Marrakech – Artificial intelligence is forcing a redesign of the infrastructure behind data centers. At Vertiv Week 2026 in Zagreb, executives argued that power, cooling, and software can no longer be planned as separate systems.
The event ran from September 21 to 25 in the Croatian capital under the theme “Converged Infrastructure for the AI Era.” It brought together European data center associations, journalists, consultant engineers, industry experts, and Vertiv leaders from across Europe, the Middle East, and Africa (EMEA).
The program ran on three tracks. The European Data Center Associations Summit (September 21-23) gathered industry bodies to discuss policy, sustainability, and AI-driven demand.
The media and analysts conference, “Vertiv Driving Innovation,” followed from September 22 to 24 with NVIDIA, operators, and associations. The Engineers’ Frontier Master Class, held September 23-25, was open only to consultant engineers designing infrastructure for AI factories, cloud environments, and enterprise deployments.
The previous edition took place in 2025 at Vertiv’s Tognana site in Italy.
NVIDIA charts an inference-driven surge
The central demand forecasts came on September 23 from Rod Evans, NVIDIA’s vice president for AI cloud infrastructure. He described AI as driving “the largest infrastructure buildout in human history.” In his account, the industry is “only a few hundred billion dollars into it,” with “trillions more” still to be built.
Citing McKinsey, Evans presented forecasts showing global data center IT demand rising from 60 gigawatts (GW) in 2023 to 219 GW in 2030. That is an annual growth rate of 22%. AI workloads would account for 71% of that demand by the end of the decade.
Inference, the process of running trained models, is expected to drive much of this growth. NVIDIA’s figures show AI inference demand reaching 93 GW by 2030, or up to 60% of all AI compute.
In Europe, most compute demand already comes from inference, which has to run close to users and their data. The presentation added that AI adoption in most European countries exceeds US levels. It also noted that the EU houses more than 7,000 companies active in the global AI race.
AI agents are adding to that demand. Goldman Sachs estimates cited by Evans project that token use by AI agents will multiply 24 times by 2030, reaching 118 quadrillion tokens per month.
Google’s own monthly token volume climbed from 9.7 trillion in May 2024 to 3.2 quadrillion in May 2026. Gartner data showed that 80% of enterprise applications shipped or updated in the first quarter of 2026 embed an AI agent, up from 33% in 2024.
Energy efficiency remains a major gap, according to NVIDIA. In a 1-GW AI factory, only 60% to 70% of grid power contributes to AI output. Roughly 20% is lost to facility power and cooling inefficiency. Another 10% goes to inefficient rack power, and a further 10% to failures and restarts.
To close that gap, NVIDIA is moving to direct liquid cooling and from AC to DC power. The company stated that its liquid-cooled rack systems deliver 300 times higher water efficiency than air-cooled Hopper systems.
A joint NVL72 reference architecture with Vertiv is projected to cut annual cooling energy by 20%. It is also projected to use 40% less rack space, deploy 50% faster, and require a 40% smaller power footprint.
The two companies are also developing Vertiv’s 800-volt high-voltage direct current (HVDC) architectures. By 2028, NVIDIA plans to move toward racks approaching 1 megawatt.
Vertiv shifts from field to factory as investment climbs
Vertiv’s case focused on how quickly that infrastructure can be delivered. Paul Ryan, Vertiv’s president for EMEA, identified five forces reshaping data center design: density, speed, scale, complexity, and dynamic load profiles. His presentation stated that “traditional field integration cannot keep pace,” and that customers want validated, repeatable infrastructure.
The company measures AI factory performance through four metrics: tokens per second, time to first token, tokens per watt, and tokens per dollar. Its strategy centers on moving assembly “from field to factory.” Under that model, converged systems such as Vertiv OneCore and Vertiv SmartRun are built and tested in factories before shipping.
Vertiv is also extending its reach toward power supply. On September 2, it announced an agreement to acquire UtilityInnovation Group, a provider of microgrid solutions and behind-the-meter power architecture. The deal carries about $1.45 billion in cash at closing. Up to $1.15 billion more is tied to EBITDA targets. The transaction is expected to close in the fourth quarter of 2026, pending regulatory approvals.
Vertiv CEO Giordano Albertazzi tied the deal to speed. For AI operators, he stated, competitive advantage “increasingly depends on how quickly they can move from site selection to first token.”
Manufacturing capacity is growing at the same time. Vertiv is adding about 22,000 square meters of production space in Nové Město, Slovakia. The site will produce power, switchgear, and thermal management systems, including liquid cooling for high-density AI. The company expects it to create hundreds of new jobs.
In EMEA, Vertiv operates 10 manufacturing locations, including one in Rugvica, Croatia. The region also has about 60 service centers and roughly 700 field engineers. Globally, Vertiv reported about $10.2 billion in 2025 revenue and around 34,000 employees serving more than 130 countries. Data centers account for 85% of that revenue.
Market analysts provided figures on the size of the opportunity. Gautham Gnanajothi, senior vice president for the data center vertical at Frost & Sullivan, valued global data center investment at $502.7 billion in 2025. The firm forecasts it will reach $1.457 trillion by 2035, a compound annual growth rate of 11.2%.
The Americas held 46.3% of 2025 investment, followed by Asia-Pacific at 29.2% and EMEA at 24.5%. Asia-Pacific is expected to grow fastest through 2035. In EMEA, colocation investment is forecast to grow 13.2% a year and hyperscale investment 7.2%.
Hyperscalers remain the main source of spending. Combined data center capital spending by Amazon, Google, Microsoft, and Meta rose from $85.7 billion in 2023 to $280.5 billion in 2025. Frost & Sullivan projects it will reach $630 billion in 2026. AI workloads accounted for 77% of hyperscale investment in 2025.
Frost & Sullivan’s analysis treated compute and energy as interdependent resources. In its framework, access to power now limits AI growth as much as access to chips does, and “AI policy = Energy policy.”
Read also: Morocco Signs MoU with Vertiv to Advance Cloud, AI, Digital Infrastructure
