China's Ministry of Industry and Information Technology (MIIT) has unveiled one of the most ambitious national computing targets in history: 9,800 exaflops (EFLOPS) of intelligent computing capacity by 2030. Announced as part of the country's 15th Five-Year Plan for the information and communications industry, the directive represents a seismic shift in how Beijing positions artificial intelligence — not as a commercial frontier, but as critical national infrastructure, equivalent in strategic importance to railways, power grids, or semiconductor fabs.
The scale is staggering. At the end of June 2026, China's intelligent computing capacity stood at just 2,185 EFLOPS — meaning the country must achieve roughly a 4.5x increase in just four years. To put that in perspective, the entire world's AI computing capacity in 2023 was estimated at approximately 10-15 EFLOPS of training-grade infrastructure. China is not just closing the gap; it is attempting to leapfrog by treating compute like a public utility.
What Does 9,800 EFLOPS Actually Mean?
EFLOPS measures exa-FLoating-point Operations Per Second — one quintillion (1018) calculations per second. When applied to intelligent computing, this specifically refers to AI training and inference workloads, not general-purpose CPU/GPU benchmarks. For comparison, a single NVIDIA H100 GPU delivers roughly 3.9 petaflops (PFLOPS) of FP8 AI training performance. Reaching 9,800 EFLOPS would require the equivalent of approximately 2.5 million H100-class GPUs working in concert — a cluster size that dwarfs even the largest US-based AI data centers today.
But the number alone misses the deeper story. This target is embedded in a broader investment framework: RMB 3.8 trillion (approximately USD 530 billion) in information infrastructure spending planned between 2026 and 2030. This is not incremental spending. It is a generational commitment comparable to China's high-speed rail buildout or its renewable energy transition — except this time, the infrastructure is computational, and the stakes are technological sovereignty.
The Hardware Backbone: Clusters, Chips, and the Sanctions Reality
The MIIT plan explicitly calls for the construction of computing clusters with 10,000 or more accelerator cards, including some with 100,000 or more cards. This is a direct acknowledgment that the future of AI belongs to massive distributed systems, not single machines. But it also raises an obvious question: where do all these accelerators come from, given that US export controls have restricted China's access to cutting-edge NVIDIA and AMD chips since 2023?
The answer lies in China's domestic semiconductor push — and it is already showing results. Huawei's Ascend 910B and 910C chips are being deployed at scale in data centers across the country. Cambricon, Biren, and Moore Threads are filling gaps in the market. While individual Chinese chips still lag behind NVIDIA's H100 in raw performance, the cluster-level strategy adopted by MIIT reflects a growing confidence that scale and software optimization can compensate for per-chip deficits. The 7.2Tbps near-package optical module showcased by Huawei last month is precisely the kind of infrastructure enabler this plan demands — high-bandwidth, low-latency interconnects that allow thousands of domestic chips to function as a single computational unit.
The plan also implies a significant expansion of China's existing 52 facilities that already house more than 10,000 accelerators each. These "East Data West Computing" nodes, straddling the vast geography from Shanghai's demand centers to Gansu and Inner Mongolia's cheaper energy, will form the physical backbone of the 9,800 EFLOPS target. The west-to-east data flow is not new — China has been building this infrastructure since 2022 — but the scale acceleration implied by the new target suggests these regions will see exponential growth in data center construction over the next three years.
Geopolitical Implications: The AI Nationalism Acceleration
China's 9,800 EFLOPS target does not exist in a vacuum. It arrives at a moment when AI nationalism — the framing of artificial intelligence capability as a matter of national security and sovereignty — has become the dominant paradigm across every major tech power. The United States has responded with export controls, CHIPS Act subsidies, and the establishment of the National AI Research Resource (NAIRR). The European Union has countered with the AI Act and the European Battery Alliance model applied to semiconductors. Now China is entering the same race with the most aggressive numerical target yet.
The implications extend far beyond compute metrics. A 9,800 EFLOPS ecosystem requires a corresponding ecosystem of Chinese AI models, applications, and talent. It creates demand for hundreds of thousands of AI engineers, data scientists, and infrastructure specialists — a labor market that China is already training through its expanding university programs in AI and computer science. It also creates a self-reinforcing loop: more compute enables better models, which attract more investment, which funds more compute.
