Beijing Plans to Add 50,000 Petaflops of Computing Power in H2, Total to Exceed 130,000P by Year-End
Claire Weston
Beijing plans to add 50,000 P of AI computing power in the second half of 2026, lifting the city's total past 130,000 P — more than double the first-half pace, and the clearest signal yet of its bid to become China's AI capital.
What does 50,000 P actually mean?
P stands for petaflops — one quadrillion floating-point operations per second, the standard unit for measuring AI training capacity.
Beijing added 22,000 P in H1, bringing its current total to 82,000 P. The H2 target of 50,000 P is more than double the H1 build-out.
This means → Beijing is shifting from steady expansion to sprint mode in compute infrastructure.
Why the sudden acceleration?
The backdrop is an intensifying U.S.–China tech rivalry. China needs to move AI from lab-stage research to industrial deployment as fast as possible.
Beijing has positioned itself as the country's "AI high ground" — and compute is AI's most fundamental resource. Without it, large-model training and applications cannot run.
In plain terms = compute is to AI what electricity is to factories — Beijing is racing to build the "power plants of the AI era."
RISC-V plus an AI operating system — what is the play?
Alongside the compute push, Beijing is nurturing an open-source "RISC-V + AI operating system" ecosystem. RISC-V is an open-standard chip instruction-set architecture — anyone can use and modify it freely, with no single company in control.
This means → Beijing wants to build a full-stack domestic technology path from chip architecture to applications, reducing dependence on Arm and Intel x86.
This reflects a deeper ambition: the compute expansion is not just about "buying more servers" — it is about controlling the underlying architecture.
Can this target actually be met?
The H2 increment of 50,000 P is 2.3 times the H1 figure of 22,000 P — a significant jump in execution difficulty.
Whether the RISC-V ecosystem can achieve real industrial synergy remains an open question.
In plain terms = the target is clear, but the proof will come at year-end.
Content is for reference only, not financial advice.