growth estimated from overall growth and DCU breakout; Source: Hygon Q1 2026 filing*

Hygon's DCU (Deep Computing Unit) series — the "Deep Calculation No. 3" — delivers 380 TFLOPS in FP16, approaching 70% of Nvidia A100 performance at about 60% of the cost. But the real moat isn't raw performance. It's ecosystem compatibility.

The X86 Advantage

Unlike Cambricon, Moore Threads, or Huawei Ascend — which all require developers to learn new architectures — Hygon's DCU is built on a GPGPU architecture that maintains CUDA-like compatibility. Models trained on Nvidia hardware can run on Hygon DCUs with minimal modification. In China's current environment, where every GPU cycle matters and migration costs are scrutinized, this is a decisive advantage.

Hygon's CPUs, meanwhile, maintain x86 instruction set compatibility, giving the company dominant market share in China's domestic server CPU market (over 70%) and deep penetration in government and financial sectors.

The synergy is powerful: customers buy Hygon servers for general compute, then add DCU accelerators for AI workloads — all within a single vendor relationship and unified software environment.

The DeepSeek Connection

Hygon's general manager, Sha Chaoqun, revealed a striking metric in the company's earnings call: the DCU "Deep Calculation No. 3" has completed compatibility certification with 365 mainstream large models, covering what the company claims is "99% of non-proprietary open-source models globally." DeepSeek, Qwen, and virtually every major Chinese LLM run natively on Hygon hardware.

Chapter 3: Moore Threads — The Startup That Finally Made Money

From "GPU Hopeful" to Profitable Reality

Moore Threads has always been the most ambitious of China's GPU startups. Founded in 2020 by a former Nvidia China executive, the company set out to build a general-purpose GPU architecture capable of everything from AI training to graphics rendering to scientific computing.

The ambition came with heavy costs. Moore Threads burned through funding rounds, delayed product launches, and watched competitors like Cambricon and Hygon grab early market share.

Q1 2026 was its vindication.

*Source: Moore Threads Q1 2026 preliminary disclosure*

The exact revenue and profit figures remain undisclosed pending full audit, but the company confirmed it crossed into profitability — a milestone that puts it in elite company among China's AI chip ventures.

Moore Threads' MTT S4000 series targets the training market, the most demanding and highest-margin segment of AI compute. If the company can maintain its trajectory, it will become the third pillar of China's AI chip triumvirate alongside Cambricon (inference) and Hygon (hybrid CPU+GPU).

Chapter 4: Kunlun Core — Baidu's Chip Bet Goes Public

From Internal Supplier to Independent Powerhouse

While Cambricon, Hygon, and Moore Threads represent the independent chip design ecosystem, Kunlun Core — Baidu's chip subsidiary — tells a different story: what happens when a tech giant spins out its silicon unit and lets it compete in the open market.

Kunlun Core's origins trace to 2018, when Baidu unveiled the "Kunlun 1" chip built on Samsung 14nm process with 260 TOPS of INT8 performance. In 2021, Baidu formally spun out Kunlun Core as an independent entity, with chief chip architect Ouyang Jian becoming CEO and the company raising funds at a ¥13 billion valuation.

The External Revenue Breakthrough

*Sources: Market estimates, industry supply chain reports*

The most significant shift isn't revenue growth — it's revenue composition. For the first time, over half of Kunlun Core's revenue comes from external customers, not Baidu. The company has secured orders from China Mobile, China Merchants Bank, China Southern Power Grid, Geely Auto, and Vivo, with deployment scales ranging from dozens of cards to tens of thousands.

In China Mobile's 2025-2026 AI server procurement — a contract worth billions — Kunlun Core reportedly captured 70% to 100% of the "CUDA-like ecosystem" category, ranking first across all segments in that bracket.

The Dual IPO Gambit

Kunlun Core has filed for listing on both Hong Kong's HKEX and Shanghai's STAR Market — a rare "A+H" dual-listing strategy that would give the company access to both international capital and domestic Chinese investors. The HKEX application was submitted in January 2026; STAR Market listing辅导 (pre-IPO guidance) commenced on May 7, 2026.

If successful, Kunlun Core would become the first major AI chip company with a dual listing, potentially raising the capital needed to compete with Nvidia's annual R&D budget of $12+ billion.

