AI算力需求爆發正重塑資料中心投資版圖：2025年全球資本支出將超兆美元，微軟、亞馬遜等巨頭單年砸下數百億美元建智算中心，中國「東數西算」疊加新能源優勢，年內智算項目已超300個，投資規模近千億元。液冷、核能、模組化與REITs平行，綠色、高密度、可擴展成為資本追逐的新標竿，資料中心正從成本中心升級為AI時代的核心資產。人工智慧正催生對算力的旺盛需求，促使企業投入數十億美元用於基礎設施建設。然而，由於未來需求的不確定性，投資人需謹慎決策。AI is fueling high demand for compute power, spurring companies to invest billions of dollars in infrastructure. But with future demand uncertain, investors will need to make calculated decisions.在人工智慧熱潮下，算力正成為本世紀最關鍵的資源之一。在全球各地的資料中心，數百萬台伺服器全天候運轉，處理支撐人工智慧的基礎模型與機器學習應用。這些資料中心所需的硬體、處理器、記憶體、儲存和能源共同構成算力，而市場對算力的需求似乎永無止境。Amid the AI boom, compute power is emerging as one of this decade's most critical resources. In data centers across the globe, millions of servers run 24/7 to process the foundation getels and machine learning 組合, that nandal, enerity, learning needed to operate these data centers are collectively known as compute power—and there is an unquenchable need for more.我們的研究表明，到2030年，全球資料中心需投入6.7億美元才能跟上算力需求的成長步伐。其中，AI處理負載能力的資料中心預計需5.2億美元支出，而支撐傳統IT應用的資料中心預計需1.5億美元支出。總體而言，到2030年資本支出總需求接近7億美元——無論以何種標準衡量，這都是一個驚人的數字。Our research shows that by 2030, data centers are projected to require $6.7 trillion worldwide to keep pace with the demand for compute power. Data centers equipped to handle AI processing loads are projected to repowerp.2 pped to handle AI processing loads are projected to repowers 25.2 5% 5.表 5% applications are projected to require $1.5 trillion in capital expenditures. Overall, that's nearly $7 trillion in capital outlays needed by 2030—a staggering number by any measure.為了滿足這一需求，算力價值鏈上的公司必須在快速部署資本和審慎決策之間取得平衡。要提高資料中心投資獲得豐厚回報率，企業可分階段推進項目，並在每一步評估投資回報率。但未來需求的不確定性使得精確的投資計算難以實現。To meet this demand, companies across the compute power value chain must strike a balance between deploying capital quickly and doing so prudently. To improve the odds that their data center investments willnvide scan return, scom smooo, shool inscan wills scoms shd, scoms shd, scom swhe scom, sstal, san shd, scom, scom, sh.com each step. Still, a lack of clarity about future demand makes precise investment calculations difficult.算力價值鏈極為複雜——從建造資料中心的房地產開發商，到為其提供電力的公用事業公司，再到生產晶片的半導體企業，以及託管數兆太字節資料的雲端服務超大規模供應商均涵蓋其中。這條價值鏈上的領導者們深知，必須加大算力投資以推動人工智慧發展加速。然而，他們面臨的挑戰是艱難的：決定將多少資本分配到那些項目，同時卻無法確定人工智慧未來的成長與發展會如何影響算力需求。超大規模雲端服務提供者是否會繼續承擔成本壓力？還是企業、政府及金融機構會透過新的融資模式介入？在人工智慧使用量持續激增的情況下，資料中心需求是否會進一步上升？還是會隨著技術進步使人工智慧對算力需求的依賴減少而下降？The compute power value chain is complex—from the real estate developers that build data centers to the utilities that power them, to the semiconductor firms that produce chips to the cloud service hypersers that hosta caler produce chips to the cloud service hypersers that hosta caliv. must invest in compute power to accelerate AI growth. But their challenge is formidable: deciding how much capital to allocate to which projects, all while remaining uncertain of how AI's future growth and dedopment will impx or will enterprises, governments, and financial institutions step in with new financing models? Will demand for data centers rise amid a continued surge in AI usage, or will it fall as technological advances make AI usage, or will it fall as technological advances make less compute有一點毋庸置疑：此事利害攸關。過度投資資料中心基礎設施可能導致資產閒置，而投資不足則意味著落後。本文基於麥肯錫的研究分析，為算力價值鏈上的各類企業整理了未來五年的投資格局。儘管這些預測經過嚴謹論證，但我們也承認人工智慧是一個快速發展的領域。我們的分析雖然基於深入研究的假設，但仍存在一些無法量化的關鍵不確定性。