With the massive push towards AI-native architectures and specialized GPUs in the cloud, I'm wondering if traditional Hybrid Cloud strategies are losing steam. For companies with sensitive data that must remain on-premises, how are you integrating your local hardware with cloud-based AI services like Vertex AI or AWS Bedrock without sacrificing too much latency or security?
3 answers
Hybrid cloud is more relevant than ever because of "Data Gravity." You can't just move petabytes of sensitive legacy data to the cloud for one AI project. The trend for 2025 and 2026 is "AI-ready data stacks" where you keep the data local for compliance but use the cloud for the heavy lifting of model training or inference. Using solutions like Azure Arc or AWS Outposts allows you to run cloud services on your local hardware. This way, you get the consistency of cloud management while keeping your most sensitive datasets behind your own firewall and under your physical control.
Do you think the cost of maintaining high-performance GPUs on-premises is becoming a deterrent for smaller firms, or is the security trade-off still worth the capital expenditure?
Hybrid is the only way for the enterprise. You use the public cloud for its infinite scale and your private cloud for your "crown jewel" data and intellectual property.
Susan is right. Most major banks are still 60% on-prem because of the exact reasons mentioned. Hybrid isn't going anywhere; it’s just evolving into a more unified layer.
For us, it’s about the legal compliance. In the healthcare sector, moving certain patient records is a non-starter. So, even if on-prem GPUs are expensive, we are looking at "Confidential Computing" in the cloud as a middle ground. It's a tough balance between the cost of local hardware and the complexity of these new secure cloud environments.