Bloomberg reported on June 24 that ByteDance is seeking about $20 billion offshore, and the size of the loan tells you how far TikTok’s parent has moved from social media into the capital arms race around AI compute.
The useful way to read this story is not as a TikTok financing story. ByteDance has approached banks for a dollar-denominated loan of roughly $20 billion, with a three-year tenor and an option to extend it to five years, according to Bloomberg. No lead banks have been named. Pricing hasn’t been disclosed. That means the deal is still being assembled, but the intent is plain enough: ByteDance is trying to finance itself like a company that expects AI infrastructure, not short video, to decide its next decade.
That is a large shift, even for a company of ByteDance’s size. Bloomberg said the new facility would be its largest offshore loan, after the company raised $10.8 billion in offshore borrowing about nine months earlier. You don’t go back to banks for nearly twice that amount unless the spending plan has changed. For ByteDance, the pressure point is compute: chips, data centers, overseas capacity and the expensive work of keeping its AI models close enough to the frontier that users and developers don’t drift elsewhere.
The chip bill already gives you the shape of the problem. Reuters reported earlier that ByteDance was preparing to spend about 100 billion yuan, roughly $14 billion, on Nvidia AI chips in 2026, up from 85 billion yuan in 2025. Tom’s Hardware, citing that Reuters reporting, noted that the company has been looking at Nvidia H200 processors after Washington allowed limited sales of the chips to approved Chinese buyers. Nvidia’s Blackwell systems remain far harder for Chinese companies to access, so ByteDance is being pushed into a messy split: buy what it can from Nvidia, develop domestic or custom alternatives where possible, and rent or build capacity outside China when sanctions make the onshore route too narrow.
This is why the offshore structure matters. A dollar loan is not just cheaper money with a different label. It gives ByteDance access to international banks, lets it fund overseas suppliers and infrastructure more cleanly, and reduces its dependence on Chinese credit channels at the exact moment U.S. export controls are deciding what hardware Chinese AI companies can buy. If you’re trying to secure chips before the next rule change lands, financial flexibility is not a nice extra. It’s part of the product roadmap.
ByteDance also has a real consumer AI business to defend. Doubao, its AI assistant in China, has been one of the country’s most heavily used chatbot products since its 2023 launch, while ByteDance’s Seed and Seedance models have pushed the company into text, image and video generation. Reuters reported in February that Seedance 2.0 had gone viral in China, and the reaction outside China was less friendly: Hollywood studios and the Motion Picture Association raised copyright concerns after clips imitating famous actors and franchises spread online. That is the awkward bargain ByteDance is making. Scale wins attention quickly. It also attracts lawyers quickly.
There is a tempting comparison with SpaceX, but it needs to be handled carefully. SpaceX also went to debt markets this week, with MarketWatch reporting that it priced a bond sale at $25 billion after initially targeting $20 billion. Axios reported that the proceeds were tied mainly to refinancing a $20 billion bridge loan after SpaceX’s IPO, while investors were also weighing the company’s capital-heavy AI and space infrastructure plans. The comparison is not that ByteDance and SpaceX are the same kind of company. They plainly aren’t. The comparison is that two of the world’s most watched private-turned-capital-market stories are both showing the same thing: AI spending has become too large to fund comfortably from operating cash alone.
Frankly, that is the point founders should sit with. ByteDance has TikTok, Douyin, CapCut, Toutiao and a pile of other cash-producing products. It still wants debt. SpaceX raised enormous IPO proceeds and still tapped bond buyers. The companies that believe AI is existential are not treating compute as an annual software budget. They are treating it like factories, telecom networks and power plants, the kind of infrastructure you finance over years because the upfront cost is too heavy to absorb neatly.
ByteDance declined to comment on Bloomberg’s report, and no banks have publicly confirmed involvement. That dry fact matters. Until the loan is signed, this is still a financing plan, not cash in the door. Banks can change terms, regulators can complicate the structure, and geopolitics can make a clean financing look much less clean by the time documents are ready.
Still, the direction is hard to miss. ByteDance is trying to buy time, chips and optionality in one transaction. If the $20 billion loan closes, the next question won’t be whether TikTok’s parent is serious about AI. That question is already answered. The better question is whether even $20 billion is enough when every serious rival is reaching for the same chips, the same engineers and the same narrow window before the next export rule rewrites the plan again.
Also read: Nvidia's banned chips are selling for twice their price in China and Washington still thinks the export controls are working • Zhipu AI bets on a Shanghai listing to lock in its 2,000% post-IPO surge • Superhuman bets on trust as it buys AI detection startup GPTZero for its 19 million users and $30M ARR
