Huawei: Data Transport Capacity Has Become AI's Biggest Bottleneck, Token Volume Up 6x in 6 Months

Huawei has made a significant declaration at a recent industry conference: data transport capacity (运力) has replaced single-chip process technology as the primary bottleneck in the AI computing ecosystem.

Citing its own data, Huawei reported that token call volume has surged 6x in just six months, exposing severe inadequacies in data movement capabilities within AI infrastructure. The company argued the industry should stop fixating on chip manufacturing process nodes and instead focus on overall system-level computing efficiency.

This statement reflects Huawei's strategic positioning in the AI infrastructure space. As China's leading AI chip and computing infrastructure provider, Huawei's "transport-first" view aligns closely with its Ascend AI computing architecture roadmap.

The 6x surge in token calls indicates that large language model applications are rapidly moving toward real-world deployment, but data transfer and scheduling efficiency cannot keep pace with the explosion in inference demand. This places new requirements on data center design, network architecture, and chip interconnect technology.

Huawei's stance also implicitly responds to industry anxiety around advanced process nodes, advocating that system-level optimization offers more practical value than pursuing ever-smaller transistor geometries.

Going forward, it will be worth watching whether the industry follows Huawei's lead in reallocating compute investment priorities, and whether Huawei introduces new products or architectures specifically targeting the data transport bottleneck.