Chapter 5: The 41% Milestone — End of an Era

*The domestic share milestone signals a fundamental shift in China's data center architecture.*

When Domestic Silicon Became the Majority Choice

The financial results are impressive, but a single market-wide statistic tells the bigger story. According to IDC's latest data on China's cloud AI accelerator market, domestic chip vendors now hold approximately 41% market share — up from roughly 29% in 2024.

*Sources: IDC China Cloud AI Accelerator Market Tracker, industry analyst estimates*

The 41% figure is a geopolitical and commercial watershed. For the first time, Chinese data center operators — cloud providers, telecom companies, government agencies — are buying domestic chips not because they must, but because the total cost of ownership (hardware + software + support + supply security) has become competitive.

The Three Drivers

Three forces are converging to accelerate this shift:

1. Inference Demand Explosion

The AI market has shifted from training to inference. Chinese AI applications — Doubao (500M+ MAU), Kimi, Tencent Yuanbao — are now serving billions of inference requests daily. Inference workloads are less sensitive to peak FLOPS and more sensitive to cost-per-token and power efficiency — areas where domestic chips have closed the gap.

2. Edge Intelligence

AI is moving from centralized data centers to edge devices: smart cameras, autonomous vehicles, industrial sensors, mobile phones. These markets favor specialized, low-power AI accelerators — exactly the segment where Chinese chip designers have focused their efforts.

3. Supply Security

Even companies with access to Nvidia hardware are diversifying. The experience of 2022-2023, when sanctions disrupted supply chains overnight, taught Chinese tech giants that single-sourcing is a strategic vulnerability. ByteDance's ¥40 billion ($5.6B) Huawei Ascend deal is the headline example, but similar diversification is happening across every major cloud provider.

Chapter 6: The Market That's About to Explode

From ¥142.5 Billion to ¥1.3 Trillion

The revenue figures from individual companies are striking, but the market forecast is staggering. According to Frost & Sullivan, China's AI chip market is projected to grow from ¥142.5 billion in 2024 to ¥1.3 trillion by 2029 — a compound annual growth rate of 54%.

*Source: Frost & Sullivan China AI Chip Market Forecast, 2024-2029*

The edge AI and automotive segments are growing fastest, driven by China's massive deployment of smart city infrastructure (400+ cities with AI-enabled surveillance) and the world's largest electric vehicle market (China produces 60%+ of global EVs, all of which require AI chips for autonomous driving and cockpit intelligence).

2026: The Year of Commercial Scale

Industry insiders are calling 2026 the "first year of scaled commercial deployment" for domestic AI chips. The reasons:

- Product maturity: The third-generation chips from Cambricon, Hygon, and Moore Threads have crossed the usability threshold

- Software ecosystem: MindSpore (Huawei), Cambricon Neuware, and Hygon's CUDA-compatible stack have reached production-grade stability

- Customer validation: Major deployments at China Mobile, China Telecom, Alibaba Cloud, and Tencent Cloud have proven reliability at scale

- Economic case: Domestic chips now offer 30-50% lower total cost of ownership for inference workloads

Chapter 7: What This Means for Global AI

*China's AI chip ecosystem is evolving from a parallel alternative to a genuine competitor.*

A Parallel Ecosystem Becomes a Competing Ecosystem

For the past three years, Western analysts have described China's AI chip ecosystem as "parallel" — a separate world that coexists but doesn't compete with the Nvidia-dominated global market. The Q1 2026 earnings suggest that characterization is becoming obsolete.

China's domestic AI chip industry is now a competing ecosystem.

*Sources: Industry analysis, company disclosures*

The competitive implications are significant. As China's domestic chip market grows toward ¥1.3 trillion, it will fund R&D investments that close the remaining performance gaps. Cambricon's current R&D budget is a fraction of Nvidia's, but if the company maintains its profit trajectory and market valuation, that gap will narrow.

More importantly, the software ecosystem is maturing. MindSpore — Huawei's AI framework — has grown from a political alternative to a genuinely capable platform. Chinese developers increasingly build models natively for domestic hardware rather than porting from CUDA.

The Talent Question

The ultimate competitive factor isn't capital or market size — it's talent. China's universities graduate more electrical engineers and computer scientists than the rest of the world combined. As the domestic chip industry generates profits and prestige, it's attracting the country's best technical minds.

