Huawei has announced the open-sourcing of two AI models from its Pangu series, along with certain model reasoning technology.

This announcement, made on June 30, 2025, is part of the company’s efforts to enhance its AI strategy amid US restrictions on advanced chip exports to China.

This initiative is similar to actions taken by other Chinese tech firms, such as Baidu, which open-sourced its Ernie large language model series on the same day.

Open-sourcing enables developers and businesses to test and adapt the models for their specific needs.

Huawei said that this initiative is part of its “Ascend ecosystem strategy,” focused on its Ascend AI chips.

These chips are considered a primary competitor to Nvidia’s products, which face export controls.

Huawei has invited global developers, researchers, and partners to download and provide feedback on its open-source products.

🔗 Source: CNBC

🧠 Food for thought

1️⃣ Hardware-software integration emerges as a strategic advantage in the AI race

Huawei’s open-sourcing of its Pangu AI models represents a strategic move that mirrors successful tech giants’ integrated hardware-software approach.

By pairing its open-source models with its Ascend AI chips, Huawei is creating an ecosystem similar to Google’s strategy of developing both AI chips and models like the open-source Gemma 1.

This vertical integration allows Huawei to optimize its AI solutions for specific sectors like government, finance, and manufacturing, differentiating itself from competitors like Baidu that focus more broadly on general-purpose AI capabilities 1.

The strategy addresses a critical business challenge: while U.S. restrictions limit Huawei’s access to advanced chip technologies, open-sourcing creates alternative pathways to market expansion by incentivizing developers to build on Huawei’s ecosystem 2.

Importantly, Huawei’s Ascend 910C chip has already generated over $2 billion in pre-orders, demonstrating strong market demand for domestic AI hardware alternatives in China 3.

This integration of open models with proprietary hardware has proven effective historically, as seen with Google’s TensorFlow (2015) and PyTorch (2016), which significantly accelerated AI adoption by creating optimized pathways between software frameworks and hardware acceleration 4.