As artificial intelligence moves from optional software to core infrastructure, IBM is making a commercial case for openness. The argument is not that every model must be free or unrestricted. It is that enterprises protect flexibility, security and margins when they can inspect, govern and replace important parts of their AI stack.
IBM software executive Rob Thomas describes a familiar progression: technology begins as a product, develops into a platform and eventually becomes infrastructure. At infrastructure scale, openness becomes a practical requirement because the technology sits inside code, security, decisions and daily operations.
Opaque services can be convenient at first. Problems emerge when companies must integrate them with sensitive systems, investigate failures, control latency or predict rapidly growing API costs.
Closed dependencies can quietly erode margins
An AI project’s headline model price rarely represents its full cost. Teams may need to sanitise sensitive data before sending it to an external service, construct additional troubleshooting layers and provision capacity for peaks. Latency can make automation slower than expected, while a provider’s technical or commercial change can force expensive rework.
The risk becomes larger as one model is embedded across many workflows. A company that cannot move workloads may have little leverage over cost or performance. It may also be unable to use a smaller, cheaper model for routine requests when a premium system is unnecessary.
IBM advocates model-swappable orchestration: applications call a governed layer that can route tasks to different models. A complex security investigation may warrant a powerful system, while classification or extraction could run locally on a smaller model. This preserves optionality without requiring employees to understand the underlying infrastructure.
Openness also improves troubleshooting. Access to architecture, code or model details can help technical teams identify whether a problem lies in data, orchestration or the model itself. A sealed service reduces that visibility and can leave customers waiting on the provider.
Openness does not mean abandoning safety
Powerful capabilities still require controls. The source points to Claude Mythos and Project Glasswing as an example of advanced cyber capability offered with restricted access. The issue is not whether every user should receive every function, but whether governance is explicit and appropriate to the risk.
Open-source components can benefit from broader scrutiny, allowing weaknesses to be found and corrected by more than one vendor. Enterprises can audit, adapt and deploy them under their own policies. Yet openness alone does not guarantee security; organisations still need tested configurations, access restrictions, monitoring and accountability.
Transparency makes those controls more informed. Teams can know which data a system uses, where it operates and what happens when it fails. That knowledge is particularly valuable in regulated or sensitive environments where a general assurance from a provider may be insufficient.
The strongest architecture combines freedom of choice with firm operational boundaries. Models can be replaceable while permissions, logging and evaluation remain consistent across the platform.
Value moves upward into implementation
IBM’s position is that open source does not eliminate commercial opportunity. It shifts value toward orchestration, reliability, governance and domain expertise—the work required to turn a capable model into a dependable business system.
That shift is already familiar from other infrastructure markets. Widely available building blocks create larger ecosystems for services and specialised products. Customers pay for outcomes, support and reduced operational risk rather than scarcity alone.
For enterprise buyers, the relevant question is therefore not simply whether a model is open or closed. They should examine how easily workloads can move, whether costs are observable, how incidents can be investigated and who controls sensitive information. A nominally open component inside a tightly locked platform may offer little practical freedom.
Robust governance protects margins because it enables deliberate choices. Organisations can reserve expensive models for tasks that justify them, keep sensitive workloads in controlled environments and change suppliers without rebuilding every application.
AI infrastructure will continue to include both proprietary and open technology. IBM’s case is that enterprises should avoid turning one provider’s model into an invisible dependency beneath the entire business. Openness, orchestration and transparency create room to negotiate, optimise and adapt. In a market where models and prices change quickly, that room may be one of the most valuable assets an AI strategy can preserve.
