ByteDance has just completed a $29.6 billion syndicated loan. It is the second largest transaction in dollars carried out in Asia this year. For the parent company of TikTok, the objective is clear: to finance its expansion in AI without drawing solely on its profits.
In brief
- Maturity of three years, extendable to five, coordinated by Citigroup and JPMorgan with around thirty banks.
- The initial target of 20 billion jumped under demand from lenders, with Chinese banks keeping more than 60% of the allocation.
- Stated purpose: general corporate use. Reality of the case: data centers and computing power.
Banks are banking on TikTok’s cash flow
The operation is reminiscent of that of SoftBank in March, when bridge financing of $40 billion was put in place around its stake in OpenAI. However, the comparison stops there.
SoftBank was borrowing against shares, awaiting longer-term refinancing. ByteDance, for its part, relies on an activity that already generates revenue. Banks are therefore betting on the cash flows of TikTok and Douyin, rather than on the value of a portfolio of securities, according to Bloomberg.
The assembly nevertheless remains particular. ByteDance is a private company and its accounts are not accessible to the public. Around thirty establishments will therefore commit billions on the basis of financial information to which users and regulators do not have access. Reuters reported on September 4 that Chinese banks had subscribed to more than 60% of the allocation, alongside U.S., European and Singaporean lenders.
Up to 70 billion capex per year, a figure to be confirmed
At ByteDance, general corporate use means computing power. ByteDance eyes up to $70 billion of AI infrastructure spending per year, a level that would put it alongside U.S. hyperscalers, according to Bloomberg sources. This figure clashes with the budget of 160 billion yuan (22.7 billion dollars) mentioned so far, without comment from the company: internal scenario, no decision made.
Money must also come out of China. ByteDance has become an anchor customer of data centers under construction in Southeast Asia, a commitment which secures their financing. The group trains a 10 trillion-parameter model in Inner Mongolia and pays more than $1 billion a year to Microsoft Azure for OpenAI’s models, while funding the hardware meant to make it self-sustaining.
American export controls close its access to the most advanced Nvidia chips, referring to custom Arm or RISC-V silicon and to Chinese suppliers. The same power therefore costs more.
A private debt that changes scale
The operation almost tripled that of 2024, 10.8 billion raised from around twenty lenders. It signals the financing of AI through bank debt, not just through profits: the AI debt of big tech has exceeded $350 billion, and the operation moves part of this credit out of the United States.
“ ByteDance competes with local hyperscalers in data centers and global ones in multimodal models », summarized Lian Jye Su, chief analyst at Omdia.
The shift also concerns crypto. While bitcoin miners reconvert their installations to AI, the rush on data centers is absorbing capital that yesterday went to hashrate. This competition for electricity and GPUs is playing out on balance sheets, while OpenAI is targeting the stock market at $852 billion, a deal that ByteDance refused.
The assembly remained to be finalized as of September 4, allocations currently being confirmed. Real test at maturity, in 2029 or beyond: if AI revenues are slow to take off, the debt will weigh, and an American legislative push against TikTok would attack the flow which bases the loan.
Maximize your Tremplin.io experience with our ‘Read to Earn’ program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