One thing is certain: The stakes are high. Overinvesting in data center infrastructure risks stranding assets, while underinvesting means falling behind. This article, based on McKinsey research and analysis, provides com cross crossue. the next five years. Despite the rigor behind these forecasts, we acknowledge that AI is a radically evolving space. Our analysis is built on thoroughly researched hypotheses, but there are critical uncertainties that ununcertainet or unccan.預測算力需求曲線Predicting the compute power demand curve企業要決定在算力上投入多少，首先應精準預測未來需求——鑑於人工智慧產業變化如此之快，這絕非易事。我們的研究表明，到2030年，全球對資料中心容量的需求可能會接近三倍，其中約70%的需求來自人工智慧工作負載（見圖1）。但這一預測取決於兩個關鍵不確定性因素：To decide how much to invest in compute power, companies should first accurately forecast future demand—a challenging task given that the AI sector is shifting so rapidly. Our research shows that global mand de dataabout so rapidly。 of that demand coming from AI workloads (Exhibit 1). However, this projection hinges on two key uncertainties:人工智慧用例。人工智慧的價值體現在應用層面——企業如何將人工智慧轉化為實際業務價值。如果企業未能從人工智慧中創造實質價值，算力需求可能達不到預期。相反，變革性的人工智慧應用可能會推動比當前預測更大的需求。AI use cases. The value in AI lies at the application layer—how enterprises turn AI into real business impact. If companies fail to create meaningful value from AI, demand for compute power could fail short of expect age 決定, current projections suggest.快速創新周期和顛覆變革。人工智慧技術的持續進步，例如處理器、大語言模型（LLM）架構和功耗，可能會顯著提高效率。例如，2025年2月，中國的大語言模型參與者DeepSeek揭露，其V3模型在訓練和推理效率方面實現大幅提升，與GPT-4o相比，訓練成本大幅降低18倍，推理成本降低36倍。然而，初步分析表明，這類效率提升可能會被更廣泛人工智慧市場中激增的實驗與訓練活動所抵消。因此，從長期來看，效率提升或許不會對整體算力需求產生實質性影響。。 that its V3 model achieved substantial improvements in training and reasoning efficiency, notably reducing training costs by approximately 18 times and inferencing costs by about 36 times, pared . types of efficiency gains will likely be offset by increased experimentation and training across the broader AI market. As a result, efficiency gains may not substantially impact overall compute power demand over the long demandm.僅人工智慧需求就需5.2兆美元的投資AI demand alone will require $5.2 trillion in investment我們測算，2030年，算力價值鏈上的各類企業僅為滿足全球人工智慧需求，就需向資料中心投入5.2億美元。這一數字基於廣泛的分析和關鍵假設，包括到2030年預計需要156吉瓦（GW）與人工智慧相關的數據中心容量，2025年至2030年期間將增加125吉瓦。這5.2兆美元的數字反映了滿足人工智慧計算能力不斷增長的需求所需的巨額投資規模——這一龐大的資本投入，也凸顯了未來挑戰的艱巨性。We calculate that companies across the compute power value chain will need to invest$5.2 trillion into data centers by 2030 to meet worldwide demand for AI alone. We based this figure on extensive anawide demand for AI alone. We based this figure on extensive analysis and key assaings projects projects, 375), g sadvvvv) AI-related data center capacity demand by 2030, with 125 incremental GW added between 2025 and 2030. This $5.2 trillion figure reflects the sheer scale of investment required to meet the wing mand for meet the compute scale of investment required to meet the wing mand for meet the compute scale ofinvestmentremed magnitude of the challenge ahead (see sidebar “The scale of investment”).鑑於未來算力需求存在不確定性，我們建立了三種投資情景，從需求受限到加速成長（見圖2）。在第一種情境中，成長大幅提速，2025年至2030年期間將增加205吉瓦，這將需7.9億美元的資本支出。本文採用的是第二種情景：需求有所增長，但不及第一種情景，預計資本支出為5.2兆美元。第三種情境為需求較受限的情況，未來五年新增容量78吉瓦，總資本支出為3.7 億美元。