For the rest of the world, the message is unambiguous: China is treating AI compute as a public good and a strategic right, not a commodity to be rationed through export controls. The 3.8 trillion yuan investment signal is a challenge to Western assumptions that technology containment can slow China's trajectory. Instead, Beijing appears to be betting that forced self-reliance will accelerate indigenous innovation — a hypothesis that the success of Huawei's chip development and the rapid deployment of domestic AI models like DeepSeek and Qwen has given the party cautious confidence.
What It Means for Global AI Development
The 9,800 EFLOPS target has immediate and long-term consequences for the global AI landscape. In the near term, it guarantees that Chinese AI companies will have access to dramatically more compute than they did even two years ago — enough to train increasingly capable models that were previously constrained by hardware shortages. The DeepSeek-V3 and R1 releases of early 2026 demonstrated that Chinese teams can achieve frontier-tier performance with significantly less compute than their American counterparts; multiplying available resources by four will only widen that gap in training velocity.
Longer-term, the infrastructure buildout will create a dual ecosystem: a Chinese AI stack powered by domestic chips and optimized for Chinese-language and China-specific applications, and a Western stack built on NVIDIA/AMD hardware and optimized for English and global markets. The two ecosystems will diverge in capabilities, safety approaches, and regulatory frameworks — mirroring the broader technological decoupling already visible in semiconductors, social media, and cloud infrastructure.
For emerging markets and the Global South, China's compute expansion presents both opportunity and risk. On one hand, Chinese AI models and infrastructure exports offer an alternative to Western-dominated tech stacks. On the other, the concentration of advanced compute in a handful of massive national projects may make it harder for smaller nations to access frontier AI capabilities on their own terms. The question is whether China will export its compute infrastructure as part of the Digital Silk Road — or keep it domestic.
🔥 Hot Takes
1. China is not playing catch-up — it is building a parallel universe. The 9,800 EFLOPS target is not designed to match the US; it is designed to make the US irrelevant to China's AI trajectory. This is digital sovereignty in its most literal form: a self-contained compute ecosystem that operates on different hardware, different models, different safety frameworks, and different geopolitical assumptions.
2. The real story is not the number — it is the investment multiplier. RMB 3.8 trillion over five years is about $530 billion in infrastructure spending. That is more than the entire AI capital expenditure budget of OpenAI, Anthropic, Google, and Meta combined for the next three years. China is not just building compute; it is building an economy around compute, from chip fabrication to data center construction to model training services.
3. Export controls backfired — they forced China to build what it cannot buy. The very sanctions designed to slow China's AI progress have accelerated domestic chip development, created a thriving market for Chinese accelerators, and pushed Chinese companies to optimize for less powerful hardware — which ironically may make their models more efficient and portable than Western alternatives that assume infinite GPU availability.
4. The 100,000-card cluster target is a stealth industrial policy. By specifying cluster sizes at the policy level, MIIT is effectively choosing winners and losers in China's semiconductor industry. Only companies that can deliver reliable, large-scale accelerator supplies — primarily Huawei Ascend, Cambricon, and state-backed foundries — will benefit from this demand signal. The rest will be left to compete in a shrinking market.
5. This is the infrastructure equivalent of the Soviet space program — but with real economic returns. Unlike the ISS or Apollo programs, China's compute buildout is directly tied to commercial AI applications: autonomous vehicles, manufacturing optimization, drug discovery, financial modeling, and surveillance. The 9,800 EFLOPS target is not just about prestige; it is about creating a domestic AI industry that can compete globally without relying on Western hardware or software.
The Bottom Line
China's 9,800 EFLOPS target is the clearest signal yet that the country's leadership views AI compute as strategic infrastructure on par with energy, transportation, and communications. The RMB 3.8 trillion investment commitment is not speculative — it is embedded in the 15th Five-Year Plan, which has historically been implemented with remarkable fidelity. The 4.5x growth requirement is ambitious but achievable given China's existing trajectory, its domestic chip industry's rapid progress, and the sheer mobilization capacity of the Chinese state.
For the rest of the world, the implication is sobering: the era of Western compute dominance is ending, and the pace of that transition may be faster than most analysts anticipate. China is not asking for a seat at the AI table — it is building its own table, its own chairs, and its own rules. The question for American and European policymakers is no longer how to maintain technological leadership, but how to compete in a world where the largest compute ecosystem on Earth operates entirely outside their regulatory and technological framework.
Source: TechNode | MIIT Official Document