報告_OCP2026_McKinsey_資料中心價值鏈_20261015

PDF 原檔：報告_OCP2026_McKinsey_資料中心價值鏈_20261015_original.pdf

講者公開簡報（OCP 官方 Recorded Content 頁提供下載）。文字以 PyMuPDF 逐頁抽取，### p<n> 為頁碼；圖表內容見原檔。

圖片清單（已驗證 2026-10-19）

原始內容

p1

Maria Goodpaster Partner, McKinsey & Company Elia Berteletti Partner, McKinsey & Company Denise Lee Partner, McKinsey & Company From Compute Demand to Capacity Delivered Unlocking the Data Center Value Chain Through OCP-Aligned Design and Supply Chain Innovation MARKET IMPACT & ADOPTION ]

p2

Global DC demand is projected to reach ~336GW by 2035 with an inflection in growth rate in ~2030 Source: McKinsey Proprietary datacentre demand model 171 254 219 336 472 2025 2026 2027 2028 2029 298 2030 2031 2032 2033 2034 2035 2.8x 3.7x 5.2x Takeaways ~3-5x demand is forecast to 2035 due to increase in compute requirements driven partly by generative AI and machine learning ~15-30% CAGR 2025-2035 can be largely attributed to major cloud providers (e.g., AWS, Google, Meta) expanding their data centre footprints to support cloud computing and streaming 65-114GW of capacity will be in the US alone with 75% of the share of total capacity driven by AI workloads in 2030 in the U.S. Conservative adoption Base case adoption Accelerated adoption Key Assumptions Accelerated adoption: GenAI adoption grows beyond current expectations; industry avoids supply chain shortages Base case adoption: GenAI growth continues without severe bottlenecks in the mid- and long- term Conservative adoption: Adoption rate constraint e.g., by supply, regulation and concern of data privacy CAGR 25-30 CAGR 30-35 29% 10% 22% 9% 16% 8% 2030-35 assumption that GenAI growth follows an S curve Estimated Global Data Center Demand, in GW IT compute, 2025 - 2035

p3

2025-2030 capital investments across data center value chain segments ($, T) Still in early innings – we anticipate $7.3 T to be cumulatively deployed through 2030 Real estate, design & construction Network & connectivity Power Cooling & electrical IT hardware & rack innovation Chips & components Data Center Software Services Hyper-scalers Colocation providers GPU-as-a- service providers Builders Energizers Designers & Manufacturers Operators & End Users 3 McKinsey & Company ~$1.0 T Builders: Cumulative CAPEX 2025- 2030 ~$1.3 T Energizers: Cumulative CAPEX 2025- 2030 ~$4.4 T Designers & manufacturers: Cumulative CAPEX 2025- 2030 ~$7.3 T (~1% of global GDP/year1) ~$0.6 T Software: Cumulative CAPEX 2025-2030 ALL FIGURES ARE GLOBAL Source: McKinsey Data Center CAPEX TAM & Demand Model, Expert interviews, Expert interviews, CapIQ, Goldman Sachs, Analyst reports 1. Global GDP: $106T (2023) | 2. Includes MEP Note: Capital investments by Operators & End users were not included in this analysis ~$0.6 T Labor ~$0.3 T Shell & Site2 ~$0.1 T Land Acquisition ~$0.8 T Electrical & Mechanical Equipment ~$0.4 T Power Generation ~$0.1 T Network Infra ~$3.5 T Servers (incl. GPUs & CPUs) ~$0.8 T Storage ~$0.4 T Stacking layers ~$0.2 T Overarching layers

