There is a natural fit for machine learning (ML) workloads on Kubernetes, as it is well-suited to meet the scalability needs of machine learning jobs as well as embracing the continuous development nature of ML models.
- High risk: Up to 87% of machine learning initiatives are abandoned before they reach production.
- Long time to value: For those initiatives that do make it to production, it can take more than 3 months for a single model to be deployed. Software provisioning at enterprises can take weeks or even months, which adds time and delays obtaining value.
- New technology: end-to-end machine learning platforms for the big data and deep learning era have only been around since 2016, with few technologies that are cloud native.
- Complexity: build-or-buy decisions for scalable platforms require immense knowledge of cloud-native infrastructure as well as the entire ML landscape.