Apple Builds AI as Its Moat: The Golden Gate Strategy
Bernhard Liebl
8 Min. read time The real message of WWDC 2026 lies in the subtext of the Siri presentation. Apple is ...
Read Article
10 min read
9 Min. Reading time
One third of DACH companies with production‑grade AI projects report cost overruns compared with the original business case. The Bitkom 2026 study cites 33 percent. For management this is not a procurement quibble but a governance issue. Whoever does not establish a solid AI investment model by Q4 2026 will have the topic slammed in the investment committee – or flagged by the auditor in the audit‑report discussion.
Key Takeaways
- Bitkom 2026: 41 percent AI usage, 33 percent cost overrun, 19 percent job cuts justified by AI – three figures that reshuffle the C‑level conversation.
- The cost overrun is not a procurement failure but a governance failure. Companies that do not track AI investments with the same model as an ERP refresh or cloud migration are flying blind.
- The board question in 2026 is not “how much AI,” but “what value contribution per euro invested.” Without an answer, investment approvals will dry up.
- Three governance levers that work immediately: value‑contribution logic before tool selection, stage‑gates instead of annual approval, risk reporting to the audit committee as a standing agenda item.
- Strategic path dependence stems not from the model but from the data stack and supplier architecture. Those who ignore it early will be stuck in 2027.
What the Numbers Really Mean
The Bitkom 2026 study was released in April and provides the first reliable data set on productive AI use in German mid‑size firms and large corporations. 41 percent of surveyed companies say they have AI in production. Of those users, 33 percent report cost overruns against the original business case. 19 percent say they have cut jobs and attribute it to AI efficiencies.
Three figures, three narratives, one common thread: AI has reached the stage of an honest balance sheet. The hype curve has run its course, and reality checks are happening in corporate accounting. What started 18 months ago as an innovation initiative now sits in the investment committee and must prove its value contribution.
What is AI investment governance? AI investment governance is the separate leadership and reporting logic for initiatives with model‑ and data‑dependent value contribution. It differs from classic IT investment governance through stage‑gate approvals, value‑contribution allocation between IT and business units, and a dedicated risk‑reporting line to the audit committee. Without this separation, two investment classes with different scaling dynamics get mixed in a single report.
The 33 percent are not a procurement phenomenon. They are the symptom of a strategy gap. Treating the cost overrun as a supplier problem is tightening the wrong screw.
33%
of DACH companies with productive AI use report cost overruns versus the planned business case
Source: Bitkom Study 2026, April. Basis: Companies with AI in production in Germany.
Why this changes the governance logic, not the procurement calendar
Traditional IT investments follow a simple logic: requirement, budget, contract, implementation, acceptance. AI investments break that logic. The value contribution is created not during implementation but in ongoing operation – and it shifts with every model change, every data enrichment, every process adjustment in the business unit.
Anyone who runs an AI initiative like a licensing project ends up with a licensing‑project forecast. It lasts 14 months. Then the first real scaling step arrives. The model meets reality: more tokens than planned, new pipelines, additional business units, external data sources with their own licensing costs. The forecast was never wrong – it was built for the wrong investment class.
From a C‑level perspective this means: AI belongs in its own investment class with its own governance logic. Not as an IT cost line item, but as a scaling‑sensitive growth investment. Mixing the two logics in one report dilutes the insight.
Industry observers from the CIO community report consistently: the first cost‑overrun discussions with the CFO are not about model prices but about who books which value contribution. When sales uses an AI‑powered lead score and reports an 8 percent higher conversion rate, the effect flows into the sales P&L. The costs, however, sit with IT. Without value‑allocation, the investment committee sees only the cost side.
The twelve months that matter now
The next four quarters will decide whether AI adoption in DACH companies continues in a controlled manner or faces a full brake in the supervisory board. Four phases, four decisions.
Q2 2026
Inventory. Consolidate AI initiatives per business unit. Compare actual run costs against the approved business case. Incomplete pilot inventory is the main cause of cost‑overrun shocks.
Q3 2026
Embed governance model. Separate AI as its own investment class in reporting. Define stage‑gate criteria. Clarify value‑allocation between IT and the business unit before the Q3 review.
Q4 2026
Standardise risk reporting. Treat AI risks (data, model, vendors, compliance) as a standing agenda item in the audit committee. Embed EU AI Act obligations into reporting. Align the first materiality analysis with the external auditor.
Q1 2027
Strategic path decisions. Multi‑vendor versus concentration strategy. Own AI platform versus hyperscaler stack. Investor talks and supervisory board meeting in spring 2027 on AI value contribution and path dependency.
Three Levers That Carry Weight in the Executive Board Meeting
Anyone who wants to have a serious cost‑overrun discussion needs three levers – and the willingness to turn any of them against your favourite initiative when necessary.
Lever 1: Value‑contribution logic before tool selection. The question isn’t “which model” or “which vendor,” but “which decision should be improved and how much that improvement is worth per year.” Starting with the tool choice means you’ve already lost the proper sequence.
Lever 2: Stage‑gates instead of an annual approval. AI initiatives need phased approvals with a right to abort. If you can’t plausibly quantify the value contribution after 90 days, you won’t receive a second tranche. That prevents the costly 14‑month cost overruns that are now becoming visible.
