Breaking

Kai-Fu Lee Says China Will Capture AI Reach While U.S. Labs Keep the Profit Lead

September 6, 2026 at 11:49 EDT

Foundation Models
Open Source
Sales & Business

Kai-Fu Lee, the former president of Google China and founder of 01.ai, said Chinese open-weight models are positioned to win global usage even as U.S. labs keep the profit lead — and argued that chief executives still underestimate how fast the technology is stretching what a single system can do.

September 4, 2026 · Kai-Fu Lee

China takes AI reach. America keeps the bill.

Former Google China chief Kai-Fu Lee says open Chinese models sit about six months behind closed U.S. frontier systems — close enough, in his view, to win global usage while OpenAI and Anthropic keep iPhone-like profits.

Magnitude

What a model can finish in one sitting

Lee says AI now solves tasks 10 times longer than a year ago. Each block is one unit of last year’s workload.

1×

4-minute task
one year ago

10×

40-minute task
now

Takeaway: the unit of work a model can finish has grown an order of magnitude in a year.

Price analogy

New Tesla vs six months old

Lee’s buyer choice for a closed frontier model versus an open one from six months earlier. Each block = $5,000.

$15,000

Open model
~6 months behind

$50,000

Closed frontier
sold at a premium

1–3%

DeepSeek GPU power vs U.S. rivals, in Lee’s estimate

~6 mo

Typical quality lag he cited versus closed U.S. labs

~1,000

Competing Chinese large language models on the market

U.S. closed labs

The iPhone play

OpenAI and Anthropic still rank first or second on quality. Lee says they will make more money by selling high-priced enterprise products.

China open models

The Android play

DeepSeek and Moonshot ship cheaper open weights. Lee sees more share and usage — and a risk that users copy the model and pay only for servers.

The company of five years from now

20 people · 100 AI workers

Lee’s starting sketch for a unit in his own business. AI cannot be accountable, so humans still design the problem, organize the work, and take the blame.

Design

Frame the right problem

Organize

Command parallel AI work

Account

A person owns the outcome

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.