Baidu's newly formed Model Committee (BMC), announced May 15, 2026, is instructive: it's staffed by "young researchers with deep understanding of large models" — a signal that China's tech giants are reorganizing to compete in AI at the organizational level, not just the product level. The same organizational intensity is happening in chip design.

Social Media Reaction: What Chinese Tech Users Are Saying

The earnings reports triggered intense discussion on Weibo, Xiaohongshu, and Zhihu. Here are representative comments:

Weibo user @科技投资笔记

"寒武纪一季报净利润10亿，现金流转正，这意味着国产AI芯片真的开始赚钱了。不再是‘国家补贴’的故事，而是‘市场需求’的故事。"

*"Cambricon's ¥1B profit and positive cash flow mean domestic AI chips are actually making money now. No longer a 'state subsidy' story — it's a 'market demand' story."*

Zhihu user @芯片从业者

"海光信息的DCU营收14亿，同比增长120%，这才是真正的大杀器。CUDA兼容不是技术妥协，是商业核武器。"

*"Hygon's DCU revenue of ¥1.4B with 120% growth — that's the real weapon. CUDA compatibility isn't a technical compromise, it's a commercial nuclear bomb."*

Xiaohongshu user @金融打工喵

"昆仑芯外销占比过半，这才是我最关注的。百度自己养的芯片公司，终于能独立走路了。IPO之后估计估值要翻倍。"

*"Kunlun Core's external sales exceeding 50% — that's what I care about most. Baidu's pet chip company can finally walk on its own. Post-IPO valuation will probably double."*

Weibo user @半导体老炮

"41%国产份额意味着每卖两块AI加速卡，就有一块是国产的。这放在三年前谁敢信？"

*"41% domestic share means for every two AI accelerators sold, one is domestic. Who would have believed this three years ago?"*

雪球用户 @价值投机2025

"摩尔线程居然盈利了！我去年还说这家公司活不过2026，打脸了。GPU四小龙里，它可能是最被低估的。"

*"Moore Threads is actually profitable! I said last year this company wouldn't survive 2026. I was wrong. Among the 'GPU Four Dragons,' it may be the most underrated."*

Twitter/X user @ChinaTechInsight (English)

"The Q1 numbers from Cambricon and Hygon aren't just earnings beats — they're proof that China's chip sanctions are accelerating domestic substitution faster than anyone in DC predicted. The 41% market share figure should terrify semiconductor strategists."

*"The Q1 numbers from Cambricon and Hygon aren't just earnings beats — they're proof that China's chip sanctions are accelerating domestic substitution faster than anyone in DC predicted. The 41% market share figure should terrify semiconductor strategists."*

The Bottom Line

China's AI chip industry has reached an inflection point that changes the strategic calculus for every player in global AI. The Q1 2026 earnings weren't outliers — they were the confirmation of a multi-year trend that has now reached commercial escape velocity.

Three conclusions stand out:

First, domestic AI chips are no longer just a political project. Cambricon's ¥10.13 billion quarterly profit, generated from cash-paying customers across multiple industries, proves that these products compete on economic merit.

Second, the 41% market share milestone in cloud AI accelerators is a tipping point. Above this threshold, network effects kick in: more customers attract more software developers, which attracts more customers. Nvidia's ecosystem advantage — the moat that has protected its dominance for a decade — is being challenged in the world's second-largest AI market.

Third, 2026 is the year the industry scales. With six major players (Cambricon, Hygon, Moore Threads, MetaX, Biren, Iluvatar) plus Huawei Ascend and Kunlun Core all producing competitive products, the Chinese market will have genuine choice. And genuine choice drives innovation faster than any subsidy.

For global AI watchers, the message is clear: the "parallel ecosystem" narrative is dead. China's AI chip industry has become a competing ecosystem — one that will shape prices, technology roadmaps, and strategic options for the entire global market.

The silicon wall that Washington tried to build? Q1 2026 proved it has a door — and Chinese chip companies just kicked it open.

*Cover image: Semiconductor wafer technology by Unsplash*

*Published on AI in China — ai-in-china.com | May 19, 2026*

Related Articles:

- The Great Silicon Wall: How China's AI Industry Is Defying US Sanctions

- China's AI Revolution: Education Reform, National Standards, and Ecosystem Breakthroughs

- DeepSeek V4: The Million-Token API Update That Signals China's AI Sovereignty Shift

- China's AI Revolution: 140 Trillion Tokens, DeepSeek V4 on Huawei, and the $700B Arms Race