Tencent invests 100 billion in AI, but transforming it into computing power is a challenge!
Tencent has invested heavily in the AI field but faces challenges due to insufficient computing power. Global advanced chip manufacturers cannot meet the GPU demand in the Chinese market, and building computing power centers takes 9-12 months, constrained by various factors. Research shows that mainstream AI products serving 100 million users require about 400,000 cards, corresponding to ten computing power centers. Liu Chiping mentioned the shortage of computing power at the 2024 revenue meeting, which has affected the fulfillment of demand
Buying cards is the first problem. Advanced process chip manufacturers worldwide are producing 24 hours a day, but the combined capacity of imported GPUs and domestic GPUs cannot meet the demand of major Chinese giants this year.
What’s more troublesome is that GPUs need to be placed in specially designed computing centers. Generally speaking, building a computing center takes 9-12 months, and the investment is also restricted by factors such as licensed contractor capacity and city power planning, making the gap even more severe than buying cards.
A secondary market person who has deeply studied AI estimates that the current mainstream AI products require about 400,000 cards of computing power to serve 100 million daily active users, which corresponds to about ten new computing centers. WeChat has 1.3 billion monthly active users globally, most of whom are in China.
This calculation does not take into account new Agent products; if products similar to Manus become popular, the computing power required per user per day may increase several times.
Liu Chiping also mentioned the issue of computing power shortage at the revenue meeting after the 2024 annual report. “We must install these GPUs in data centers, which takes some time. Therefore, we couldn't capture that much demand in the first quarter of this year.”