News linked to both this project and an event.
Odaily Odaily News: FUNDA posted on X platform that North American AI data center demand in 2027, measured by IT load, is approximately 35GW, while actual deliverable supply is only 16.5 to 23.4GW. The overlapping range of two independent supply projections shows a gap of 1.8 to 2.1 times between demand and deliverable supply, which serves as the core of this report.Demand was cross-validated using two methods, yielding similar results. Based on CoWoS capacity build-out projections, global chip-level electricity demand in 2027 is estimated at 44 to 49GW; based on platform-by-platform bill of materials projections, the result is 48.4GW. The current market discussion range is 40 to 60GW, and the report estimates global demand at approximately 50GW, with North America accounting for about two-thirds.Supply is largely locked in, and capital investment cannot change this. Grid interconnection pathways can deliver 11 to 14GW of IT load, depending entirely on which projects enter the interconnection queue before mid-2025; applications submitted today cannot contribute to 2027 supply. Behind-the-meter installations can add 5.5 to 9.4GW of IT load after accounting for overlap, but large gas turbine orders are already booked through 2031. Since it typically takes 4 to 5 years from order placement to commercial operation, turbines ordered today will only come online between 2030 and 2032.Power delivery is the binding constraint and sits upstream of all links that project developers can accelerate through capital spending. Permitting hurdles prevent 78% of the 43GW of shelved projects from moving forward. Over the past 12 months, power-related factors have caused 60% to 70% of project delays on a megawatt-weighted basis. Next is the shortage of certified labor: skilled electricians require 10,000 hours of training, and only about 30% of the current electromechanical, piping, and HVAC workforce is located in regions hosting 70% of projects. Modular construction can compress build timelines but cannot shorten queue times, only reducing the probability of commercial operation delays by 10% to 20%.Scarcity is already reflected in pricing. Annual recurring revenue per GW across the four comparison companies ranges from $8.3 billion to $50 billion, and NBIS has signed contracts over an 18-month period at prices 3 to 4 times its existing installed base unit pricing. The significant repricing in Q1 and Q2 has concluded, and the report expects market stabilization with modest increases starting in Q4 2026. As new supply comes online, rents on a generation-by-generation basis are expected to decline by approximately 20% to 30%, while utilization rates are expected to remain high.The four power procurement pathways correspond to four different risk profiles. xAI trades speed for doubled power costs and permitting risk, holding the shortest contract duration among the four companies; CRWV converts power delivery risk into refinancing risk; NBIS operates on customer prep
SemiAnalysis stated on X platform that traditional backup power equipment for data centers——Reciprocating Engines——are transitioning from an "emergency backup" role to a primary round-the-clock power source to meet the massive electricity demand driven by AI computing expansion.SemiAnalysis points out that in the past, data centers typically only activated on-site reciprocating engine backup power during grid failures, operating for just a few hours annually. However, with increasing grid capacity constraints, more data centers are repurposing these devices for Prime Power, enabling continuous electricity supply.According to SemiAnalysis' energy model estimates, manufacturers of reciprocating engines, including Caterpillar, INNIO, and Cummins, are expected to have signed approximately 1 GW in Behind-the-Meter (BTM) project orders this year; order volumes for 2027 and 2028 are each projected to exceed 4 GW.SemiAnalysis notes that current order volumes are still very limited compared to future potential demand. After modeling US grid constraints, they found that existing power supply margins are expected to be exhausted between 2027 and 2028, and planned utility-scale new generation capacity through 2030 is insufficient to meet the growing data center load demand.Based on its data center model, SemiAnalysis estimates that approximately 140 GW of identified data center projects still need to sign power supply contracts and may adopt behind-the-meter power supply models.Among the technologies available for behind-the-meter power supply, including reciprocating engines, aeroderivative gas turbines, and fuel cells, SemiAnalysis expects reciprocating engines to capture the largest share.The institution believes that reciprocating engines offer the best balance between cost and deployment speed, while also providing modular advantages that support phased data center expansion. Compared to fuel cells, they boast higher commercial maturity and financing feasibility. As manufacturers ramp up production capacity in the coming years, reciprocating engines are poised to become critical infrastructure for AI data centers to solve power bottlenecks.SemiAnalysis states that AI computing growth is driving data centers to shift from relying on traditional grid power to a hybrid model of "grid plus self-owned energy," with energy infrastructure emerging as a new core constraint for AI industry expansion.