Serverless stopped being a niche
For most new workloads, the question is no longer whether to run servers but which parts genuinely need them. Managed runtimes now cover long-running jobs, streaming and background processing — the workloads that used to force teams back onto always-on infrastructure.
The practical benefit is not the cost curve, which is often neutral at scale. It is the operational surface: fewer patched hosts, fewer capacity decisions, and a smaller estate for your security team to defend.
Edge computing moves work closer to the data
Latency and data residency are pushing inference and filtering out of the central region and toward the point of collection. We see this most in industrial and infrastructure settings, where bandwidth is limited and the interesting events are rare.
The architectural pattern that works: process and discard at the edge, and send only what carries signal to the centre. It reduces cost, and it sidesteps a whole category of privacy problems by never moving the raw data at all.
Multi-cloud, honestly assessed
Multi-cloud is often sold as insurance against lock-in. In practice, teams that spread workloads across providers to stay portable usually pay for it twice — once in engineering complexity and again in the managed services they can no longer use.
The version that does work is deliberate and narrow:
- Pick a primary provider and use its managed services fully
- Place specific workloads elsewhere for a concrete reason — data residency, a unique service, a customer requirement
- Keep the portable layer thin: containers, IaC and your own data formats
- Plan exit paths on paper before you need them, rather than building for them everywhere
What to plan for
Budget for GPU capacity as a first-class infrastructure concern, not an experiment line item. Treat FinOps as an engineering practice with owners and targets. And assume your compliance obligations will tighten rather than relax — building the audit trail now is far cheaper than retrofitting it later.
Working on something like this?
We are happy to talk through the approach, whether or not it turns into a project.
Start a conversation
