Amazon Web Services (AWS) provides EC2 instances with instance storage. This provides storage on drives that are physically attached to the host computer. With local NVMe drives this can provide excellent I/O performance, better than available for networked storage such as S3 or EBS. For example i3en instances provide up to 60TB storage, 2 million random IOPS (4kB block size) and total disk throughput of 16GB/s (as of May 2019).
By contrast, latency is relatively high for EBS - it can easily be a few milliseconds, compared to microseconds for a local NVMe drive.
The instance storage persists for the life of the EC2 instance (as long as the EC2 instance isn't stopped/started and the drive doesn't fail). Note in particular that the EC2 instance can be rebooted (intentionally or unintentionally) without losing persistent data.
For valuable long-term data, instance storage should only be used when sufficient fault tolerance is built in. For example, this can involve use of a replicated, distributed DBMS such as CEDA. The data can be replicated and synchronised across any number of EC2 instances, perhaps spanning multiple data centres. To lose data requires all nodes to fail and this is extremely unlikely with multiple replicas.