Liwei Two Minutes #1: Why Are Tokens Getting More Power-Hungry?

In the past two years, many people realized for the first time: AI is this power-hungry. It's even starting to compete for electricity.

Isn't it just chatting, writing articles, generating some images? How did it suddenly become an energy monster?

Because today's large models are, at their core, burning tokens at massive scale. Once tokens enter industrial production, the power consumption will be staggering.

The internet is about information transmission. AI is about real-time computation. A search engine is like looking up a dictionary. A large model is like writing an essay from scratch.

The model predicts one token at a time. Behind every bit of generated content is a sea of matrix computation.

Today, GPUs have essentially become token generators. What you consume isn't chat sessions — it's token throughput.

After agents emerged, AI itself started consuming tokens. Thousands, tens of thousands of times — invisible to humans.

It's like the Industrial Revolution. Coal went from heating homes to driving factories, trains, and ships. Tokens went from chatting to industrial fuel.

The whole world is frantically building data centers, power plants, nuclear reactors. AI competition is no longer about algorithms — it's about who can burn tokens continuously, stably, and cheaply.

AI companies increasingly look like subsidiaries of a new energy-industrial complex. The internet flows with bits. AI burns tokens.

That's today's 立委两分钟. Thanks for watching. Goodbye. by Tuya

发布者

立委

立委博士,多模态大模型应用咨询师。出门问问大模型团队前工程副总裁,聚焦大模型及其AIGC应用。Netbase前首席科学家10年,期间指挥研发了18种语言的理解和应用系统,鲁棒、线速,scale up to 社会媒体大数据,语义落地到舆情挖掘产品,成为美国NLP工业落地的领跑者。Cymfony前研发副总八年,曾荣获第一届问答系统第一名(TREC-8 QA Track),并赢得17个小企业创新研究的信息抽取项目(PI for 17 SBIRs)。

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注

这个站点使用 Akismet 来减少垃圾评论。了解你的评论数据如何被处理