According to Vlad Galabov, Omdia’s Research Director for Digital Infrastructure, substantial productivity gains across sectors fueled by AI will account for more than 50 % of the global data centre power and over 70 % of profit opportunities. Galabov made a number of other business estimates during Data Center World 2025’s scientist time:
- NVIDIA and hyperscalers ‘ goals of producing one MW per rack are unlikely to occur until engineering technology catches up to the demand for power and heating.
- Off-grid and behind-the-meter solutions are no longer recommended for those looking to build fresh data centers because many utilities struggle to provide the necessary energy because over 35 GW of data center power is anticipated to be self-generated by 2030.
- By 2030, data center annual capital expenditure ( CAPEX ) investments are projected to be worth$ 1 trillion, up from$ 500 billion at the end of 2024.
- Natural system, such as electricity and cooling, has the highest level of CAPEX spending, increasing annually by 18 %, according to spending trends.
The expenditure in physical infrastructure increases as compute density and plate concentrations rise, according to Galabov. We anticipate a smaller amount of scaled-up systems to replace a scaled-out client strategy in terms of server consolidation. The cost of a byte/compute pattern is also decreasing.
Energy in a data center erupts
Galabov cited the authority outages that AI has caused for info centers. Less than 150 GW of installed strength was present in data centers around the world when the AI storm first started in late 2023. He predicts nearly 400 GW of accumulated data center power by 2030, though 600 k cabinet styles are only about two years away and 120 kW cabinet designs are just beginning to appear. By the end of the decade, there will be almost 50 GW of new data centers ‘ annual power changes.
Not all did, however, survive the wild east of the DC and AI markets. Many new DC school developments and neoclouds fail to develop a long-term company model because some lack the necessary business acumen to live. Galabov warned against focusing on just one service because some are very unlikely to succeed.
More policy of NVIDIA’s Vision for AI Factory World 2025
Technology in liquid cooling is driven by AI.
Shen Wang, a principal analyst for Omdia, described the cooling effects of the AI flood. He claimed that heat heating reached its maximum in 2022. It maყ provide up tσ 80 W per çm2, with a feω manufacturers cIaiming ƫo raise the temperature of air heating.
Beyond that range, single-phase direct-to-chip ( DtC ) cooling, where heat is removed by placing water or other fluid on cold plates that are placed directly on top of computer chips, is required. Single-phase DC you reach 140 W/cm2.
The best way to great cards right now, according to Wang, is single-phase DtC. The most recent containers did have” by 2026″ exceeded the threshold for single-phase DtC. ” When implementation costs for two-phase wet cooling should start to increase. Fluids aɾe heated up in ƫhe device dưring two-phase cσoling, which causes them ƫo evaporate as vapor as tⱨey cool, increasing cooling performance.
The heaviest implementation of liquid cooling is seen in “advancers in the 600 ohm and over selection,” Wang said. ” By 2028, 50 % of cards in that category may use liquid cooling,” according to the report.
