The AI divide between activity and impact
The hype around generative AI has given way to a sober question. Where does measurable value actually come from? The 29th Annual Global CEO Survey by PwC, which polled 4,454 CEOs in 95 countries from 30 September to 7 November 2025, 93 of them in Germany, gives a sobering answer (PwC 2026). Only 30 percent of CEOs worldwide are confident they can grow revenue over the next twelve months, down from 38 percent the year before (PwC 2026). And only 12 percent managed to both cut costs and generate additional revenue from AI in the past year (PwC 2026).
The pattern is clear. AI is in use almost everywhere, yet the economic return is concentrated in a few hands. A complementary study by MIT finds that roughly 95 percent of organisations record no measurable contribution in their profit and loss statement despite significant pilot budgets (MIT, Project NANDA, 2025).
2026 is shaping up as a decisive year for AI. A small group of companies are already turning AI into measurable financial returns, while many others are still struggling to move beyond pilots. That gap is starting to show up in confidence and competitiveness, and it will widen quickly for those that do not act. (Mohamed Kande, Global Chairman PwC, 2026)
Germany is falling behind
The gap is especially clear in Germany. Only 11 percent of German CEOs report additional revenue from AI, compared with 29 percent globally. On cost reduction, Germany also trails the global average at 16 percent versus 26 percent (PwC 2026).
Striking is the contradiction with self-perception. 74 percent of German CEOs consider their technology environment and culture fit for purpose, globally 67 and 69 percent respectively, yet results fail to materialise (PwC 2026). The German Economic Institute offers an explanation. Many companies overestimate their maturity, and only about 6 percent use AI across several business areas at once (IW Köln, 2025). A single chatbot is mistaken for transformation.
The foundation decides, not the tool
The most important finding of the study is also its most hopeful. Success is plannable. Companies that are strong in at least five of six core AI foundations, such as data availability, targeted investment and talent development, are 2.3 times more likely to grow revenue and 1.7 times more likely to cut costs (PwC 2026).
Foundations matter as much as scale. CEOs whose organisations have established strong AI foundations are three times more likely to report meaningful financial returns. (Mohamed Kande, Global Chairman PwC, 2026)
In concrete terms, AI does not succeed through yet another tool, but through strategy, a clean data architecture, clear governance and people who master both. Anyone who layers AI onto broken processes mainly scales the errors. That is precisely why you need expertise and a strategy before the first model goes into production.
What this means for your company, and how we help
This is exactly where our AI consulting and implementation come in. We combine strategy, data architecture and delivery rather than shipping isolated pilots. The choice of partner matters. According to MIT, only about 5 percent of internally built custom solutions reach productive scale, while external partnerships show roughly twice the success rate (MIT, Project NANDA, 2025).
We ourselves build and operate dozens of adapted agentic systems across different domains. Here are a few examples from our practice.
- In quote and invoice processing, agents extract documents, reconcile them with the ERP and pre-post entries.
- In customer service, RAG-based assistants answer questions from your own, vetted knowledge base.
- In procurement and supply chain, agents monitor stock, detect shortages and trigger orders based on rules.
- In manufacturing, process mining and machine learning work together to create a clear understanding of the production process. This makes it possible to identify the relevant influencing factors and shows in what way and to what degree they shape output quality.
- In contract and compliance, models check documents against policies and flag risks for human approval.
- In marketing and content, product data is turned automatically into consistent, search-optimised copy.
Notably, many companies misallocate their budget. Up to 70 percent flows into visible front-office applications, while the high-volume back-office processes with the higher and more scalable return stay underfunded (MIT, Project NANDA, 2025). That is where the sustainable ROI sits, in procurement, finance and operations, and that is where we focus.
Deliberately reducing dependence on big tech
A second strategic lever is often overlooked, namely reducing dependencies. 58 percent of German companies source most of their digital technology, including software, AI and cloud, from the United States, while European solutions account for just 27 percent (PwC 2026). A good third of German CEOs, namely 36 percent, therefore plan to rely more on EU providers over the next three years (PwC 2026).
The concentration is dramatic in infrastructure too. Around 80 percent of the global compute capacity for AI sits in the United States, while Europe accounts for roughly 5 percent (industry estimates, 2025). According to Deloitte, 62 percent of German executives say that more than 40 percent of their AI infrastructure is under the control of foreign providers (Deloitte, 2025).
This dependence is a concrete business risk. Licence and API prices fluctuate, models and terms change suddenly, data leaves your legal jurisdiction and a lock-in erodes all negotiating power. Even PwC notes that European solutions can be a complement, but not a full replacement (PwC 2026). All the more reason for a deliberate architecture that brings critical capabilities back in house.
Local models, open source and your own AI hardware secure the strategy
This is where the circle closes. An AI strategy only becomes durable when it rests on a foundation the company owns itself, namely local and fine-tuned models, open-source solutions and AI hardware operated in house.
- Sensitive data never leaves your premises and stays sovereign. This meets GDPR requirements and builds trust with customers and regulators.
- Running your own models on your own hardware makes you independent of API prices and term changes, and lets you calculate ROI reliably.
- Open-source models cannot be switched off, repriced or unilaterally changed. Availability stays in your hands.
- Specialised, smaller models handle many tasks more efficiently than giant generalists and thereby cut energy use and cost. This is a relevant factor given a projected electricity demand of German data centres of 21.3 billion kilowatt hours for 2025 (Bitkom, 2025).
- With the EU AI Act, whose obligations for high-risk systems take effect from August 2026, traceable and controllable AI becomes a requirement. Owned models and clear governance keep audits manageable.
Only this combination of local models, open source and your own AI hardware ensures a truly sustainable use of AI. Anyone who wants to create value over the long term cannot build that value on rented, externally controlled foundations. Only ownership of model, data and compute makes AI plannable, secure and viable both economically and ecologically.
Lay the foundation now
The message of the 29th Global CEO Survey is unambiguous. The era of pure experimentation is over, and measurable success goes only to those who integrate AI strategically and across the enterprise (PwC 2026). The window is narrow, but open. Whoever lays a solid foundation now, builds expertise and secures sovereignty will not only defend their margins but gain a lead that hesitant competitors will struggle to close.
We support you on exactly this path, from strategy through data architecture to the operation of sovereign, agentic systems. Talk to us.
