← All posts / Industry

ByteDance's $29.6 Billion Loan: The Biggest AI Debt Deal Yet Shows Banks Can't Stop Lending

TikTok's parent upsized its record offshore loan from $20B to $29.6B after banks piled in $30B+ of orders — 68bps over SOFR, the cheapest AI money on the market — to bankroll a $70B capex year.

ByteDance's $29.6 Billion Loan: The Biggest AI Debt Deal Yet Shows Banks Can't Stop Lending

In the summer of 2026, the question hanging over AI infrastructure was no longer whether banks would keep financing the buildout — it was how cheaply they would do it. ByteDance just answered that question in spectacular fashion. The Beijing-based parent of TikTok has closed a $29.6 billion syndicated loan, Asia’s second-largest dollar-denominated borrowing of the year, after demand from lenders overwhelmed the deal’s original $20 billion size, according to Bloomberg News.

The mechanics of the deal tell the real story. When ByteDance first sent feelers to banks in late June, it asked for about $20 billion — which would already have been its largest-ever offshore borrowing. By mid-August, orders had swelled past $30 billion, and the company upsized the facility to $29.6 billion. When a borrower raises 50% more money than it originally sought and still leaves orders on the table, that is not a loan negotiation. That is a stampede.

The price of the money is the headline

The most striking detail is not the size. It is the opening margin of 68 basis points over the Secured Overnight Financing Rate (SOFR) — one of the lowest borrowing costs on record for a corporate facility of this scale, and, as multiple outlets noted, a figure that repriced downward from the 85 basis points initially floated when the deal was being marketed. In other words, the more banks that crowded into the deal, the cheaper the money got. ByteDance will pay less to borrow thirty billion dollars than many investment-grade companies pay to borrow one.

The facility carries a three-year tenor that can be extended to as long as five years, with the margin adjusting if the tenor stretches. That structure matters: it gives ByteDance near-term flexibility on a war chest sized for immediate infrastructure deployment, while preserving the option to stretch repayment out across the useful life of the data centers it is building.

The deal was led by a syndicate including Citigroup, with roughly two dozen lenders ultimately joining a general syndication phase that ran through August, according to 9fin’s tracking of the process.

Where the money goes

ByteDance is not borrowing $29.6 billion to buy back stock. The proceeds are earmarked for the company’s AI infrastructure push, which has ballooned into one of the largest capital programs in the technology industry. Reports through the year have put ByteDance’s 2026 capital budget at as much as $70 billion for data centers and AI infrastructure — a figure the company reportedly raised by at least 25% from earlier internal plans as the AI race intensified.

The physical footprint is becoming concrete. Just hours before the loan news broke, the South China Morning Post reported that ByteDance plans to expand a massive AI data centre cluster in Ulanqab’s Jining district, in Inner Mongolia — a region that has quietly become China’s AI compute heartland. Nearly 100 data centers have opened or broken ground in Ulanqab since 2016, with Chinese firms committing some 12.5 gigawatts of planned capacity, drawn by cheap land, abundant wind and solar power, and cool, dry air that cuts cooling costs.

Inner Mongolia’s green-computing push gathered its own momentum in late August, when the region signed 186.46 billion yuan (about $27 billion) of AI computing project agreements, with Volcano Engine — ByteDance’s cloud platform — named among the leading participants alongside Cambricon and China Telecom. ByteDance’s loan is the private-sector fuel for exactly this kind of buildout: domestic clusters in Inner Mongolia and elsewhere, plus offshore capacity (including Malaysian clusters that give it access to top-tier Nvidia silicon) for frontier model training.

The strategic logic: Doubao, Seed, and the compute race

ByteDance’s spending makes sense only against the backdrop of its model roadmap. The company’s Doubao chatbot is one of China’s most-used AI assistants, and its Seed model family has been climbing leaderboards all year — Seed 2.1 Pro drew level with frontier Western models in June, the Seedance video models ship at a relentless cadence, and Volcano Engine sells the whole stack to enterprises as the default AI cloud in China. Every one of those businesses is compute-hungry, and none of them can be paused while rivals like Alibaba’s Qwen organization and DeepSeek push release after release.

There is also a geopolitically pragmatic flavor to the financing. A dollar-denominated offshore loan lets ByteDance fund its international AI infrastructure — the offshore clusters that keep its frontier training within reach of leading-edge chips — without touching its onshore renminbi operations. It is corporate structure as sanctions insurance: compute held offshore, in jurisdictions where export controls bite less, financed by global banks that plainly have no intention of being left out of the AI lending boom.

What it signals for the industry

Three signals are worth reading out of this deal.

First, debt has officially become an AI fuel. The first wave of AI infrastructure was funded from balance sheets and equity. The 2026 wave is funded from loan syndicates. ByteDance’s deal follows a pattern visible across the market — from Meta’s mega-facility financings to the private-credit-backed data center funds now raising tens of billions. When banks accept 68 basis points over SOFR for AI capex, they are telling you they view hyperscaler-grade AI debt as among the safest assets they can hold.

Second, the China AI buildout is not slowing — it is accelerating on cheap money. For every headline about export controls, there is a counter-headline about a record loan, a new gigawatt-scale cluster in Inner Mongolia, or a fresh model release. ByteDance borrowing $29.6 billion at near-treasury spreads is a stronger statement about China’s AI trajectory than any policy white paper.

Third, the gap between AI haves and have-nots is compounding. A company that can borrow at SOFR+68 and spend $70 billion a year on compute plays a different game than one that cannot. The loan effectively locks in ByteDance’s seat at the frontier-model table for the duration of the facility — and pushes the cost of staying competitive that much higher for everyone below it.

The backdrop of the deal also lands at a peculiar moment for ByteDance internationally, with TikTok’s regulatory position in the West still contested even as global banks queue up to lend to its parent. The banks, at least, have rendered their verdict: whatever political risk attaches to the name, the cash flows behind it — advertising, Douyin, Volcano Engine, Doubao — are as solid as any in global tech.

One loan does not decide the AI race. But a $29.6 billion facility, upsized on a tidal wave of bank orders, priced at 68 basis points over SOFR — that is the financial system voting, with unusual unanimity, on which way the future is trending.