← All posts / Industry

Revenue Up 30%, Profit Down: ByteDance's $120 Billion Half Shows the Price of the AI Arms Race

ByteDance's first-half revenue surged ~30% to $120 billion, but net profit slipped by a single-digit percentage to $20 billion as AI infrastructure spending climbed — the clearest evidence yet that the AI arms race is an earnings statement problem.

Revenue Up 30%, Profit Down: ByteDance's $120 Billion Half Shows the Price of the AI Arms Race

The most consequential AI financial report of the year did not come from a Silicon Valley earnings call. It came, quietly and without an IPO prospectus, from Beijing.

According to an exclusive report by The Information published September 15, 2026, ByteDance — the private Chinese company behind TikTok, Doubao, and the world’s most aggressive AI infrastructure build-out outside the United States — saw its first-half 2026 revenue rise roughly 30% year-over-year to about $120 billion. Yet its net profit fell by a single-digit percentage, settling at around $20 billion. The culprit, per the report: surging spending on artificial intelligence infrastructure.

For a company that has been one of the world’s most reliably profitable internet businesses, that combination — booming top line, shrinking bottom line — marks a genuine turning point. It is the financial signature of the AI arms race finally reaching the income statements of even its richest combatants.

The numbers

The Information’s reporting, drawing on internal financials seen by the outlet, sketches ByteDance’s half-year in stark terms:

  • Revenue: ~$120 billion, up roughly 30% year-over-year — an extraordinary figure for a company of this scale, and one that keeps ByteDance comfortably ahead of Meta in quarterly revenue terms.
  • Net profit: ~$20 billion, down by a single-digit percentage from the first half of 2025.
  • The decline was driven primarily by AI-related capital expenditure and operating costs, including compute purchases, data center build-outs, and the subsidised pricing that keeps Doubao — China’s most-used AI assistant — cheap for consumers.

To put that in context: a single-digit profit decline on 30% revenue growth implies margins compressed meaningfully. In 2024, ByteDance posted roughly $33 billion in net profit for the full year on $155 billion in revenue. The company then saw profit fall more than 70% in 2025 as its “all-in” AI drive kicked in, according to earlier reports from Caixin and the South China Morning Post. The first half of 2026 suggests the margin squeeze has not relented — but the revenue engine has more than compensated on the top line.

TikTok Shop is the growth story

The Information notes that first-half revenue growth was “led partly by TikTok” — and specifically by TikTok Shop, the commerce arm that has become the fastest-growing piece of ByteDance’s empire. Earlier reporting on 2025 figures showed overseas revenue growing nearly 50% while domestic (Chinese) revenue grew about 20%, with TikTok Shop’s gross merchandise value expanding aggressively in Southeast Asia, Europe, and Latin America. Forbes reported in August that TikTok’s Europe, Africa and LatAm operations turned their first profit, with international revenues booming to roughly $91 billion.

That international engine matters for the AI story in a direct way: TikTok Shop is itself an enormous consumer of AI — recommendation models, live-commerce moderation, logistics optimisation, and increasingly AI-generated advertising creative. ByteDance is, in effect, running a flywheel where AI spending boosts commerce and advertising, which funds more AI spending.

The $70 billion capex question

The profit decline would be alarming if it were not so clearly deliberate. Bloomberg reported in May 2026 that ByteDance was discussing capital expenditures of as much as $70 billion for the year — more than double the roughly $25 billion it spent in 2025, and a figure that puts it in the same spending tier as Microsoft and Alphabet on an absolute basis, despite ByteDance having no public-market access to capital.

To finance this, the company pulled off the largest private debt deal in Chinese tech history: a $29.6 billion syndicated loan arranged in early September 2026, which this blog covered at the time. Banks, evidently, are willing to underwrite the bet. The first-half financials now show what that borrowed money is doing to earnings: it is being converted, immediately and at enormous scale, into GPUs, data centers, and model training runs.

On the model side, the spending has produced real results. ByteDance’s Doubao has become the most broadly adopted AI assistant in China, and the company claims its latest models compete with — and in some visual benchmarks beat — frontier models from Anthropic and OpenAI. Its Seedance video model set records for 4K generation, and its open-weight releases have become fixtures on global leaderboards. ByteDance now makes roughly $4 billion a year from AI-related revenue, the largest publicly claimed figure for any Chinese company, according to South China Morning Post reporting cited by The Next Web — still a rounding error against its ad business, but growing fast.

Why it matters beyond ByteDance

Three broader signals sit inside this report.

First, the AI capex tax is universal. ByteDance’s results arrive in the same season that Alibaba reported a 76% profit drop on AI spending in its June quarter, with cloud revenue up 45% on surging AI demand. Meta, Microsoft, Alphabet and Amazon have all told versions of the same story: unprecedented revenue growth partially absorbed by unprecedented infrastructure bills. That ByteDance — a company with fatter margins and a more powerful commerce flywheel than most — cannot escape the squeeze suggests the economics of frontier AI remain brutal for everyone.

Second, private companies can play this game longer. Unlike its listed peers, ByteDance faces no quarterly earnings pressure, no activist investors, and no obligation to disclose anything at all. It can run margins to zero for years if its leadership — with founder Zhang Yiming now personally leading real-time world-model efforts — believes the endpoint justifies it. The fact that we only learn of these results through investigative reporting is itself a competitive advantage: rivals must show their cards every 90 days; ByteDance does not.

Third, the US–China spending gap is functionally closed. A year ago, the comfortable Western narrative was that Chinese labs were compute-starved and would fall behind. Instead, ByteDance’s planned 2026 capex of up to $70 billion — funded by domestic loans, sovereign-cloud deals, and a re-exported Nvidia chip pipeline — puts it alongside the biggest American spenders. The margin compression its financials now reveal is not a sign of weakness. It is the cost of admission.

The road ahead

The second half of 2026 will test whether the bet pays off. ByteDance is reportedly preparing to fold its Trae and Coze developer tools into Doubao and launch a “Doubao Work” enterprise suite aimed squarely at Tencent’s WeChat-anchored office business. TikTok Shop continues its push into Europe and Japan, markets ByteDance views as a $770 billion growth engine. And the company’s world-model team, led personally by Zhang Yiming, is racing OpenAI and Google toward real-time interactive video — arguably the most compute-hungry ambition in the industry.

None of that is cheap. But the first-half numbers suggest ByteDance has made its choice: it would rather be the company that spent everything and won the AI transition than the one that protected its margins and lost it. For an industry still arguing about whether AI spending is a bubble, the most aggressive spender outside the US just showed its receipts — revenue up 30%, profit down, and no plans to slow down.