Key points
- Atlassian uses Nvidia Nemotron and Mistral models for its AI features.
- The CEO stated there is no customer pushback against open models.
- Microsoft recently released a decision model built on Alibaba’s Qwen.
Atlassian chief executive Mike Cannon-Brookes confirmed that the company does not operate any Chinese artificial intelligence models within its production systems. He made this statement during an interview with 20VC, a podcast hosted by investor Harry Stebbings. The episode was published on Monday. According to The Next Web, the recording took place in Amsterdam. This location hosted Atlassian’s Team ’26 Europe conference last week. Stebbings initiated the discussion regarding Chinese models. He noted that SaaStr founder Jason Lemkin had indicated large enterprises generally avoid Chinese open models. Cannon-Brookes acknowledged potential customer resistance to specific Chinese models. However, he emphasized that Atlassian follows its customers' preferences. He clarified that there is no customer pushback against open models in general. The company aims to select the best model for each task. Cost efficiency for customers is a primary driver in these selections. Atlassian maintains a clear stance on its current production environment. The CEO stated they do not use any Chinese models in production. He argued that the distinction between models can be difficult to draw.
He claimed many US models are trained partly on Chinese ones. He provided no specific examples or evidence for this claim. Refusing inspectable open technology may be unsustainable, he suggested. Enterprises might struggle to reject cheaper alternatives with better results. He expressed skepticism about companies claiming to have built their own AI models. Many such firms likely fine-tuned Chinese open-weight models for specific jobs. Atlassian also fine-tunes open models but does not claim them as proprietary. The benefits of building specialized models remain uncertain, he noted. Specialized models must consistently outperform general models on price and performance. Few achieve this, according to his assessment. Many early startups have been overtaken by large foundation models. Chinese open-weight models are already embedded in some US products. Microsoft released a small decision model built on Alibaba’s Qwen on Monday. Cost remains a significant theme in AI adoption for large customers. Atlassian employs a full AI gateway team to route tasks. This team sends each task to the most suitable model. Product teams are advised not to worry about cost too early.
By the numbers
- Team ’26 Europe: The conference where the interview recording took place.
- Monday: The day the podcast episode was released.
- Zero: The number of Chinese AI models running in Atlassian production.
What the figures mean
The absence of Chinese models in Atlassian's production reflects a strategic choice. This decision aligns with current customer preferences in certain markets. The use of Nvidia Nemotron and Mistral models highlights a partnership approach. Atlassian leverages deep partnerships with chipmakers like Nvidia. This allows them to access specific models for their services. The company's AI gateway team optimizes task routing. This ensures efficient use of resources across different models. The focus on cost and performance drives these technical decisions. Customers care significantly about the cost of AI features. Atlassian aims to provide the best results at the lowest cost.
This strategy involves selecting models based on specific task requirements. The company does not prioritize proprietary model development. Instead, they utilize existing open models and fine-tune them. This approach allows for flexibility and cost efficiency. The market landscape is shifting rapidly with new model releases. Microsoft's recent release of a Qwen-based model illustrates this trend. It shows that Chinese models are entering US products indirectly. Atlassian's stance may evolve if customer preferences change. The CEO indicated that the situation with Chinese models could change. He believes the line between models is harder to draw than it appears. This suggests a nuanced view of the AI supply chain.
What to watch
Monitor changes in Atlassian's model partnerships and selections. Watch for updates on customer acceptance of open models. Track the performance and cost of specialized AI models. Observe the integration of Chinese models in US products.
What to do and how to stay safe: Atlassian
- Review your organization's policy on using AI models from specific regions.
- Verify the origin of AI models used in your production environments.
- Assess the cost and performance trade-offs of different AI models.
- Monitor vendor announcements regarding model partnerships and updates.
Step-by-step guide: Detect Shadow AI: Network Logs, Process Trees and Hidden Signals
General security guidance from the Sandbox Gazette newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Does Atlassian use Chinese AI models in production?
No, Atlassian CEO Mike Cannon-Brookes confirmed they do not use any Chinese models in production.
Which AI models does Atlassian use?
Atlassian uses Nvidia’s Nemotron models and models from Mistral, among others.
Why does Atlassian avoid Chinese models?
The CEO cited customer pushback in some places against Chinese models, though not against open models.
