Security fears put on-prem systems back in the buyer conversation

Strategic adviser Itay Sagie says demand for on-premises systems is reappearing among some buyers, based on a recent conversation with a PBX vendor and his read of security concerns around AI and quantum computing. No spending figures, customer names, or vendor data were disclosed, so the claim is best treated as an early signal rather than evidence of a broad market reversal.

Sagie, a guest contributor to Crunchbase News and an adviser to technology companies, investors, CEOs and boards, wrote that a PBX vendor told him customers are again asking about on-premises deployments. He argues that the shift is being driven by anxiety over where core infrastructure and sensitive data sit, particularly as AI-enabled fraud improves and companies put more proprietary data into AI systems.

The argument does not reject cloud computing. Sagie notes that the cloud still gives companies speed, scale and lower upfront costs, and that many providers can secure infrastructure better than a single company could on its own. The change he describes is more specific: for systems involving communications, payments, identity and customer data, some decision-makers may be placing a higher value on direct control.

AI fraud is pressuring trust systems

Sagie points to AI-enabled phishing, more credible fake invoices and voice impersonation as reasons buyers may reassess exposure. In his view, a request that appears to come from a senior executive, including by voice, is becoming harder for employees and systems to evaluate.

That broadens the security problem beyond servers. Sagie lists identity tools, SaaS products, APIs, employee processes, permissions, contractors and support portals as part of the exposed surface area. On-premises deployment does not make those risks disappear, he says, but it can reduce reliance on outside platforms and give companies more direct ownership of systems they consider too sensitive to outsource.

Production AI changes the cost and risk equation

Sagie also argues that enterprise AI pilots and enterprise AI operations have different infrastructure needs. Cloud AI APIs are useful when teams want to test quickly without buying GPUs, managing models or building a large infrastructure group. Once AI systems move into production, he says, the economics and data risks change.

The AI applications with the clearest business use often depend on proprietary material: contracts, source code, customer records, financial documents, support tickets, security logs, medical files and internal messages. Sagie’s view is that private or on-premises AI infrastructure can make more sense for those workloads because models can run closer to the data, with tighter access rules and clearer retention, compliance and audit controls.

He also raises a cost issue familiar to operators scaling AI products internally. Per-token or per-query pricing can be convenient during testing, but recurring usage can become expensive when large numbers of employees or customers use AI tools daily. For stable, high-volume workloads, Sagie argues that owned or controlled infrastructure may prove cheaper than paying for each interaction through a cloud service.

Quantum risk is entering security planning

Sagie is not claiming that quantum computers are breaking enterprise encryption today. His point is that companies with long-lived sensitive data are already considering what commercial quantum computing could mean for encrypted information stored outside their direct control.

He identifies banks, healthcare providers, telecom companies, governments, defense-related organizations and infrastructure providers as groups with the most at stake. Whether on-premises systems are the right technical answer is unresolved, but Sagie argues that the perception of greater control could still push more buyers to revisit older deployment models.

This story draws on original reporting from Crunchbase News.