Munich Startup: What does your startup, Pruna AI, do? What problem do you solve?
Bertrand Charpentier, Pruna AI: AI requires enormous computing resources. For example, if everyone in Germany generated five pages of ChatGPT every day, we would need two nuclear power plants running continuously to generate enough energy. To solve this problem, Pruna AI is the go-to resource for ML teams looking to simplify scalable inference. We use advanced (combination) compression methods to automatically make every model faster, smaller, cheaper, and more environmentally friendly – on any hardware and for any application.
Munich Startup: But that's been around for a long time!
Bertrand Charpentier: AI, and especially deep learning, are constantly evolving fields where new data, models, and hardware continue to push the boundaries of what's possible. At Pruna AI, we automatically adapt to any scenario, effortlessly, by combining the best compression methods based on the user's goals and constraints. This ensures that our solution is always up-to-date, optimized and ideally suited for the respective task, which sets us apart from existing approaches.
Pruna AI wants to make AI accessible to everyone
Munich Startup: What is your founding story?
Bertrand Charpentier: During my PhD, I noticed that, regardless of the company or lab, people were constantly struggling to get enough computing power for their models. I quickly realized that it was not possible to run SOTA models (Editor's note: SOTA = State of the Art) on local computers without consuming enormous resources. This motivated me to delve deeply into efficient AI methods. I then built a team of co-founders—Stephan Günnemann, John Rachwan, and Rayan Nait Mazi—all of whom bring deep expertise in AI and entrepreneurship to found Pruna AI and make AI sustainable for the planet and accessible to everyone.
Munich Startup: What have been your biggest challenges so far?
Bertrand Charpentier: One of our biggest challenges was the widespread lack of understanding of how AI works and the real-world technical challenges it presents. Communicating to people the critical issues driving AI costs and the solutions available is crucial. Our team has focused on raising awareness of these challenges while demonstrating the importance of efficient and sustainable AI development.
Leading company in terms of model downloads
Munich Startup: How are things going?
Bertrand Charpentier: We've already made a significant impact by deploying our models in production for hundreds of thousands of hours, saving our customers millions of compute hours and costs. We've also publicly shared more than 7,000 compressed models, which have been downloaded over 1.3 million times on Hugging Face, making us the leader in model downloads. These milestones demonstrate the growing demand for our technology and the value we bring to the AI community.
Munich Startup: How have you experienced Munich as a startup location so far?
Bertrand Charpentier: Munich is a fantastic location for building Pruna AI. With its leading universities, vibrant startup ecosystem, and strong focus on technology and innovation, Munich offers numerous opportunities to further develop both our team and our business. The proximity to top-level research institutions has also been a huge advantage in advancing our technology and our mission.
Munich Startup: Risk or security?
Bertrand Charpentier: There are two types of risks: those you can control and mitigate through testing and experimentation, and those you can't control because they depend on external factors. The key is to position yourself to explore opportunities while minimizing the risk of factors you can't control. This allows you to continue to grow and thrive, even under uncertain conditions.
