By 2025 China’s biggest internet companies have turned artificial intelligence into a capital-intensive war of ecosystems. Alibaba, Tencent and ByteDance have together committed spending measured in the low trillions of yuan over the next few years, but each firm is pursuing a markedly different playbook: Alibaba is building a fortified cloud-and-commerce moat; Tencent is squeezing immediate monetisation from social, games and ads; ByteDance is trying to rewire the user gateway itself with system-level AI.
Alibaba has been the most explicit about scale. The group announced a three‑year plan to pour roughly RMB 3,800bn into cloud and AI infrastructure — an amount that eclipses its investment over the previous decade — and has already recorded some RMB 1,200bn of capital expenditure in recent quarters. Those outlays have shown up in the accounts: free cash flow swung from a net inflow to a net outflow as the company sacrifices short‑term financial polish for what it hopes will become a durable infrastructure advantage. Alibaba’s strategy is classical platform economics: deepen services inside its commerce and local‑service stack so an AI “conversational” layer becomes a direct transaction terminal.
Tencent’s approach is the inverse. Its capital spending has risen sharply but the group has steered AI development toward products that translate quickly into higher margins — upgraded ad targeting, automated content and game production tools — and those moves have helped lift gross margins and operating profit even as capex grows. Tencent is betting that its private‑domain strengths — chiefly WeChat’s daily attention and payments plumbing — can host an AI agent that increases monetisable engagement without needing the same scale of heavy infrastructure outlays as Alibaba’s cloud push.
ByteDance has taken the most aggressive aim at the plumbing of the smartphone era. Its AI assistant and a new handset strategy seek to sidestep traditional app icons and stores by intercepting user intent at a lower system level. With daily active users for its AI assistant reportedly already surpassing 100 million, ByteDance hopes to make search, discovery and commerce decisions before users even open an app, a move that would fracture the classic app‑based distribution model and concentrate enormous control over what users see.
The contest for what the industry calls the “super entrance” — the single gateway through which consumers discover content, shop and transact — is thus playing out across software, cloud infrastructure and hardware. Events such as the viral technical breakthroughs of early 2025 hastened a phase transition: the race is less about headline model parameters and more about who controls data, payment rails, latency‑optimised inference and the user relationship.
That shift has exposed sharp practical vulnerabilities. Organisational churn is under way as groups reconfigure around model research, data platforms and inference infrastructure; talent poaching has driven core technical salaries up by more than 20% in the past year, raising running costs. Most alarmingly for all players, high‑performance GPUs remain a choke point. Heavy reliance on external suppliers — and on limited allocations from a handful of advanced foundries and chip vendors — means a firm’s algorithmic lead can be throttled by hardware scarcity.
Investors and executives alike are watching another pressure: open models and cost deflation on API pricing have shrunk margins on pure model provisioning, forcing companies to look for higher‑value bundles (commerce conversion, advertising yield, gaming content) rather than monetising raw compute. The result is a more pragmatic, gruelling phase of competition in which capital intensity, ecosystem control and supply‑chain resilience will decide winners as much as research breakthroughs.