The debate is shifting. Not because AI’s potential has diminished, but because for many companies, the numbers just don’t add up anymore. When AI becomes too expensive for companies, it’s not just about budgets. It’s about control, dependency, and the question of what a solid ROI for AI should even look like.
A recent LinkedIn News article sums up this shift perfectly: the mood surrounding AI is becoming more sober. Rising costs for models, infrastructure, and integration are colliding with an uncomfortable reality. Many initiatives look promising in pilot phases but fail to deliver a clear economic advantage when fully operational.
This is an important signal, especially for European companies. After all, AI has never been merely a question of tools here. It is always also a matter of governance, architecture, and sovereignty.
The real problem isn’t AI. It’s the economic design.
Many early AI roadmaps were developed under pressure to deliver results. Boards wanted to demonstrate their ability to take action. Innovation teams wanted to launch pilot projects quickly. Vendors promised productivity, automation, and speed. No one wanted to give the impression that they were late to the AI game.
This led to the development of ambitious programs featuring copilots, assistants, search systems, and automation initiatives. On paper, it looked like a fresh start. In practice, however, it quickly became clear that a pilot program is not yet a robust operational model.
What matters is not whether a model can produce impressive results. What matters is whether it makes an end-to-end process measurably better, more cost-effective, and more manageable.
1. The costs of AI often rise faster than the actual benefits
The obvious costs are easy to identify: development and API costs, infrastructure, platforms, and licenses. However, the costs that only become apparent as the system scales are the ones that really matter.
- Integration effort in existing system landscapes
- Quality assurance and human review loops
- Security, data protection, and compliance efforts
- Monitoring, governance, and ongoing workflow maintenance
- Change Management, Training, and Internal Adoption
This is precisely where the logic breaks down in many companies. At first glance, heavy use of AI appears to be a step forward. However, if increased token consumption cannot be translated into proportional business value, the promise of productivity turns into a new layer of costs and coordination.
Anyone who wants to seriously implement AI must therefore measure not only model performance but also the total cost of ownership. Only then will it become clear whether an initiative is economically viable.
2. With every AI project, vendor lock-in increases
The second point is even more strategically significant. With many AI initiatives, companies are not only building new capabilities; they are also creating new dependencies:
- Price sensitivity due to variable rates, limits, and access conditions
- Technology lock-in caused by external model roadmaps
- Operational dependence on availability, latency, and quality
- Data dependency in sensitive contexts and knowledge bases
- Governance Dependency in Regulatory and Contractual Risks
Once AI is integrated into critical workflows, it is no longer just a procurement issue. It becomes an architectural issue. Companies that build their core knowledge, decision-making, or customer processes on a handful of external platforms are not merely outsourcing computing power. They are outsourcing part of their operational control.
This is precisely why the debate over AI strategies in companies is intensifying. Not because AI is becoming less relevant, but because its strategic implications are becoming more apparent.
3. Autonomous AI is evolving from a political concept into a management challenge
Sovereign AI is not an ideological alternative to innovation. It is a business response to rising costs, growing dependencies, and the desire to maintain control over one’s own intelligence stack.
In this context, self-sufficiency does not mean building everything yourself. Above all, it means:
- To maintain freedom of choice among models and providers
- to maintain control over critical data and knowledge assets
- Designing architectures so that components remain interchangeable
- Systematically incorporating governance and traceability
- To be able to actively manage costs, risks, and quality
This is particularly crucial in Europe. Data protection, traceability, and digital sovereignty are not peripheral issues here, but rather fundamental design requirements. What may seem like a hindrance during periods of hype can become an advantage in the next market phase.
The winners of the next wave of AI won’t be those who have implemented as many tools as possible. They will be those who can effectively manage costs, control, trust, and value creation all at once.
4. The ROI of AI must be calculated much more rigorously
A robust AI ROI is not simply a matter of comparing time savings before and after implementation. Companies must consider at least six dimensions:
- direct productivity gain
- Improved quality and better decision-making
- ongoing AI operating costs
- Costs of Human Oversight and Governance
- Lock-in and dependency costs
- Transformation costs for processes, roles, and organization
The bottom line is simple: AI only delivers a good ROI if it not only generates output but also improves workflow efficiency. If the costs of monitoring, integration, and vendor lock-in rise faster than the actual process benefits, the ROI is worse than any demo might suggest.
That is why the key question should not be: How effective is this model? It should be: Which process are we actually improving, what costs will be incurred over the lifecycle, and what new dependencies are we creating in the process?
5. What needs to replace the first AI roadmaps
When initial AI initiatives start to lose momentum, the solution shouldn’t be to simply collect even more use cases. Companies need a different operating model for AI.
- From Use Cases to Workflow Redesign
- From Tool Euphoria to Orchestration
- From Provider Convenience to Sovereignty by Design
- From short-term productivity to long-term profitability
- From AI Hype to AI Discipline
A sustainable advantage rarely stems from a single model. It arises from the orchestration of people, knowledge, models, rules, and systems. For many organizations, this is precisely where the real test of maturity lies.
If you’d like to explore this topic further, our article on the role of AI in the S/4HANA transformation also offers a practical look at its benefits, limitations, and organizational requirements.
Conclusion
If AI becomes too expensive for companies, that doesn’t mean AI is over. It means the next phase is beginning—moving away from symbolic innovation and toward cost-effectiveness, architectural discipline, and strategic control.
The future does not belong to companies that buy as many AI tools as possible. It belongs to companies that calculate their ROI realistically, actively manage their dependencies, and build a robust, trustworthy human-AI stack.
That is precisely where the real management challenge lies. And that is exactly what determines which organizations use AI productively and which ones merely create an expensive middle layer.
Anyone currently reassessing their AI stack shouldn’t start by talking about the next model. It makes more sense to take an honest look at costs, control, governance, and process impact.
