ByteDance, the Beijing‑based owner of TikTok, is preparing a major AI spending plan as Chinese tech companies try to stay competitive with U.S. rivals. According to the Financial Times, ByteDance has drafted early budgets to spend RMB160 billion, about $23 billion, in capital expenditure in 2026, a step up from the RMB150 billion the company put into AI systems this year. Around half of the 2026 budget is set aside for advanced semiconductors used to build and run AI models and applications. ByteDance has also set aside RMB85 billion specifically for AI processors, even though access to Nvidia chips is still a struggle. ByteDance increases chip spending as Trump allows limited Nvidia sales Chinese tech companies have been restricted by U.S. export controls that block access to Nvidia’s most powerful chips, pushing companies like ByteDance and Alibaba to design models that cost less to run and need fewer computing resources. The restrictions are still in place, but there was a policy change this month when Donald Trump lifted a ban that allows Nvidia to sell its H200 processor to approved buyers in China, which admittedly is weaker than Nvidia’s top hardware, but it still matters. Trump described the policy as allowing sales to “approved customers in China” but the approval process still faces resistance from lawmakers in Washington, and President Xi Jinping has since said he doesn’t even want the chips anymore. If the H200 sales move forward, ByteDance plans to place a test order of 20,000 H200 chips, with each unit priced at roughly $20,000, according to the report . ByteDance also continues to spend billions of dollars leasing data centers overseas, so it can legally use Nvidia’s most advanced hardware to train AI models and support users outside China, but these payments are booked as operating costs, not capital expenditure, which means they are not included in the $23 billion budget. ByteDance grows consumer AI use as U.S. rivals fund AI with debt While ByteDance’s open‑source Doubao models trail Alibaba’s Qwen and DeepSeek on independent benchmarks, data from QuestMobile shows the Doubao chatbot has overtaken DeepSeek in monthly active users and downloads, making it the most widely used AI chatbot in the country. ByteDance is also pushing its Volcano Engine cloud platform to corporate clients, putting it in direct competition with Alibaba’s cloud business, which have driven high usage across ByteDance’s AI services, according to Goldman Sachs. Goldman analysts reported that in October, ByteDance recorded more than 30 trillion daily tokens, while Google logged 43 trillion tokens, a tiny margin in the grand scheme of things. Despite this growth, ByteDance’s spending remains far below that of U.S. tech giants. Microsoft, Alphabet, Amazon, and Meta together spent over $300 billion this year building data centers and power systems for AI models and products. Much of that U.S. expansion has been financed through borrowing. American companies sold $1.7 trillion of investment‑grade bonds in 2025, close to the $1.8 trillion raised in 2020 during the Covid crisis. Trade body Sifma tracked issuance through the end of November, showing a surge tied to AI infrastructure funding. Goldman Sachs estimates that AI‑related borrowing now represents about 30% of net investment‑grade issuance and is expected to rise again in 2026, even as concerns grow about the debt levels taken on by AI hyperscalers. If you're reading this, you’re already ahead. Stay there with our newsletter .
