16 SEP 2026 — ByteDance took about $120 billion in revenue in the first six months of this year, up roughly 30 per cent on the same period last year. Net profit fell to $20 billion.
Both numbers matter, but the second one is the story. A company growing revenue at 30 per cent while profit falls is spending the difference on something specific.
Where the money went
The decline is attributed to AI infrastructure. ByteDance has been reported as budgeting around 160 billion yuan, roughly $23 billion, for AI infrastructure this year, including the servers, data centres and accelerators that any company running models at this scale has to buy or rent.
That number sits against a first half in which the business itself did well. Growth came substantially from TikTok's international advertising and its e-commerce operation, the two lines that have been under the most political pressure and are, on these figures, the ones paying for the model build.
For scale: the company's 2025 revenue was about $200 billion, up 29 per cent. A first half of $120 billion is, by our arithmetic, about three fifths of that entire year.
The profit line moves the other way. A single-digit percentage decline on $20 billion is somewhere between a few hundred million and about two billion dollars of forgone profit in six months. Set against a reported infrastructure budget of roughly $23 billion for the year, the company is not spending its profits on AI. It is spending several times them, funded out of a business that throws off cash faster than the build consumes it.
Margin is the cleaner way to see it. Twenty billion dollars of profit on $120 billion of revenue is about a sixth of the top line, by our arithmetic, in a half-year when the company was buying accelerators at scale. That is not the margin of a business in trouble. It is the margin of one converting a very profitable advertising and commerce operation into infrastructure.
A single-digit fall is not a warning
The profit decline is described as a single-digit percentage. A single-digit fall is not a collapse, a distinction that headlines routinely lose.
A company that earns $20 billion in half a year while building out infrastructure is not under strain. It is making a choice about where the next decade's margin comes from, and the choice is visible in the accounts rather than in a strategy deck.
The useful comparison is with the American platforms making the same bet. All of them are spending current margin on compute, hoping the products built on it defend the revenue that paid for it. ByteDance differs in one respect: nobody outside the company can check the homework.
The chips underneath the number
Part of the spending is hardware that ByteDance is not free to buy at will. Reporting around these results referred to a potential purchase of 20,000 Nvidia H200 accelerators, the chip whose sale into China Beijing spent a year restricting and then, as we reported in August, moved to permit for named companies including ByteDance.
That is the constraint behind the capital expenditure figure. A Chinese platform's AI budget is not only a question of what it can afford, but of what it is allowed to buy, from whom, and with which government's approval on each side of the transaction.
What "private company" means for these figures
ByteDance is not listed. There is no filing behind any of this, no audited statement, and no management call where someone can ask a follow-up question.
The figures come from The Information, citing sources, and from Zhitong Finance, citing market sources. Two outlets reporting similar numbers from unnamed people is the normal state of knowledge about this company, but weaker than a quarterly report. Nothing here has been confirmed by ByteDance.
These numbers should not be ignored, because they have been broadly reliable in the past. They should be held loosely. That is particularly true of the profit line, which is sensitive to how internal accounts allocate the cost of infrastructure still under construction.
What to watch
Watch whether the capital expenditure figure rises again. Reports through this year have ranged from $23 billion upwards, and the gap between plan and outturn is where the real commitment shows.
Then there is TikTok's contribution. International advertising and e-commerce carried this half, and both are exposed to regulatory decisions that have nothing to do with how well the products work.
Finally, watch whether any of this becomes checkable. A listing would turn sourced estimates into filings, and the same question hangs over several of the largest AI spenders in Asia, whose infrastructure commitments are mostly reported rather than disclosed.
