Atlassian does not use Chinese AI models in production, its CEO says

Atlassian does not run any Chinese AI models in its products, chief executive Mike Cannon-Brookes said in an interview with 20VC, the podcast hosted by investor Harry Stebbings.

The episode came out on Monday and was recorded in Amsterdam, where Atlassian held its Team ’26 Europe conference last week.

Stebbings raised the issue, saying SaaStr founder Jason Lemkin had told him big companies do not want Chinese open models. Mike Cannon-Brookes said that may change, but that Atlassian follows its customers.

“There is certainly some customer pushback in some places to, quote unquote, Chinese models. There is not customer pushback to open models,” he said.

Instead, Atlassian uses Nvidia’s Nemotron models through what he called a deep partnership with the chipmaker, as well as models from Mistral and others. Its job is to pick the best model for each task as cheaply as possible for customers, he added.

“We don’t use any Chinese models in production, we’re very clear about that,” he said.

Even so, he argued that the line is harder to draw than it looks. He claimed that many US models are trained partly on Chinese ones without saying so, but gave no examples or evidence.

Refusing open technology that firms can inspect will be hard to sustain, he argued.

“For an enterprise to say I’m not going to use a cheaper alternative that’ll give me better results when I know what’s in it, I think is going to be a very hard long-term choice,” he said.

He was also sceptical of companies that say they have built their own AI models. Most of them, he said, have taken a Chinese open-weight model and fine-tuned it for one job. Atlassian fine-tunes open models too, he added, but does not call them its own.

“The jury is really out on the benefits of building specialised models at the moment,” he said.

To change his mind, he said, specialised models would have to keep beating the latest general models on price and performance. Few do, in his view, and many early start-ups have since been overtaken by the big foundation models.

Chinese open-weight models are already embedded in some US products. On Monday, Microsoft released a small decision model built on Alibaba’s Qwen.

Cost was the other theme of the interview. Large customers now care a great deal about what AI costs them, Cannon-Brookes said.

Atlassian has a full AI gateway team that sends each task to the model that suits it, but he tells product teams not to worry about cost too early.

“If five people use a feature, I don’t care how much it costs. If 5 million people use a feature, I’m going to care a lot about cost,” he said.

Asked what he had changed his mind on this year, he said AI running costs will fall further than he once thought.

In July, Atlassian gave its own engineers monthly AI budgets, and in August it reported 28% revenue growth for its fourth quarter.