OpenAI’s latest report on AI’s impact on the European labour market goes beyond merely listing endangered professions. It paints a dynamic picture where some jobs will be automated, others will grow, and many will change shape. Read through the lens of those building or managing on-premise AI infrastructure, this mapping raises specific questions about the workforce needed to make an adoption away from the cloud sustainable.

From cloud to local: why skills matter

The dominant narrative presents AI as accessible via API, with ever-larger models served by a handful of providers. Yet across Europe, the push toward self-hosted deployments is accelerating: sensitive data to protect, GDPR constraints, inference costs at high volumes. In these scenarios, the choice is not just technical but organizational. Putting an LLM into production on dedicated hardware, perhaps with fine-tuning on internal documents, requires people able to manage pipelines, quantization, and performance monitoring. Buying GPUs is not enough.

OpenAI’s map touches this exposed nerve: if many administrative or back-office tasks are headed for heavy automation, who will orchestrate the models inside the company? Hybrid profiles – system administrators with ML knowledge, data engineers with infrastructure awareness – become the real bottleneck. And they gain value precisely when the decision is not to outsource everything to the cloud.

Data sovereignty and the cost of skilled labour

A well-executed on-premise deployment returns full control over data flows. But that control comes at a price: training or hiring specialized staff. European Union estimates on the shortage of ICT experts intersect with the report’s forecasts: the roles most exposed to automation are often the least tied to direct management of AI stacks, while demand for skills to design, train, and maintain in-house models is set to rise.

This tension goes to the heart of TCO evaluations. While cloud inference costs can explode as token volumes increase, the cost of specialized labour for a local infrastructure remains a fixed, not always predictable, line item. OpenAI’s report, without diving into architectural choices, offers a helpful frame: understanding which tasks will be absorbed by AI helps size training investments, avoiding automating before having built the team that will govern automation.

Beyond the report: the AI-RADAR perspective

For those evaluating whether to bring models in-house or stick with external APIs, workforce data adds a strategic variable. It is no longer just a matter of VRAM, low-latency throughput, or quantization strategies to fit a 70-billion-parameter model on a single machine. It is also a problem of people who can interpret workloads, choose the right frameworks, and keep the entire stack aligned with business goals.

AI-RADAR steps precisely into this space, offering analytical tools to assess trade-offs between on-premise and cloud deployments. OpenAI’s mapping becomes a piece to design adoption plans that do not overlook the human component. Because in a local stack, data sovereignty is defended with hardware, software, and skills: if one of the three legs is missing, the project stumbles.