Everything About AI
NextFin News - ByteDance Ltd., the Beijing-based owner of TikTok, has drawn more than $30 billion in orders from lenders for its jumbo offshore bank loan, people familiar with the matter said, a level of demand that overshadows the roughly $20 billion facility the company was seeking and lays bare how fiercely banks are competing to finance the artificial-intelligence buildout.
The order book opened this month on a deal coordinated by Citigroup Inc., Goldman Sachs Group Inc. and JPMorgan Chase & Co. and swelled to more than $30 billion as global and Chinese lenders rushed to participate, according to the people, who asked not to be identified because the discussions are private. The final size of the facility has not been disclosed, and the terms remain subject to change until the loan is signed and syndicated to the wider market.
The demand is a striking reversal for a company that, less than two years ago, was raising debt under the shadow of a possible U.S. ban on its flagship app. It is also a referendum on two questions at once: whether banks believe the AI buildout will pay for itself, and whether they are willing to underprice risk to keep a seat at the table with the companies leading it.
The Deal, the Terms, and the Precedent
ByteDance first approached banks in June about what would be its largest offshore borrowing ever: a three-year facility with two one-year extension options. The structure echoes the company's September 2024 loan, which raised $10.8 billion from more than 20 lenders and stood, at the time, as the biggest dollar-denominated corporate facility in Asia outside Japan.
This time, the mandate is nearly double that. A $20 billion facility would rank among the largest syndicated loans ever arranged for a Chinese technology company, and the order book suggests the final size could exceed the initial target. Under the proposed tiering, banks can join as mandated lead arrangers with commitments of $500 million to $999 million, or as lead arrangers for smaller amounts, and a greenshoe option - an additional tranche that lets the borrower increase the size if demand holds - is under discussion.
Pricing is expected to come in tighter than the roughly 85 basis points over the Secured Overnight Financing Rate that ByteDance paid on the three-year portion of its 2024 facility, according to a loan banker briefed on the terms. That is a notable concession for a borrower raising tens of billions in a market where syndicated lending across Asia Pacific, excluding Japan, fell 20.2% in the first half of 2026 to $217.6 billion - the weakest half-year tally since 2013, according to the London Stock Exchange Group.
"We have strong appetite in China's TMT sector with the country's rapid advancements in artificial intelligence technologies," said a Hong Kong-based loan banker. "This is evident from financial reports from multiple Chinese tech giants like Alibaba and Tencent, with the booming AI sector highlighting their future business strategies."
The 2024 precedent supports the appetite. That facility was oversubscribed and grew from an initial $9.5 billion target to $10.8 billion at signing, even as U.S. scrutiny of TikTok intensified. It was coordinated by the same three banks and attracted 16 other lenders, including Bank of China, HSBC Holdings Plc, Industrial and Commercial Bank of China Ltd., Standard Chartered Plc and UBS Group AG. Proceeds from the 2024 deal were used in part to retire a $5 billion dual-tranche facility raised in December 2021. For the new borrowing, the specific use of proceeds has not been confirmed, and people familiar with the discussions have said details may still change.
Why the Money Is Needed: The AI Arms Race
Behind the financing is one of the steepest capital-spending ramps in the global technology industry. ByteDance is discussing expenditures of as much as $70 billion in 2026 on data centers and AI infrastructure, a figure that could climb to $100 billion in 2027 if economic and business conditions remain favorable, people familiar with the company's planning have said. For context, the company spent roughly $25 billion in 2025.
That puts ByteDance in the same league as the U.S. hyperscalers. Amazon.com Inc., Alphabet Inc., Microsoft Corp. and Meta Platforms Inc. have earmarked as much as $725 billion for capital projects this year, most of it flowing into AI data centers. ByteDance's preliminary 2026 capital-expenditure plan of around 160 billion yuan - about $22.7 billion - was already directed mostly at AI infrastructure before the spending ceiling was pushed higher.
The spending is not optional. ByteDance is building its own inference chips, known internally as SeedChip, procuring advanced processors where export controls allow, and has reached an agreement with Qualcomm Inc. for application-specific integrated circuits tailored to its data centers. It is also one of Microsoft's largest AI customers, spending more than $1 billion a year to access OpenAI's models through Azure in markets where OpenAI does not operate directly. Chips, data centers, and model access all require dollars - and a lot of them.
For a private company of ByteDance's scale, with no public listing to tap for equity, a syndicated bank loan is one of the few mechanisms available to raise tens of billions of dollars quickly. The company reported roughly $186 billion in revenue in 2025, up about 20% from the prior year, with overseas revenue rising nearly 50% and crossing 30% of the total for the first time. That cash generation is what makes the loan market comfortable: ByteDance is not a speculative AI start-up, but a profitable internet giant whose earnings are being redirected into infrastructure.
There is a cost to that redirection. Net profit in 2025 fell more than 70% from the prior year as AI investment accelerated, reversing earlier expectations of roughly $50 billion in full-year earnings. The company generated about $40 billion in net income through the first three quarters of 2025 before a heavy second-half investment surge. In other words, the loan is not funding growth on top of untouched profits; it is funding growth as profits compress.
The Market Is Starved for Quality Paper
To understand why more than $30 billion chased a $20 billion loan, look at the market ByteDance is borrowing into. Syndicated lending in Asia Pacific, excluding Japan, dropped to $217.6 billion in the first half of 2026, down 20.2% from a year earlier and the lowest half-year reading since 2013. China and Hong Kong together accounted for just 28.2% of regional volume. Australia was the largest loan market in the region at 26.1%, with volume down 4.6%.
