The AI will lead to more than 50% of the capacity of the global data center and to more than 70% of income opportunities, according to OMDIA's research director for the digital infrastructure Vlad Garcové, who said that massive productivity gains in all sectors piloted by the AI will feed this growth. Speaking during the day of the World 2025 analysts, Garlalabo has made a number of other predictions on the industry:
- The ambitions of 1 MW per Rack of Nvidia and Hyperscalers will probably not materialize yet a few years until engineering innovation catches up with power and cooling requests.
- By 2030, more than 35 GW of the energy of the data center should self-generate, which makes solutions outside the network and behind the meter which are no longer optional for those who seek to build new data centers, because many public services have trouble providing the necessary power.
- Annual capital expenditure (CAPEX) of the data center should reach $ 1 billion worldwide by 2030, compared to less than $ 500 billion at the end of 2024.
- The strongest area for CAPEX is physical infrastructure, such as power and cooling, where spending is increasing at a rate of 18% per year.
“As the calculation densities and Rack densities are climbing, investing in physical infrastructure is accelerating,” said Garlalavé. “We expect a consolidation of the number of servers where a small number of systems on a scale are preferred to a server strategy on a scale. The cost per byte / calculation also decreases. ”
The energy center capacity explodes
Garlalavé highlighted the explosion that AI has caused the power needs of the data center. When the AI wave started at the end of 2023, the power capacity installed in the world data centers was less than 150 GW. But with conceptions of 120 kW racks on the immediate horizon and 600 kW racks at only two years, it provides almost 400 GW of capacity of cumulative data center by 2030. With new additions of data center capacity approaching 50 GW per year by the end of the decade, it will not be long before the semi-terawatt became the standard.
But not everyone will survive the Far West of the AI and DC market. Many developments and neoclouds on the DC Startup campus will not be able to build a long-term business model, because some do not have the expertise and know-how of companies to survive. Do not focus on a single supplier, has warned Garlabov, as some are likely to fail.
More cover of Data Center World 2025: Nvidia's vision for AI factories
AI causes liquid cooling innovation
OMDIA's main analyst Shen Wang presented the cooling repercussions of the AI wave. Air cooling reaches its limit around 2022, he said. The consensus is that it can deliver up to 80 watts per CM2, with some suppliers claiming that they can take over in air higher.
Beyond this beach, direct monophase chip (DTC) – in which water or liquid is transported in cold plates which are directly on computer fleas to eliminate heat – is necessary. The single -phase DTC can go up to 140 W / CM2.
“The single -phase DTC is the best way to cool the fleas right now,” said Wang. “By 2026, the Monophase DTC threshold will be overwhelmed by the latest racks.” It is at this time that the liquid cooling in two phases should start to see an increase in adoption rates. Biphase cooling performs fluids at higher temperatures at the chip, making them turn to steam within the framework of the cooling process, thus increasing the cooling efficiency.
“The chips put forward in the 600 watts and the higher scope of the heaviest adoption of liquid cooling,” said Wang. “By 2028, 80% of fleas in this category will use the liquid cooling, compared to 50% today.”