Lever 3: Risk reporting in the audit committee. AI belongs in the same reporting stream as ERP risks or cloud risks – a standing agenda item in the quarterly review. Three indicators are enough: run‑costs versus plan, value contribution versus plan, supplier concentration. It’s lean but audit‑ready. Anyone looking for a reference framework for strategic anchoring will find it in the Merck decision story on Agentic AI.
“Honest AI discussions now happen between the CFO and the CIO, not between marketing and the innovation lead. Walk in without a clean value‑contribution logic and you’ll lose the next investment round – no matter how well the pilot performed.”
Tone from the CIO round‑table in the DACH mid‑market, spring 2026
The Path Dependence No One Shows on the Slide
Today’s cost overruns are tomorrow’s path dependence. Scaling AI initiatives in the first 18 months without a clear architecture decision creates lock‑ins that become expensive later. Three of them are so common they barely register any more.
First: data‑stack dependence. Models are interchangeable, training and context data are not. Building the training‑data workflow in a hyperscaler tool without cleanly specifying data extraction means you won’t switch later. That’s not a tech question; it’s an investment question.
Second: supplier concentration. Hyperscaler stacks look like smart consolidation in the first 12 months. By month 24 they become a single point of failure for your negotiating position. Several board members in the DACH region learned this during cloud migrations – the AI generation is repeating the mistake right now. Those who want to avoid it systematically can find the architecture‑decision logic in the analysis of three observations from the Constellation Enterprise Intelligence Report.
Third: skill concentration. When AI expertise sits in a central team that speaks only one platform’s toolchain, the switching‑cost corridor isn’t a licensing issue. It’s a people issue. This path dependence arrives with a time lag – but it arrives.
Ignore all of this and you’ll face a strategy problem on the table in 2027, not just a cost‑overrun problem. From a supervisory board perspective, the latter is uncomfortable; the former is liability‑relevant.
What Controlled Steering of Audit‑Stop Differs From
Controlled Steering
- AI as its own investment class in corporate reporting
- Value‑contribution allocation between IT and business clarified
- Stage‑gate approvals with documented termination rights
- Architecture decisions made explicit (data, suppliers, skills)
- Quarterly TOP in the audit committee
Path to Audit‑Stop
- AI booked as an IT cost line without a value‑contribution view
- Annual approval instead of phase‑gate release
- Pilot wild growth without central inventory
- Supplier concentration without explicit architecture decision
- Risks presented to the audit committee only on request
The dividing line is not speed. Both columns describe companies that take AI seriously. The dividing line is the question of whether steering is ahead of scaling or lagging behind. Those who lag will face a tough discussion in 2027 – with the auditor, the supervisory board or the investor, depending on ownership structure.
Who Should Put What on the Table Now
The CIO provides the inventory: which AI initiatives are actually running, what their operating costs are, and against which approved plan. Without this list, every further discussion is speculation.
The CFO brings the governance model: AI as its own investment class, stage‑gates, value‑contribution allocation. That is his original playing field – and the point at which most mid‑size corporate groups need to rethink the architecture of their reporting.
The CEO raises the strategic question: Where do we want to be in 2028? Which AI investments will get us there? Without this anchoring, every stage‑gate debate becomes a tactical squabble. With anchoring, it becomes strategic steering.
And the chairman of the supervisory board asks the honest question: Can we defend the investment approvals of the past 18 months to the investor? If the answer is “I don’t know,” the topic will be on the agenda by the spring meeting of 2027 at the latest. Better to have it there in a controlled way beforehand.
What’s Next on the List
The Bitkom figures are a snapshot. Upcoming surveys will show whether 2026 was the year DACH companies matured their AI governance – or the year the first CEOs were dismissed because of uncontrolled AI investments. Both scenarios are plausible. Which one materialises will be decided not by the IT roadmap but by the investment logic.
A complementary CFO perspective on the architecture and procurement side of the same Bitkom finding can be read in the MBF Media network: AI more expensive than planned: What the 33‑percent cost‑overrun rate means for mid‑size CFOs. Those who want to deepen the value‑contribution logic will find the strategic lines in the analysis AI Governance 2026 at System Level. And the data on productive AI use are expanded in the three figures from the Deloitte State of AI April 2026.
Frequently Asked Questions
How does AI investment governance differ from classic IT investment governance?
The AI value contribution is generated during ongoing operation and changes with data, model, and process. Classic IT investments have a fixed scope of delivery. Therefore AI requires its own investment class, stage‑gates instead of an annual approval, and a clean value‑contribution allocation between IT and the business unit.
Which three indicators are sufficient for AI risk reporting to the audit committee?
Run costs versus planned value, measured value contribution versus forecast, supplier concentration per application area. Narrow enough for quarterly reporting, audit‑ready for the external auditor, and meaningful enough for the supervisory board.
When does AI path dependency become liability‑relevant from a C‑level perspective?
When strategic options are lost without documented decisions. Supplier lock‑in or data‑stack dependency without an explicit investment justification is regarded in a dispute as a gap in due‑diligence. An early explicit architecture decision protects against this discussion.
Source cover image: Pexels / Vitaly Gariev (px:36713442)