Amid the uncertainty about future needs for compute power, we created three investment scenarios ranging from constrained to accelerated demand (Exhibit 2). In the first of our取 three scenarios, growth relbit 2). In the first of our取 three scenarios, growth relelerates signal accly 20 月added between 2025 and 2030. This would require an estimated $7.9 trillion in capital expenditures. The second scenario is the one we use in this article: Demand grows, but not groas we use in this article: Demand 實體$5.2 trillion. In our third scenario, in which demand is more constrained, with 78 incremental GW added in the next five years, the total capital expenditure is $3.7 trillion .2025-2030 年預測：人工智慧驅動的全球資料中心資本支出總額（按類別與情境劃分）單位：兆美元Global data center total capital expenditures driven by Al,by category and scenario, 2025-30 projection, $ trillion無論那種情景，這些投資數額都極為驚人，背後有多重因素驅動：In any scenario, these are staggering investment numbers. They are fueled by several factors:生成式人工智慧的大規模應用。支撐生成式人工智慧的基礎模型，其訓練與運作需要大量算力資源。訓練和推理工作負載都在推動基礎設施的成長，預計到2030年，推理將成為主要的工作負載。Mass adoption of gen AI. The foundation models that underpin gen AI require significant compute power resources to train and operate. Both training and inference workloads are contributing to infrastructure growth, with inference expected to ben contributing to infrastructure growth, with inference expected to become 30.企業整合。在各行業（從汽車到金融服務）部署人工智慧驅動的應用程序需要大量的雲端運算能力。隨著應用場景不斷增多，人工智慧應用將愈發複雜，會整合為特定領域量身定製的專用基礎模型。Enterprise integration. Deploying AI-powered applications across industries—from automotive to financial services—demands massive cloud computing power. As use cases grow, AI applications will grow more sophisticated, integrating specialed.激烈的基礎設施競賽。超大規模供應商和企業正在競相建立專屬人工智慧算力以獲得競爭優勢，這推動了越來越多的資料中心的建設。這些「建設者」（下文將進一步描述）希望透過實現規模效應、優化資料中心技術堆疊，最終降低算力成本，從而鞏固競爭優勢。Competitive infrastructure race. Hyperscalers and enterprises are racing to build proprietary AI capacity to gain competitive advantage, which is fueling the construction of more and more data centers. These "builders” (as furbe des pib) descr3 3000m ”(achieving scale, optimizing across data center tech stacks, and ultimately driving down the cost of compute.地緣政治考量。各國政府正大力投資人工智慧基礎設施，以增強安全、經濟領導和技術獨立性。Geopolitical priorities. Governments are investing heavily in AI infrastructure to enhance security, economic leadership, and technological independence.這些投資將流向何處？Where is the investment going?需要說明的是，我們對人工智慧基礎設施 5.2 兆美元的投資預測存在一定侷限性—— 該分析可能低估了所需的總資本投入，因為我們的估算僅量化了五類算力投資者中的三類，即建設者、能源供應商以及技術研發與設計商。這三類投資者直接為人工智慧發展所需的基礎設施和基礎技術提供資金支援（詳見側邊欄「五類資料中心投資者」）。約15%（0.8兆美元）的投資將流向建設者，用於土地、材料和場地開發。另有25%（1.3兆美元）將分配給賦能者，用於發電和輸電、冷卻和電氣裝置。最大的投資份額，即60%（3.1兆美元），將流向技術開發者和設計師，他們為數據中心生產晶片和計算硬體。另外兩種投資者原型，運營商（如超大規模提供商和共址提供商）和人工智慧架構師（他們構建人工智慧模型和應用）也投資於計算能力，特別是在人工智慧驅動的自動化和數據中心軟件等領域。但由於這類投資與他們的整體研發支出有重疊，因此難以單獨量化其在算力上的具體投入規模。To qualify our $5.2 trillion investment forecast for AI infrastructure, it's important to note that our analysis likely undercounts the total capital investment needed, as our estimate quantifies capital investment for only three out of five compute power investor archetypes—builders, energizers, and technology developers and designers—that directly finance the infrastructure and foundational technologies necessary for AI growth (see sidebar “Five types of data center investors”). Appximate 是 15 cent 你。 materials, and site development. Another 25 percent ($1.3 trillion) will be allocated to energizers for power generation and transmission, cooling, and electrical equipment. The largest share of investment, 60 will ($3. produce chips and computing hardware for data centers. The other two investor archetypes, operators, such as hyperscalers and colocation providers, and AI architects, which build AI models 和center software. But quantifying their compute power investment is challenging because it overlaps with their broader R&D spending.