ByteDance's 70% Profit Plunge: Is the AI Bet a Gamble or a Must?

On April 20th, a financial report hit the tech world like a bomb: ByteDance’s 2025 full-year net profit dropped more than 70% year-over-year.

Don’t jump to conclusions yet. The number is ugly, sure, but you have to look at the structure. The main reason for the profit decline is AI compute investment — in plain terms, ByteDance has been buying GPUs aggressively, building data centers, and training large models. The spending has been enormous, but the returns haven’t arrived yet.

Meanwhile, ByteDance’s overseas revenue grew approximately 20% year-over-year, and TikTok’s performance in global markets remains strong. Domestic business is also growing — it’s just that the AI investment has eaten up most of those profits.

A friend of mine who works in algorithms at ByteDance told me they internally define 2025 as the “strategic loss year” — meaning the short-term profit picture is deliberately made to look bad. It’s a calculated strategic choice. The logic goes: whoever falls behind on AI infrastructure first could be out in the next wave of competition.

I actually buy this logic. AI-era compute is like land during the mobile internet boom — if you don’t buy now, it’ll be even more expensive later. But “agreeing with the logic” and “not worrying about reality” are two different things.

What really concerns me is the “prisoner’s dilemma” across the entire industry.

ByteDance invests, so Tencent must invest. Alibaba invests, so JD can’t sit back either. The result: every major tech company is pouring money into AI, but nobody dares to stop first. Because the moment you pause, you’re perceived as having “conceded defeat” in the AI race.

This collective arms race typically ends with: everyone spending enormous amounts of money, yet nobody building a truly irreplaceable moat. Because when everyone can buy the same GPUs and train the same models, the differentiation at the foundation model layer shrinks — competition eventually returns to product experience and user networks.

For ByteDance specifically, this profit plunge also exposes a structural problem: they don’t have a cash cow like Tencent’s WeChat. Tencent’s gaming and fintech businesses continuously generate profits, giving them more breathing room in AI investment. ByteDance’s TikTok overseas revenue looks impressive, but geopolitical risk has always been hanging over it.

So ByteDance’s current situation is: domestic profits down, overseas risks present, and AI investment that can’t be paused. With all three pressures combined, 2026 will be a tough year for Zhang Yiming.

What I’m genuinely curious about: how’s ByteDance’s AI product (Doubao) doing in terms of C-end penetration? If the AI application layer fails to break through, then this round of AI investment truly becomes a gamble rather than a necessity.

The data tells the story: in the domestic LLM application market, Doubao’s monthly active user numbers vary wildly across different sources. This also shows that AI application layer competition is even more chaotic and unresolved than the foundation model layer.

In summary, ByteDance’s financial report serves as a wake-up call for the entire industry: AI investment is a long-term war, not a short-term sprint. Whoever survives this war won’t be determined by who spends the most, but by who can fastest find the balance between AI investment and commercial return.

Will Zhang Yiming find that balance?