The notion that artificial intelligence will power a new era of blistering growth for the West has found a high-profile critic. Christopher Pissarides, who shared the 2010 Nobel Memorial Prize in Economics and teaches at the London School of Economics, has said the boom years are likely over for good and that AI alone will not resurrect them.

His stance lands at a moment when enterprises are pouring capital into LLM infrastructure, betting that advanced automation and real-time inference will translate into disruptive efficiency gains. Pissarides instead calls for a realism with far broader implications than the academic debate. For those weighing on-premise deployments, the warning reframes the decision: if a macro productivity leap is not guaranteed, it makes sense to shift the lens from immediate economic return to variables such as data sovereignty, predictable operating costs, and the ability to manage sensitive workloads without external dependencies.

Inside Italian and European organizations, often bound by GDPR requirements and a culture of information protection, this reasoning finds fertile ground. A self-hosted model deployment, perhaps running INT8 quantization on enterprise-grade GPUs, is no longer judged solely on tokens per second served, but also on the resilience of internal control. If the expected growth does not materialize rapidly, TCO is no longer seen as an investment with guaranteed returns, but as the price of an infrastructure that provides autonomy and compliance.

Those who have architected on-premise fine-tuning pipelines already feel this tension. Moving inference locally requires significant upfront CapEx, and the absence of a productivity explosion could stretch amortization timelines. Yet sectors such as advanced manufacturing, public healthcare, and regulated finance continue to prefer on-premise precisely because, in a low-growth scenario, the risk of exposing sensitive data or becoming locked into a cloud vendor becomes a hidden cost few can afford.

Pissarides’ point, at its core, does not deny the advances of generative AI; it restrains their transformative scope across the entire economic system. In doing so, it exposes the gap between the disruptive narratives of many providers and the reality of companies that must balance their books every quarter. For those deciding how to allocate technology resources, this discrepancy means that on-premise deployment choices should be evaluated with a broader criterion: not as levers for growth that may never come, but as tools for operational independence in a low-acceleration economy.

AI-RADAR pays growing attention to these balances, offering analytical frameworks at /llm-onpremise that help examine the trade-offs among cost, control, and performance without succumbing to promises of miraculous growth.