ByteDance shifts to AI infrastructure

ByteDance is starting to look less like a high margin ad machine and more like an AI infrastructure company that still happens to own TikTok. Revenue kept climbing to an estimated $186B in 2025, but profit collapsed as spending shifted into GPUs, data centers, chips, and model training. That changes how the business should be read, because more cash is now being reinvested into compute capacity before it shows up in earnings.1 2 3 4

• The squeeze is happening on top of a still growing core business. ByteDance's revenue dataset shows growth slowing from 38% in 2024 to 20% in 2025, which means profit pressure is being driven less by demand weakness and more by a deliberate cost reset tied to AI buildout.1 2 4
• This mirrors the capital intensity now seen at the largest AI platforms. Meta spent $69.7B on property and equipment in 2025, mostly for servers, data centers, and network infrastructure, while ByteDance is reportedly discussing up to $70B of 2026 capex, putting it in the same spending class even without public market financing.3 5
• The strategic logic is that ByteDance already has huge consumer surfaces to feed. TikTok U.S. alone was estimated at $27B of 2024 revenue and international revenue was the fastest growing part of the company, so owning more of the compute stack can support Doubao, recommendation systems, creator tools like CapCut, and future AI products sold into that installed base.2 6

The next phase is likely a multi year tradeoff where ByteDance accepts lower margins in exchange for building its own AI supply chain and product stack at massive scale. If capex stays near the reported 2026 range, the company will be competing less on quarterly profit and more on who can turn compute into durable consumer and advertising advantage fastest.3 5 6