五類資料中心投資者Five types of data center investors儘管預計需要如此龐大的資本投入，但我們的研究顯示，當前投資規模仍落後於需求。在數十次客戶訪談中我們發現，執行長們不願全力投資算力產能，因為他們對未來需求的洞察力有限。對人工智慧採用是否會繼續快速上升的不確定性，以及基礎設施項目有很長的前置時間，使得企業難以做出明智的投資決策。許多公司不確定今天對人工智慧基礎設施的大規模資本支出是否會在未來產生可衡量的投資回報率。那麼，企業領導者如何能自信地推進他們的投資呢？首先，他們可以確定自己的組織在計算能力生態系統中的位置。Despite these projected capital requirements, our research shows that current investment levels lag demand. In dozens of client interviews, we found that CEOs are hesitant to invest in compute power capacity at maximum levels be hesitant to invest in compute power capacity at maximum levels be hesit. about whether AI adoption will continue its rapid ascent and the fact that infrastructure projects have long lead times make it difficult for companies to make informed investment decisions. Many comcommend are unsure wheake informed investment decisions。 measurable ROI in the future. So how can business leaders move forward confidently with their investments? As a first step, they can determine where their organizations fall within the compute power ecosystem.人工智慧基礎設施投資者的五大類型Five archetypes of AI infrastructure investors這場兆級美元的人工智慧算力投資競賽背後，究竟是誰在主導？我們已明確說明五類核心投資者類型，每類都面臨獨特的挑戰與機遇，同時也詳細測算出它們未來五年的潛在投入規模。Who are the investors behind the multitrillion-dollar race to fund AI compute power? We have identified five key investor archetypes, each navigating distinct challenges and opportunities, and detailed how much they could spstinct challenges and opportunities, and detailed how much they could spend in the next five spendy.1. 建設者 Builders核心身份：房地產開發商、設計公司和建築公司，正在擴大大量資料中心的容量。Who they are: real estate developers, design firms, and construction companies expanding data center capacity人工智慧工作負載的資本支出：8000億美元。AI workload capital expenditure: $800 billion非人工智慧工作負載的資本支出：1000億美元。Non-AI workload capital expenditure: $100 billion關鍵投資：土地和材料採購、熟練勞動力、場地開發。Key investments: land and material acquisition, skilled labor, site development機遇。建設者若能優化選址，可搶佔核心區位、縮短建設周期，並儘早融入營運反饋，從而實現數據中心更快部署與更高運營效率。Opportunities.Builders that optimize site selection can secure prime locations, reduce construction timelines, and integrate operational feedback early, ensuring faster deployment and higher data center effency.挑戰。技術人員與建築工人的勞動力短缺可能影響人力供給，而選址限制可能壓縮可選場地範圍。與此同時，機架功率密度的增加可能會帶來空間和冷卻方面的挑戰。Challenges.Labor shortages could impact technician and construction worker availability, while location constraints could limit site selection options. Meanwhile, increased rack ges density could create space and cooling lenlen.解決方案。具有前瞻性的建設者能夠找到核心挑戰的應對之策，為其投資決策增添確定性。例如，一些建設者透過採用模組化設計來解決勞動力短缺問題，這種設計簡化了建設過程，例如在場外建造大型元件，然後在現場組裝。Solutions.Forward-thinking builders can find solutions to core challenges, adding certainty to their investment decisions. For example, some are solving the labor shortage issue by adopting mod alsigns that solving the labor shortage issue by adopting mod alsigns that cesstreams sargeion can be assembled on-site.2. 賦能者 Energizers核心身份：公用事業公司、能源供應商、冷卻/電氣裝置製造商和電信運營商，他們正在為人工智慧數據中心建設電力和連接基礎設施。Who they are: utilities, energy providers, cooling/electrical equipment manufacturers, and telecom operators building the power and connectivity infrastructure for AI data centers人工智慧工作負載的資本支出：1.3兆美元。