The so-called AI revolution is actually built on the shaky foundation of traditional cloud infrastructure.
From infrastructure to intelligence: an uncut chain of dependence.
The false independence of AI.
Many people think that AI infrastructure is a brand-new technology stack, but in fact it is just the "deluxe version" of Cloud Infra.
Take training a large language model as an example: Computing resources: Still relying on EC2 or equivalent virtual machines Instance storage: S3 or similar object storage to save training data and model checkpoints Network: VPC and load balancer to coordinate distributed training Container orchestration: Kubernetes, a decade-old technology to manage GPU clusters, the only real new additions are AI accelerators such as GPU/TPU and some dedicated frameworks.
In 2025, Google Cloud announced that 9 of the 9 top AI laboratories will use its infrastructure, Microsoft emphasized that Azure supports the operation of OpenAI, and Oracle won a US$30 billion five-year order from OpenAI.
Behind these figures is the deep binding of the AI industry to traditional cloud giants.