ByteDance's $30 Billion Loan Shows the AI Boom Is Now a Debt Story

The most valuable technology company you cannot buy just borrowed roughly $30 billion. You can't own ByteDance — the company behind TikTok and China's Douyin has never gone public — so the natural reaction to the headline is to scroll past it. Don't. The terms of this loan are a quiet tell about who now finances the AI buildout, and that is a shift every investor in this theme should see, because it moves the risk from one set of balance sheets to another.

Here is what actually happened. ByteDance secured a $29.6 billion unsecured syndicated loan from nearly 30 banks, coordinated by Citigroup and JPMorgan, with Chinese institutions taking more than 60% of the facility and ICBC and HSBC reported among the lenders. A syndicated loan just means a group of banks each chips in a slice of one big loan. The "unsecured" part is the detail worth pausing on: no collateral, so the lenders are betting on ByteDance's creditworthiness alone. It is the second-largest dollar loan in Asia this year, behind only SoftBank's $40 billion bridge loan raised in March for OpenAI.

The tell: cheaper money on a bigger bet

Now the part that deserves a "get this." ByteDance asked the market for $20 billion and lenders lined up more than $30 billion — roughly 1.5 times oversubscribed — so it took the bigger number. And it got cheaper. The margin, the interest paid on top of the benchmark rate known as SOFR, came in at 68 basis points. That is below the 85 basis points it paid on its 2024 offshore loan, even though this facility is nearly three times larger.

Price is a crowding signal. Lenders accepted thinner pay, on a bigger amount, with no collateral, during the weakest first half for Asia's syndicated loan market in 16 years. They had plenty of other places to park the money. They chose an unsecured bet on an AI buildout instead. When the marginal buyer of a risk is a bank competing to lend into it, that is not the same thing as the market quietly digesting it.

Where the risk now sits

So far this reads like a green light, and on its face it is. You don't get 1.5 times oversubscription on a $29.6 billion unsecured note to a private company unless lenders genuinely expect to be paid back. Cheap credit is fuel. That is the bull case, stated fairly.

But look at what the money is for. The official use is general corporate purposes, and people close to the deal say the funds are mainly for AI projects outside China — data centers and infrastructure. This is not a speculative placeholder. ByteDance is said to be weighing capital spending as high as $70 billion this year, more than double last year's total, and its China AI budget alone was raised to roughly 200 billion yuan (about $30 billion), up 25% from an earlier forecast. It is already an offtaker — a buyer that commits to capacity — for data centers under construction in Southeast Asia. The adoption is observable; the loan is funding something real.

Here is the reversal. A floating-rate loan tied to SOFR has to be serviced regardless of whether AI revenue arrives on the planned schedule. ByteDance, being private, can quietly absorb a slow year — no public earnings to explain, no shareholders watching a stock. The banks that lent it $29.6 billion do not have that luxury, and more than 60% of the exposure sits on Chinese state lenders. That is the concentration mirage in the form of a balance sheet: one theme, one borrower, one funding base carrying the weight. The bigger the loan and the thinner the margin, the more the lenders are competing for the privilege of underwriting a single bet.

And the bet is not small. The five biggest U.S. cloud and AI providers have committed to somewhere around $660–690 billion in capital spending this year, nearly double last year. Add ByteDance, the Chinese hyperscalers, and OpenAI's backers, and the AI buildout is a multi-trillion-dollar project increasingly financed with borrowed money rather than reinvested profits.

What it means for money you can actually deploy

You cannot buy ByteDance, and there is no clean proxy that exists for you to own it. But this loan is the live plumbing reading for the AI trade you can make. It shows the marginal funder of AI risk is now the banking system, not the equity holder, and that the risk has pooled where the leverage is. Cheap, oversubscribed credit is a vote of confidence today; it is also evidence that everyone already got in.

The condition that would make this reading wrong: credit stops being this willing. If these margins start widening, if the next big AI syndication comes in undersubscribed, then the buildout's fuel gets more expensive precisely when the largest bills come due. That is the number to watch — the cost of AI capital, not another earnings multiple. It is the thing that decides whether this cycle ends in real ownership or in a quiet handoff from the companies to the banks.