Opinionaited
01 / 05
EXIT
Chip performance gap between Huawei Ascend and Nvidia systems
THE GAP

China is behind where AI advantage is most capital-intensive

U.S. export controls tightened again in January 2026, while Huawei's Ascend 910C remained far below Nvidia's leading hardware on real-world performance, inference efficiency, and silicon footprint.

Editor's Insight

This is not a story about China closing the per-chip gap. It is a story about what happens after that becomes implausible.

System-level architecture built from many weaker chips
THE WORKAROUND

CloudMatrix shows the systems strategy

Huawei's CloudMatrix 384 can approach GB200 NVL72 system-level competitiveness by linking far more chips together, trading elegance for aggregate throughput.

Editor's Insight

China did not solve scarcity. It changed the optimization function.

Open model ecosystem and software compatibility layer
THE STACK

Open weights and domestic clouds lower the hardware bar

Open-weight releases and Qwen's derivative sprawl matter because they expand what can run on good-enough inference hardware. The strategic object is the toolchain, not only the frontier model.

State-backed capital and procurement routing across the AI stack
THE STATE

Beijing is routing money, standards, and demand

Big Fund III, budget investment, treasury bonds, procurement approvals, and content rules make the state a scheduler of capital and market structure.

AI deployment spreading across manufacturing, robotics, science, and logistics
THE PAYOFF

Deployment breadth is becoming a form of power

China may matter in AI because it can embed AI broadly across an economy, even if it never wins cleanly at the frontier chip layer.

Editor's Insight

The commercial picture is ugly, but geopolitical capability does not require pretty margins.