Beijing targets 130,000 petaflops by end-2026 in AI compute push
Synopsis
Beijing plans to more than double its H1 compute additions in H2 2026, targeting 130,000 petaflops total — while simultaneously building a RISC-V-based open-source AI stack to reduce dependence on Arm and Intel architectures.
Key Takeaways
Beijing aims to reach 130,000 petaflops of intelligent computing capacity by end-2026 , according to the Beijing Municipal Bureau of Economy and Information Technology .
The city added 22,000 petaflops in the first half of 2026 , bringing current capacity to 82,000 petaflops ; a further 50,000 petaflops are targeted for H2 2026 .
Beijing is promoting a 'token economy' framework to commercialise AI outputs and integrate AI into traditional industries.
A new 'RISC-V + AI OS' open-source ecosystem is being developed as a full-stack alternative to Arm and Intel x86 architectures.
The initiative positions Beijing as a national 'AI highland' , reflecting China 's broader strategy to achieve technology self-sufficiency amid US export controls.
Beijing is aggressively scaling its artificial intelligence infrastructure, targeting a total intelligent computing capacity of 130,000 petaflops by end-2026 as China accelerates its technology race with the United States. The Beijing Municipal Bureau of Economy and Information Technology announced the target at a press briefing on Tuesday, 22 July 2026, framing the push as central to establishing the capital as a national 'AI highland'.
The numbers behind the compute surge
The city's intelligent computing capacity stood at 82,000 petaflops as of mid-2026, following a 22,000-petaflop addition in the first half of the year, according to the local technology agency. Reaching the year-end goal requires adding another 50,000 petaflops in the second half of 2026 alone — a pace that significantly outstrips the first-half build-out. One petaflop equals 1 trillion calculations per second, underscoring the sheer scale of compute being deployed.
The 'token economy' and what it means
Alongside raw compute expansion, Beijing unveiled new policies to develop what officials are calling a 'token economy' — a framework designed to monetise AI-generated outputs and integrate artificial intelligence into traditional industries to drive new economic growth. The initiative reflects a broader national strategy to move beyond AI research and into large-scale industrial deployment, with the capital serving as a test bed for the rest of China.
The RISC-V and open-source OS bet
Beijing is also backing an open-source 'RISC-V + AI OS' ecosystem, pairing AI-focused operating systems with RISC-V — an open-standard chip architecture positioned as a strategic alternative to proprietary Western architectures such as Arm and Intel's x86. The goal is to build a full-stack framework spanning chips through to applications, reducing dependence on foreign intellectual property at every layer of the technology stack. This positions Beijing's compute ambitions not just as a capacity play, but as a sovereignty play.
Why it matters
The compute build-out comes as US export controls continue to restrict China's access to advanced semiconductors, making domestic alternatives and raw scale increasingly critical levers. By investing in open-standard architectures and proprietary AI operating systems, Beijing is attempting to construct an ecosystem that can function independently of Western chip supply chains. Industry analysts have noted that compute capacity is now treated as a strategic national resource in China, on par with energy infrastructure.
What's next
The pace of the second-half compute addition will be a key indicator of whether Beijing's targets are achievable given ongoing hardware constraints. Broader adoption of the RISC-V + AI OS stack across Chinese enterprises will signal how quickly the open-source alternative can displace incumbent Western architectures. Observers will also watch whether the 'token economy' policies translate into measurable AI revenue for industries beyond the technology sector.
Point of View
Not merely a cost consideration. The 'token economy' framing is notable: it suggests Chinese policymakers are trying to solve the monetisation problem that has stalled AI ROI globally, not just the infrastructure problem. What mainstream coverage tends to underplay is that petaflop targets are easy to announce but hard to hit under export controls — the real test is whether domestic chip suppliers like Cambricon or Huawei's Ascend line can fill the gap. If Beijing meets its year-end number, it will mark a significant proof point that China's AI infrastructure can scale without access to Nvidia's highest-end hardware.
NationPress
22 Jul 2026
Frequently Asked Questions
What is Beijing's AI computing power target for 2026?
Beijing is targeting 130,000 petaflops of intelligent computing capacity by end-2026 , according to the Beijing Municipal Bureau of Economy and Information Technology . The city currently has 82,000 petaflops after adding 22,000 petaflops in the first half of the year.
What is China's 'token economy' in AI?
The 'token economy' is a policy framework announced by Beijing officials designed to commercialise AI-generated outputs and integrate artificial intelligence into traditional industries to create new economic growth. It represents a shift from AI development to large-scale AI deployment and monetisation.
Why is Beijing investing in RISC-V and open-source AI operating systems?
Beijing is backing a 'RISC-V + AI OS' ecosystem as a strategic alternative to Western chip architectures from Arm and Intel , aiming to reduce dependence on foreign intellectual property. The initiative is part of a broader push for technology self-sufficiency amid ongoing US export controls on advanced semiconductors.
How does Beijing's AI push fit into the China-US tech race?
China is accelerating AI adoption across key industries as part of its technology competition with the United States , with Beijing positioned as a national 'AI highland' . The compute expansion and open-source chip strategy are designed to sustain AI infrastructure growth even as US export restrictions limit access to top-tier foreign semiconductors.
What is a petaflop and why does it matter for AI?
A petaflop equals 1 trillion calculations per second and is a standard measure of computing performance critical for training and running large AI models. Beijing 's target of 130,000 petaflops by end-2026 reflects the enormous computational demands of modern AI workloads at an industrial scale.
