Cloud sovereignty becomes a boardroom issue: What the EU tech sovereignty package means for DACH
Tobias Massow
6 min read The EU unveiled its Tech Sovereignty Package on 27 May. It proposes restricting the use of ...
Read Article
7 min read
On 20 May 2026, Nvidia CEO Jensen Huang dropped a number during the Q1-FY2027 earnings call that every German corporate IT department needs to take note of. Annual AI capital expenditures by the hyperscalers are projected to rise to between €2.6 and €3.4 trillion by 2030, up from roughly €860 billion today. Any CIO or CTO negotiating a three-year roadmap today is effectively locking in terms against a market that, within five years, will shoulder three to four times the infrastructure investment.
Key Takeaways
- Huang’s figure is concrete-and CFO-verified. Jensen Huang cites €2.6–3.4 trillion in AI CapEx by 2030; CFO Colette Kress even calls the target achievable before 2030-this is earnings guidance, not marketing fluff.
- Vendor power shifts, not negotiating leverage. With €2.6–3.4 trillion in hyperscaler CapEx, corporate IT becomes a rounding error in pricing models. Cloud-contract consolidation grows costlier, not cheaper.
- Three decisions in 2026 are unavoidable. Decouple compute procurement, pull talent to the interface, and price sovereignty costs honestly. Those who sit on the fence will foot the full sprint bill in 2028.
Related:Microsoft’s AI bet gobbles CIO budgets / Compute becomes the scarce production factor
What €2.6–3.4 trillion in AI CapEx means
What is AI CapEx? AI CapEx (Artificial Intelligence Capital Expenditure) refers to the annual capital investments made by large cloud providers and hyperscalers in AI-specific infrastructure-GPU clusters, data-centre construction, power supply, network backbones, and proprietary accelerator chips. Unlike traditional IT spend, these are long-lived physical assets with depreciation periods stretching six to twelve years.
The current baseline is documented in the earnings call. Hyperscaler CapEx in 2026 totals roughly €623 billion, of which Nvidia attributes €468 billion directly to AI infrastructure. Translate that to the DACH market and you see: today’s hyperscaler cohort already invests more in a single year of AI infrastructure than the combined annual IT budgets of all DAX-40 companies-repeatedly.
The headline doubling isn’t the story. What matters is that Nvidia CFO Colette Kress confirmed in the same call that the €2.6–3.4 trillion could arrive before 2030. CIOs signing three-year contracts with AWS, Azure or GCP in 2027 are therefore not entering a stable market-they’re locking in terms during a period of aggressive consolidation and rapidly rising vendor market power.
The Numbers from the Q1-FY2027 Call
On 20 May 2026, Nvidia reported record figures: €70.1 billion in quarterly revenue, up 85 percent year-on-year. Of that total, €64.6 billion came from the data-centre segment, a 92 percent increase. These numbers are not just Nvidia’s success; they reflect the hyperscaler CapEx wave in full force.
2.6–3.4 trillion €
Annual AI capital expenditures by the hyperscaler cohort through 2030. Current tally, per Nvidia, is roughly €860 billion.
Source: Jensen Huang, Nvidia Q1-FY2027 earnings call, 20 May 2026; confirmed by CFO Colette Kress.
Huang’s choice of words in the call will echo in boardrooms for months. Building AI factories, he said, is the largest infrastructure expansion in human history-and it is accelerating. Those who follow Huang know this is no rhetorical flourish; it is the slide deck his CFO will use to calibrate investor expectations in the coming quarters.
The second metric that matters more to CIOs than the headline figure comes from the compute-growth slide. Global AI compute capacity is projected to scale at a 2.25× annual rate through 2030. Translated into monetary terms, that implies average annual growth of 32–41 percent. For CIOs planning compute budgets today, the market is one in which the largest players double their needs roughly every 2.5 years.
When the CapEx Wave Hits DACH
The €2.6–3.4 trillion figure sounds abstract. To make it tangible, look at the timeline of hyperscaler investments, from the first AI wave to today. Three milestones should be top of mind for DACH CIOs.
