Huawei's chip roadmap to 2029 aims to sidestep manufacturing limits with photonic computing and folded architecture

Huawei disclosed at its Huawei Connect conference in Shanghai a comprehensive hardware roadmap for AI chips and compute clusters through the end of the current decade, aiming to break bottlenecks caused by export restrictions on advanced Nvidia processors. Eric Shu, vice chairman of the board, presented the next generation of Ascend GPUs, announcing the launch of the Ascend 960 DT processor for model training in the first quarter of 2027, followed by the Ascend 960 PR processor for inference and data processing in the third quarter of the same year, with the next-generation Ascend 970 and Ascend 980 chips slated for 2028 and 2029 respectively.

These announcements rely on the pivotal role of the currently available Ascend 950 processor in bridging a wide computing gap for AI-developing organizations in China and abroad, but the company’s major leap focuses on a super-computing cluster architecture. Shu introduced the Atlas 960E SuperPod, describing it as the industry’s first system to use integrated photonics near the chip, replacing optical links with conventional electrical connections between circuits and modules. The system aggregates a massive number of GPUs in a unified operating environment to accelerate the training and inference of large models that can reach ten trillion parameters with higher operational efficiency.

The Chinese bet does not stop at photonic interconnects; it also extends to bypassing the restrictions on photolithography equipment denied to Beijing.Shu explained the details of applying the “Tao Gradient Law” principle that the company announced last May, which relies on a folded structural engineering that shortens signal-transfer time to overcome the physical limits described by Moore’s Law. He affirmed that this architecture will enable the development of precise chips with performance comparable to 1.4-nanometre technology by 2031 without needing advanced lithography equipment subject to the ban, by multiplying frequency range, memory capacity and interconnect speeds rather than relying solely on traditional geometric transistor scaling.

The company affirmed that its view of AI as a major technological revolution requires building an ecosystem that is not owned by a single entity but rests on the integration of application agents across multiple technology layers. Huawei’s core strategy is to provide processors and solid-state infrastructure while adhering to open-system and open-source principles, giving model builders greater flexibility in managing operational loads and developing their tools without relying on closed, proprietary environments.

This hardware shift outlines a new strategic reality for technology managers and digital-infrastructure leaders in the Gulf, Egypt and the Arab Levant, where massive data-center investments are accelerating amid heightened scrutiny of advanced-processor supply chains and energy-consumption constraints. The emergence of complete hardware alternatives that rely on photonic interconnects and target trillion-parameter models opens a practical path to diversify risk and avoid confining sovereign projects to a single hardware supplier, and it requires regional engineering teams to align their software stacks with Ascend processors and open environments in anticipation of changes in inference and training costs at local data centres over the next three years.