AI workload capital expenditure: $1.3 trillion非人工智慧工作負載的資本支出：2000億美元。Non-AI workload capital expenditure: $200 billion核心投資領域：發電（發電廠、輸電線路）、冷卻解決方案（空氣冷卻、直接晶片液體冷卻、浸沒式冷卻）、電氣基礎設施（變壓器、發電機）、網絡連接（光纖、電纜）。Key investments: power generation (plants, transmission lines), cooling solutions (air cooling, direct-to-chip liquid cooling, immersion cooling), electrical infrastructure (transformers, generators), net connectivity) (networkn,able)機遇。能源供應商若能擴大電力基礎設施規模，並在可持續能源解決方案方面開展創新，將能更能掌握超大規模雲服務供應商日益增長的能源需求所帶來的機會。Opportunities.Energizers that scale power infrastructure and innovate in sustainable energy solutions will be best positioned to benefit from hyperscalers' growing energy demands.挑戰。現有電網的薄弱環節可能導致資料中心供電受阻，而處理器密度不斷提升所帶來的熱管理難題仍是一大障礙。此外，賦能者還面臨清潔能源轉型的要求和漫長的電網連接審批流程。Challenges.Powering data centers could stall due to existing grid weaknesses and solving heat management challenges from rising processor densities remains an obstacle. Energizers also face clean-energy transition re解決方案。鑑於超過 1 兆美元的投資面臨風險，賦能者正在尋找方法提供可靠的電力，同時推動投資回報率。他們正在對新興發電技術進行大量投資，包括核能、地熱能、碳捕獲與儲存以及長期能源儲存。同時，他們正加強，盡快提升各類能源的上線容量既涵蓋再生能源，也包括天然氣、化石燃料等傳統能源基礎設施。當前的變化在於，能源需求的規模已極為龐大，這催生了以空前速度建設發電產能的新緊迫性。隨著需求——尤其是對清潔能源的需求——的激增，預計發電量將迅速增長，可再生能源預計到2030年將佔能源結構的45%到50%，而今天僅佔約三分之一。Solutions.With over $1 trillion in investment at stake, energizers are finding ways to deliver reliable power while driving ROI. They are making substantial investments in emerging power-generation technologies—including captures, ) capgeage, 片面They are also doubling down on efforts to bring as much capacity online as quickly as possible across both renewable sources and traditional energy infrastructure, such as gas and fossil fuels. What ismands nowa is a遠 that the build power capacity at unprecedented speed. As demand—especially for clean energy—surges, power generation is expected to grow rapidly, with renewables projected to account for approximately 45 to 50 percent to 20 energy 2003, 000 energy 30day 50 enerper 20030,3. 技術研發與設計商 Technology developers and designers核心身份：為資料中心生產晶片和計算硬體的半導體公司和IT供應商。Who they are: semiconductor firms and IT suppliers producing chips and computing hardware for data centers人工智慧工作負載的資本支出：3.1兆美元。AI workload capital expenditure: $3.1 trillion非人工智慧工作負載的資本支出：1.1兆美元。Non-AI workload capital expenditure: $1.1 trillion核心投資領域：圖形處理單元（GPU）、中央處理單元（CPU）、記憶體、伺服器和機架硬體。Key investments: GPUs, CPUs, memory, servers, and rack hardware機遇。技術研發與設計商若能投資於具備可擴展性、面向未來的技術，且擁有明確的需求洞察力作為支撐，將能在人工智慧計算領域獲得競爭優勢。Opportunities.Technology developers and designers that invest in scalable, future-ready technologies supported by clear demand visibility could gain a competitive edge in AI computing.挑戰。少數幾家半導體公司控制著市場供應，抑制了競爭。產能建設仍不足以滿足當前需求，與此同時，人工智慧模型訓練方法與工作負載的變化，使得特定晶片的未來需求難以預測。Challenges.A small number of semiconductor firms control the market supply, stifling competition. Capacity building remains insufficient to meet current demand, while at the same time, shifts in AI model trainture method and while at the same time, shifts in AI model 大chips.解決方案。在算力競賽中，技術研發與設計商的潛在效益最大，因為正是它們提供了承擔實際計算工作的處理器與硬體。目前市場對其產品的需求旺盛，但它們的投資需求也最為龐大——未來五年將超過 3 億美元。少數幾家半導體公司對行業供應有著不成比例的影響，使他們成為計算能力成長的潛在瓶頸。技術開發者和設計師可以透過擴大製造能力並多樣化供應鏈來緩解這一風險，以防止瓶頸。