ByteDance Profit Falls 70% Amid AI Investment Surge, Insider Reveals

Select Language:

In 2025, the net profit of ByteDance dropped by over 70% compared to the previous year, as the company boosted investments across the entire artificial intelligence ecosystem. This move highlights the increasing prioritization of AI as a key factor for sustaining long-term competitive advantage among leading Chinese tech firms, according to an industry insider.

The company significantly ramped up spending on computing resources, infrastructure, and research and development during the latter half of last year, which impacted its profitability, the source revealed.

While ByteDance has not officially released specific figures, it reported approximately $33 billion in net profit in 2024. Based on that, the company likely earned around $9 billion in net profit last year.

The variation in financial results among top technology companies underscores that investing in computing power is crucial to winning the race in AI development, the insider explained. The substantial decline in profit last year indicates that ByteDance is working to compensate for the ‘computing power gap’ caused by talent retention challenges, rising computational costs, and complexities linked to shifting technological pathways, through new initiatives poised for high growth.

According to the CEO at a company-wide meeting earlier this year, AI holds a bigger strategic opportunity for the industry than both personal computers and the internet. In the near term, ByteDance plans to advance its AI chatbot Doubao, its international version Dola, and its enterprise-focused Model-as-a-Service products.

The company’s revenue in China increased by 20% in 2025 from the previous year, while overseas revenue surged nearly 50%, the source mentioned. International earnings now make up a record 30% of total revenue, up from 25% in 2024, driven notably by TikTok’s expanding e-commerce segment.

Investments include proprietary development of Dayu AI liquid-cooled server cabinets supporting 64 to 128 GPUs each, a MegaScale training cluster capable of supporting thousands of GPU units for stable training of trillion-parameter Mixture-of-Experts models, diversified procurement strategies, and advances in optical chip technology.