Atlassian Does Not Use Chinese AI Models in Production, CEO Cannon-Brookes Says

Atlassian CEO Mike Cannon-Brookes in a 20VC podcast interview (host Harry Stebbings) clearly stated: the company does not use Chinese AI models in production. The interview was recorded in Amsterdam at the Team '26 Europe conference.

The conversation was prompted by SaaStr founder Jason Lemkin's statement that large companies avoid Chinese open-weight models. Cannon-Brookes agreed that this could change, but Atlassian follows its customers: 'Yes, there is customer resistance to Chinese models in some places. But there is no resistance to open models.'

What Models Are Used at Atlassian

Instead of Chinese solutions, the company relies on:

• Nvidia Nemotron models — through deep partnership with the chipmaker • Mistral models • Open models from other vendors, which the company fine-tunes for specific tasks

Atlassian's goal: to choose the best model for each use case at the lowest cost for customers. Routing is handled by its own AI gateway team.

The Line Between Proprietary and Open Models Blurs

However, Cannon-Brookes acknowledged the complexity of the position. According to him, many American AI models are partially trained on Chinese models, but this is not disclosed. Moreover, Microsoft on Monday released a small decision model based on Qwen from Alibaba — a Chinese company.

Cannon-Brookes is skeptical of companies claiming to have their own AI models. Most simply take a Chinese open-weight model and fine-tune it for one task, but don't call it their own. Atlassian also fine-tunes open models, but is honest about it.

'The jury is still out on the value of specialized models,' says the CEO. In practice, specialized solutions rarely beat the latest general models on either price or performance. Many early startups with their own models have lost to the mass market.

Cost Is the Main Concern for Enterprise Customers

The second major topic of the interview is finances. Enterprise customers are increasingly concerned about what AI will cost them. Atlassian has a full AI gateway team that routes each task to the optimal model.

Cannon-Brookes gives product teams clear guidance: don't obsess over cost too early. 'If five people use a feature, I don't care how much it costs. If five million — I'll be very concerned about the price.'

One of the major shifts in his thinking this year: AI running costs will fall even more than he expected.

Facts and Figures

• July 2026: Atlassian gave its engineers monthly AI budgets • August 2026: 28% revenue growth in Q4 • Key partners: Nvidia, Mistral, open models • Key tool: AI gateway team for task routing

Cannon-Brookes' conclusion: companies will be forced to adopt cheaper and more effective solutions, even knowing what's inside them. It will become increasingly difficult to abandon open technologies that can be verified.