CIOs entering 2026 expecting cloud prices to stabilise on the horizon are misreading the curve. Hyperscalers keep building because they must. Yet they are building for workloads that do not yet exist. In this phase, prices per token or per compute-hour do not automatically fall; they only decline where providers face real competition-and with today’s concentration, that space is shrinking fast.
Three strategic priorities for DACH CIOs by 2026
Huang’s figure yields three priorities for the next budget cycles. They’re not headline-grabbing, but they’re exacting because they break established governance structures.
First: decouple compute from the main negotiation. If you continue to bundle cloud workloads and AI compute in the same master agreement, you surrender negotiating power. A better route is to source AI compute through separate vehicles-neo-cloud providers, dedicated inference platforms, or in-house GPU clusters. This lets you negotiate the compute slice independently of the overall cloud contract and re-price it more frequently than a three-year framework.
Second: pull top talent to the interface. Senior tech talent who simultaneously grasp compute economics, model tuning, and vendor negotiation is scarce today and will be a strategic resource through 2028. Waiting to hire until an AI project fails is too late. These profiles belong in the IT strategy team, not in a siloed innovation unit.
Third: price sovereignty honestly. The push for European cloud sovereignty operates in a trillion-euro world and is operationally more expensive. Choosing BSI-C5 or GAIA-X options carries a premium that grows structurally as hyperscalers scale. The question isn’t whether sovereignty matters, but where it truly mitigates risk and where it’s a politically driven surcharge. Both answers are valid, but they must be quantified.
What the figure does not imply
Two misreadings are common. The first claims that €2.6–3.4 trillion in AI CapEx will make token-level compute cheaper. The opposite is likely. Hyperscalers build for peak models and markets where premium pricing is feasible. Inference costs at the mainstream-model frontier will keep falling, but more slowly than in 2023–2025. Planning on another annual halving is betting against the providers’ earnings logic.
The second misreading is that this CapEx wave will automatically spawn European sovereignty alternatives. That won’t happen. Europe-specific vendors will occupy niches-especially in regulated sectors and the public sector-but they won’t match the volume curves of U.S. hyperscalers. If you’re banking on a European counter-model, treat it as a lobbying or grant thesis, not a procurement strategy.
What remains is a clear message for DACH boards. The €2.6–3.4 trillion is not a marketing number. It’s a market setup in which corporate IT must clarify its stance within the next 18 months-or else it will do so by default contracts.
Frequently Asked Questions
When did Jensen Huang make the trillion-euro statement?
During Nvidia’s Q1-FY2027 earnings call on 20 May 2026. CFO Colette Kress added in the same call that the figure could be reached even before 2030.
What exactly does AI CapEx mean compared to traditional IT CapEx?
AI CapEx covers capital expenditures for AI-specific infrastructure: GPU clusters, AI data-centres, power delivery, high-speed networks. Unlike classic IT CapEx, depreciation periods are markedly longer and far more tied to proprietary hardware roadmaps.
How high are current hyperscaler CapEx figures for 2026?
According to Nvidia, total hyperscaler CapEx for 2026 is about €623 billion, of which €468 billion is directly earmarked for AI infrastructure. The AI share doubles in magnitude within two years.
What three decisions should a DACH CIO make right now?
First, decouple AI compute from cloud framework agreements and source it through separate vehicles. Second, build senior tech talent at the intersection of compute economics, model tuning and vendor negotiation. Third, quantify sovereignty costs honestly instead of treating them as political line items.
Will cloud prices fall as a result of this CapEx wave?
Not automatically. Hyperscalers are investing in premium tiers and flagship models. Inference costs at the mainstream-model frontier will continue to decline, but more slowly than between 2023 and 2025. Anyone banking on annual halvings is betting against the providers’ earnings logic.
Editor’s Reading List
More from the MBF Media Network
Source of cover image: Will Buckner / Wikimedia Commons (CC BY 2.0)