In a thin market, a borrower with ByteDance's profile - profitable, global, and tied to the most compelling growth narrative in technology - becomes a scarce asset. Banks are not just earning fees; they are buying a relationship with one of the fastest-growing private companies in the world, and positioning themselves for a future that may include a public listing. The risk that once hung over the deal - a U.S. forced divestiture of TikTok's American operations - has eased after the company agreed to sell its U.S. entity to a consortium of American investors, removing a major overhang that would have made lenders hesitate two years ago.
The Second-Order Read: Cheap Credit Is the Signal
The first-order story is simple: banks want ByteDance's paper. The second-order story is more uncomfortable. Pricing that comes in tighter than 85 basis points over SOFR on a three-year loan to a Chinese internet company - in a volatile rate and regulatory environment - is not just a sign of demand. It is a sign that the loan market may be underpricing AI risk.
One loan banker put the tension plainly: "Pricing on the latest loan is likely to come inside the roughly 85bp over SOFR interest margin on the 2024 facility, raising questions over whether lenders can take large holds at that level. We already viewed the pricing on the 2024 deal as tight, and in today's more volatile rate and regulatory environment we would require a premium." In plainer terms: the bankers winning allocations may be doing so at a price that does not fully compensate them for the risk they are taking, because losing the allocation would cost them the relationship.
This is the mechanism by which AI capex gets mispriced across the financial system. It is not only equity investors bidding up chip stocks. It is also credit markets offering bargain rates to the companies building the infrastructure, on the assumption that the AI revenue will arrive in time to service the debt. When the discount rate on AI infrastructure falls in both the equity and credit markets at the same time, the entire buildout looks cheaper than it really is.
There is also a structural shift running underneath the cyclical scramble. The demand surge itself is cyclical - a flight to scarce, high-quality paper in a thin market, which will normalize as issuance picks up. But the funding channel is structural. For most of the past decade, a company of ByteDance's size would have funded a buildout of this scale through an IPO or a follow-on equity offering. The public-equity window for Chinese technology companies has been effectively closed for years, and ByteDance has shown little appetite for a listing even as secondary trades have valued the company at hundreds of billions of dollars. The result: the syndicated loan market has become the default funding channel for AI capex that would once have been equity-financed.
That matters because it shifts risk. Equity investors mark their positions to market daily; bank loan books carry exposure at amortized cost until a problem surfaces. A wave of AI infrastructure that is debt-financed rather than equity-financed stores more leverage inside the banking system, where it is less visible and slower to reprice.
ByteDance is not alone in this. SoftBank Group Corp. secured a $40 billion bridge loan in March to boost its investment in OpenAI, maturing in March 2027 with an initial margin of around 250 basis points over SOFR. CoreWeave Inc. has secured more than $30 billion of debt and equity capital this year alone. The pattern is consistent: AI infrastructure is being funded with debt at a pace that assumes the revenue will catch up.
The Counter-Thesis: This Is Rational, Not Reckless
The strongest case against the mispricing view is straightforward: ByteDance is not a leveraged gamble. It is a borrower with investment-grade-like fundamentals - roughly $186 billion of revenue, a dominant position in short-form video in China and globally, and a profit stream that, even after the AI-driven compression, remains in the tens of billions of dollars. A three-year loan with two one-year extension options is short-duration exposure. At a spread over a floating benchmark, lenders are paid as rates reset. The 2024 facility was oversubscribed and signed at tighter terms than ByteDance initially sought - evidence that the company can access deep pools of capital even under scrutiny, and that the syndicate of banks that lent to it remains intact.
From this angle, the tight pricing is not a bubble signal; it is the rational price of a high-quality borrower in a market where high-quality borrowers are scarce. The bankers demanding a premium are the ones who did not make the cut. The banks inside the deal have done the credit work and are comfortable that ByteDance's cash flow covers debt service many times over, even if AI returns disappoint.
There is weight to that argument. But it rests on one assumption: that ByteDance's revenue keeps growing while its capex discipline holds. If spending climbs toward $100 billion in 2027 while advertising revenue slows - or if the international revenue stream that now makes up more than 30% of the total comes under pressure - the credit profile changes faster than a three-year loan's pricing can adjust. Cyclical demand can mask a structural shift in risk, and this loan sits at the intersection of both.
What to Watch
The final size of the facility will be the first tell. If ByteDance keeps the loan near $20 billion despite more than $30 billion in orders, it signals discipline and a preference for pricing over capacity. If it upsizes toward $25 billion or $30 billion, it signals that the company is stockpiling cash for a capex push larger than even the $70 billion figure implies.
The syndicate composition matters too. A broad lender base with scaled-back allocations would confirm strong, broad demand. A deal held largely by the three arrangers and a small group of relationship banks would suggest the order book was softer than reported.
Most importantly, watch ByteDance's cash flow. The falsifying signal for the bullish credit read is specific and observable: if the company's 2026 capital expenditure approaches $70 billion while full-year revenue growth slows below 10%, or if free cash flow turns negative for two consecutive quarters, the assumption that this is a low-risk, investment-grade-like borrowing breaks down. At that point, the tight pricing on this loan - and any follow-on issuance - would look less like rational scarcity pricing and more like the early stage of a credit cycle that got ahead of itself.
The Bottom Line
More than $30 billion in orders says lenders believe in both the borrower and the buildout, and they are willing to pay for the privilege in the form of tighter spreads. The bill comes due in three years - with two one-year extensions if everyone agrees to kick it down the road.
The real story is not that ByteDance can borrow $20 billion. It is that the market is lending it at a price that assumes the AI future arrives on schedule. That is a bet on the technology, disguised as a bet on the borrower.
Explore more exclusive insights at nextfin.ai.
