China's AI Capital Expenditure Set to Hit $98 Billion by 2025 Amid Intensifying US Rivalry

China is projected to dramatically increase its artificial intelligence (AI) capital expenditure to an estimated $98 billion by 2025, marking a substantial 48% rise from 2024 levels. This significant investment underscores Beijing's aggressive push to narrow the technological gap with the United States in global compute capacity, a critical domain for advanced AI development and deployment. The surge in spending comes as both nations intensify their rivalry for supremacy in the rapidly evolving AI landscape, particularly under the shadow of ongoing U.S. export controls.

As stated by Rohan Paul on social media, "Some recent reporting says 2025 AI capital expenditure in China at up to $98B, up 48% from 2024, with about $56B from government programs and about $24B from major internet firms." This comprehensive state-led and corporate-backed funding, confirmed by Bank of America analysis, aims to bolster China's AI infrastructure, emphasizing the construction of new data centers and supporting energy facilities. This approach contrasts with the U.S. focus on IT hardware like semiconductors.

The United States currently holds a commanding lead in global AI compute capacity, controlling the absolute majority of known AI training compute on the planet. This dominance stems from its leadership in advanced AI chip design, with companies like Nvidia, Google, AMD, and Intel controlling over 90% of the market, alongside extensive data center infrastructure and major cloud service providers. U.S. export controls on advanced semiconductors and related equipment are strategically designed to limit China's access to cutting-edge technology, thereby aiming to slow Beijing's aspirations for AI leadership.

In response to these restrictions, Chinese firms are increasingly relying on domestic accelerators, especially for AI inferencing. Huawei, a key domestic player, plans mass shipments of its Ascend 910C in 2025, an architectural evolution integrating two older Ascend 910B chips. However, as noted by Rohan Paul, "Independent analysts compare Nvidia’s export-grade H20 with Huawei’s Ascend 910B and find the Nvidia part still holds advantages in memory capacity and bandwidth, which matter hugely for training large models." Paul further highlighted, "Also there's this huge issue on software maturity gaps around Huawei’s stack, that reduce effective throughput, even when nominal specs look close to older Nvidia parts like A100."

Despite China's aggressive capital expenditure and domestic innovation efforts, translating this investment into truly competitive training compute capacity remains a complex and time-consuming endeavor. The challenges are exacerbated by U.S. export controls and persistent issues such as low chip yield rates for domestic production. This intense technological rivalry highlights a strategic divergence, with China prioritizing self-reliance through infrastructure development and homegrown chip solutions, while the U.S. seeks to maintain its lead through advanced hardware and strategic restrictions. The outcome of this competition will significantly shape the future of global AI innovation and its geopolitical landscape.