p4

Power | Power is the single biggest barrier to data center growth, and the industry is responding with onsite generation at scale ~100 GW by 2030 Industry experts project U.S. data center demand will reach ~100 GW by 2030, but only 65% of announced capacity is expected to come online Grid access is the bottleneck Interconnection timelines, transmission constraints, and equipment lead times are the top three barriers, with energization taking up to 4-5 years 65% deploying onsite power Most Data Center players are turning to onsite generation, with ~60% of that capacity expected to be permanent, not temporary bridge Reliability and speed over cost Power reliability and time to power come top of the list in siting, onsite power selection and utility selection, and both command the highest willingness to pay premiums (~15%) Natural gas is the near- term preferred solution Natural gas and renewables are the most preferred onsite generation technologies today, with gas technologies (turbines, engines, fuel cells) expected to hold ~52% of 2030 onsite share Clean energy ambition outpaces execution Most players target ~50% clean energy by 2030, but only 27% are confident they'll hit it, and willingness to pay for clean attributes is the lowest of any power attribute at 9.3% premium The problem What it means Power

p5

US supply gap| Expected to widen to ~15GW in 2030 driven by time-to-power constraints 37 49 65 82 101 121 36 47 59 73 89 106 2025 26 27 28 29 2030 Demand: 27% p.a. Supply: 24% p.a. Demand Supply US Data Center Supply and Demand, GW X% Demand supply ratio 101% 105% 109% 113% 113% 115% Key Insights US demand is expected to outpace supply by 3 p.p. from 2025 to 2030, resulting in a 15GW supply shortfall by 2030 Price premiums are expected to persist, driven by this sustained structural imbalance Demand gap widens to 113% in 2028 when demand spikes (i.e., server shipments forecasted to peak as technology cycles mature) and gap expected to persist to 2030 Capacity unable to meet demand due to construction-related challenges (e.g., time to power, community pushback, delays in securing MEP equipment, labor shortages) Data center demand and key trends Source: McKinsey Proprietary Data Center demand and supply model

p6

Equipment lead times | Lead times have increased by 2-3x for critical equipment (e.g., generators, chillers, etc.) Generator UPS Transformer1 Switchboard Chillers ~30 ~100 ~24 ~28 ~15 ~30 ~20 ~33 ~20 ~75 2019 2025 Generators and chillers have more than tripled their lead times vs. 2019 (~100 and ~75 weeks, respectively) Lead times are longest for colocation and enterprise data centers that: Do not use standardized equipment Have no strategic supplier base Utilize project-based procurement as opposed to centralized/portfolio procurement process Lead time of major data center critical equipment (in weeks) NOT EXHAUSTIVE Source: Press search, Expert Interviews, McKinsey analysis 1. Medium Voltage

p7

The current approaches make an impact but are not enough to resolve the supply chain constraints Building bottom-up cost baselines for various types of greyspace and whitespace equipment to quantify impact of standardization Should-cost modeling End-to-end supply chain optimization Rethinking each step involved throughout the integration, logistics and assembly points Identifying and qualifying credible new entrants against open specs, broadening a supply base too concentrated to meet projected 2030 demand New supplier market entry Working on helping with manufacturing throughput and debottlenecking to compress lead times and enable delivery Supplier capacity ramp-up

p8

Bespoke specifications amplify every bottleneck; OCP standardization breaks the cycle From… To… Each operator qualifies independently, creating months of redundant testing Power and cooling require custom rack designs, reducing repeatability and extending lead times Equipment is procured project by project, limiting purchasing scale IT and facility require unique interfaces, slowing reuse and commissioning Open specifications enable shared validation, eliminating redundant testing Standardized power and cooling designs improve repeatability and shorten lead times Standardized components enable purchasing scale and stronger supplier economics Standardized interfaces between IT and facility enable reuse and faster commissioning The result: OCP-validated, standardized deployment can potentially shorten lead times by 3-61 months and enable faster scale The result: Fragmented, bespoke deployments extend lead times by 6-12 months 1. Lead time reduction via standardization will vary by component type (e.g., power gen vs. chillers)

p9

How can you be involved? Focus community effort on the interfaces that limit deployment speed Domain OCP Initiative Expected outcome Power Power Distribution Faster rack deployment Cooling Cooling Environments Liquid cooling interoperability Rack Rack and Power Fungible AI racks Network Open Cluster Designs for AI Portable AI clusters Grid Open Data Centers for AI Standardized power integration

p10

（本頁無可抽取文字，內容在圖中）