Richard Yu, the executive who transformed Huawei from a telecommunications giant into a consumer electronics powerhouse, has been tasked with a new mission: securing the company’s dominance in the artificial intelligence era. Speaking at the Huawei Developer Conference (HDC 2026), Yu signaled a defiant return to the AI front, announcing the release of openPangu 2.0, a series of open-source large language models designed to operate within the constraints of China’s current hardware landscape.
Under Yu’s leadership, Huawei is pivoting toward a strategy of 'compute efficiency' over raw scale. The new openPangu 2.0 Pro, despite its 505-billion parameter architecture, utilizes an 'active parameter' strategy of 18 billion to optimize performance on Huawei’s own Ascend chips. This move is a calculated response to the global GPU shortage and U.S. export restrictions, focusing on high throughput and low latency rather than the trillion-parameter 'vanity metrics' pursued by American rivals.
Yu’s reassignment as the Director of the Product Investment Review Board (IRB) marks a critical consolidation of power. In this role, he oversees the strategic allocation of resources and budget for Huawei's most vital projects. Internal observers view this appointment as the start of a 'decisive AI battle,' where Yu is expected to replicate the success he achieved with the Mate smartphone series and the company’s burgeoning automotive division.
The strategic roadmap also includes a deep integration between AI and Huawei’s hardware ecosystem. By autumn, a 30-billion parameter version of the Pangu model is slated to run natively on Kirin mobile chips, promising a fivefold increase in throughput. This 'cloud-device synergy' aims to create a closed-loop ecosystem where Huawei’s software, silicon, and AI models are perfectly tuned for one another, bypassing the need for Western-designed hardware architecture.
However, the path to the 'number one' spot Yu covets is fraught with challenges. Huawei acknowledges that domestic demand for its Ascend chips has limited its own internal training capacity, preventing the development of the massive trillion-parameter models seen in the West. By open-sourcing not just model weights but the entire training and inference toolchain, Huawei is betting that a broad developer community will help refine its ecosystem and establish Pangu as the foundational architecture for Chinese industry.