In a significant move that could reshape the landscape of China's artificial intelligence infrastructure, internet giant ByteDance is reportedly preparing a massive procurement of domestically produced AI chips. This strategic pivot comes amid ongoing global supply chain uncertainties and a surge in demand for computing power to fuel next-generation AI services.
ByteDance's Reported $55 Billion Ascend Chip Procurement
According to industry reports citing informed sources, ByteDance is planning to purchase over CNY 400 billion (approximately USD 55 billion) worth of Ascend AI chips from Huawei in 2026. This represents a dramatic increase from the company's current procurement levels, which are said to be nearly zero for this year. The initial delivery phase is expected to involve chips valued in the hundreds of billions of yuan, signaling a rapid and large-scale deployment. This move is seen as a direct response to the supply gap created earlier in 2025 when NVIDIA's H20 chips were cut off from the Chinese market, coupled with explosive growth in tokens processed by ByteDance's cloud service, Volcano Engine, and its AI application, Doubao.
Reported ByteDance AI Chip Procurement Plans:
- Huawei Ascend (2026): Potential order value > CNY 400 billion (approx. USD 55 billion).
- NVIDIA H200 (Conditional): Potential initial order of 20,000 units (approx. USD 40,000 per unit).
The Broader Push for Domestic AI Compute
The potential deal with Huawei is not an isolated event but part of a broader strategic consensus among Chinese tech firms to accelerate the adoption and substitution of homegrown computing power. Industry analysis suggests the market is evolving into a "Huawei Ascend + Multiple Strong Players" structure. While Huawei's Ascend series may capture a significant portion of ByteDance's order, other domestic manufacturers like Hygon and Cambricon are also positioned to secure shares of this sizable market. Diversifying supply sources is considered a prudent strategy to mitigate risk in a critical sector.
ByteDance's Ambitious AI Investment Roadmap
This chip procurement aligns with ByteDance's previously reported ambitious capital expenditure plans for AI development. Reports from late 2025 indicated the company preliminarily budgeted CNY 1.6 trillion for 2026, up from CNY 1.5 trillion in 2025. A substantial portion of this, roughly CNY 850 billion, was earmarked specifically for acquiring AI processors to train models and power applications. The scale of this investment underscores the company's commitment to becoming a leader in the global AI race and the immense computational resources required to do so.
ByteDance's Broader AI Investment Context:
- 2026 Preliminary Capital Expenditure: CNY 1.6 trillion (up from CNY 1.5 trillion in 2025).
- Portion Allocated for AI Processors: CNY 850 billion.
The Complicated Global Supply Context
The shift toward domestic suppliers exists within a complex global trade environment. In December 2025, the U.S. administration announced it would allow NVIDIA to sell its H200 processors to "approved customers" in China. Reports suggest that if such sales are approved, ByteDance and other Chinese tech firms have expressed interest in substantial orders, with ByteDance itself considering an initial purchase of 20,000 H200 chips. This indicates a pragmatic, multi-sourcing strategy where companies are preparing to leverage both newly available foreign high-end chips and rapidly scaling domestic alternatives to ensure their AI ambitions are not hindered.
Implications for the Domestic Semiconductor Ecosystem
A procurement commitment of this magnitude from a leading cloud and AI service provider like ByteDance would provide a massive boost to Huawei's Ascend ecosystem and validate the progress of China's domestic AI chip industry. It serves as a critical "anchor order" that can drive volume production, lower costs, and foster software optimization. Furthermore, it tests the capacity of the domestic supply chain, as seen with Hygon's Deep Compute 3 chip, which entered mass production ramp-up in August 2025 and has reportedly passed full-scenario testing at ByteDance and Alibaba Cloud.
Domestic AI Chip Landscape Note:
- Hygon's Deep Compute 3: Entered mass production ramp-up in August 2025. Initial monthly capacity: 10,000 units, projected to rise to 30,000 units per month in Q4 2025. Reported to have passed full-scenario testing at ByteDance and Alibaba Cloud.
Navigating a Future of Dual Sourcing
The emerging picture is one of strategic hedging. Chinese AI companies are building plans that no longer rely on a single geographic source for critical technology. While pursuing advanced options like NVIDIA's H200 where possible, the parallel and accelerated build-out of a domestic AI compute backbone, led by deals such as the potential ByteDance-Huawei agreement, is becoming a central pillar of technological sovereignty and business continuity. The success of this dual-track approach will be a key determinant of China's AI competitiveness in the coming years.