ByteDance is set to crank its capital expenditure up to $70 billion this year to build out AI infrastructure, nearly tripling last year's roughly $25 billion spend. The new budget, estimated at 400 billion to 500 billion yuan (about $59 billion to $74 billion), will be supported by projected 2025 profits of roughly $50 billion. If business conditions hold, the company could increase spending to $100 billion next year.
ByteDance's expansion dwarfs its domestic rivals, who have been more cautious with AI infrastructure. Tencent's 2025 capital expenditure was 79.2 billion yuan (about $11 billion), and it has pledged to double AI investment this year to over 36 billion yuan. Alibaba spent 126 billion yuan (about $18 billion) in its fiscal year ending March, maintaining a rolling three-year target of over $50 billion. ByteDance's single-year plan already exceeds the combined spending of its Chinese competitors by several times.
Constrained by export controls on high-end chips, ByteDance leverages the low cost of local infrastructure. Data center construction and operation costs in China are far lower than in the US, allowing the company to buy equivalent capacity with less money. By comparison, the four US hyperscalers — Microsoft, Google, Amazon, and Meta — are spending a combined $725 billion this year, averaging $181 billion each. ByteDance is diversifying its compute resources, recently signing a deal with Qualcomm to purchase millions of custom ASIC chips to support its Doubao large language model (with over 300 million monthly active users) and the Seedance video and text model ecosystem.
The company is also streamlining its business lines to focus on AI R&D. Over the past year, ByteDance has divested non-core businesses like gaming, funneling capital into AI and social products. In the latest round of equity transactions, ByteDance was valued at roughly $550 billion. While the spending plan will be adjusted quarterly and final outlays depend on hardware and power availability, ByteDance's accelerated push into large models, commercialization, and hardware procurement signals its determination to dominate compute and agent capabilities.