Solutions.Technology developers and designers have the most to gain in the compute power race because they are the ones providing the processors and hardware that do the actual computing. Demand for their products is currently high, but the grle 造成 highnes the grolducts is currently0,five years. A small number of semiconductor firms have a disproportionate influence on industry supply, making them potential chokepoints in compute power growth. Technology developers and designers can mitigate this ifyrby expanding and unywperion this risk by bottlenecks.4. 運營商 Operators核心身份：超大規模提供者、共址提供者、GPU即服務平台以及透過提高服務器利用率和效率來優化計算資源的企業。Who they are: hyperscalers, colocation providers, GPU-as-a-service platforms, and enterprises optimizing their computing resources by improving server utilization and efficiency人工智慧工作負載的資本支出：未包含在此分析中。AI workload capital expenditure: not included in this analysis非人工智慧工作負載的資本支出：未包含在此分析中。Non-AI workload capital expenditure: not included in this analysis核心投資領域：資料中心軟件、人工智慧驅動的自動化、訂製矽片。Key investments: data center software, AI-driven automation, custom silicon機遇。業者若能有效率擴大規模，同時平衡投資回報率（ROI）、性能與能源消耗，將可望引領產業長期發展。Opportunities.Operators that scale efficiently while balancing ROI, performance, and energy use can drive long-term industry leadership.挑戰。人工智慧託管應用尚不成熟，可能導致長期投資回報率（ROI）的計算難以清楚量化。資料中心營運效率低正推高成本，但人工智慧需求的不確定性仍在繼續擾亂長期基礎設施規劃和採購決策。Challenges.Immature AI-hosted applications can obscure long-term ROI calculations. Inefficiencies in data center operations are driving up costs, but uncertainty in AI demand continues to disrupt long-term.解決方案。儘管如今的數據中心已處於較高的運行效率水平，但人工智慧創新的迅速發展仍要求運營商同時優化能源消耗與工作負載管理。部分運營商透過投資更有效的冷卻解決方案和增加機架堆疊能力在不犧牲處理能力的前提下減少空間需求，從而提高資料中心的能源效率。另一些運營商正在投資人工智慧模型開發本身，以建立需要較少計算能力來訓練和運行的架構。Solutions.While data centers today operate at high-efficiency levels, the rapid pace of AI innovation will require operators to optimize both energy consumption and workload management. Some operators are improving energy dataeffect inject insticution and 周長increasing rack stackability to reduce space requirements without sacrificing processing power, for example. Others are investing in AI model development itself to create architectures that need less compute power to be trained and operateditectures that need less compute power to be trained and operated.5. 人工智慧架構師 AI architects核心身份：人工智慧模型開發者、基礎模型提供者、建構專有AI能力的企業。Who they are: AI model developers, foundation model providers, and enterprises building proprietary AI capabilities人工智慧工作負載的資本支出：未包含在此分析中。AI workload capital expenditure: not included in this analysis非人工智慧工作負載的資本支出：未包含在此分析中。Non-AI workload capital expenditure: not included in this analysis核心投資領域：模型訓練與推理基礎設施，演算法研究。Key investments: model training and inference infrastructure, algorithm research機遇。人工智慧架構師若能發展出平衡效能與低算力需求的架構，將引領下一波人工智慧應用浪潮。而投資於AI能力的企業可以透過開發符合其需求的專用模型來獲得競爭力。Opportunities.AI architects that develop architectures that balance performance with lower compute requirements will lead the next wave of AI adoption. Enterprises investing in proprietary AI capabilities can eds competitiveness by 字詞 specialtail bys specialtail.挑戰。人工智慧治理相關問題，包括偏見、安全和監管，增加了行業複雜度，可能會減緩發展。與此同時，推理構成了一個主要的不可預測成本組成部分，企業正面臨難以證明人工智慧投資的明確投資回報率的困難。Challenges.AI governance issues, including bias, security, and regulation, add complexity and can slow development. Meanwhile, inference poses a major unpredictable cost component, and enterprises are facing dicis ultare facs解決方案。大規模人工智慧模型不斷升級的計算需求正在增加訓練它們的成本，特別是在推理方面，即訓練有素的人工智慧模型將他們學到的知識應用到新的、未見過的數據上以做出預測或決策的過程。具有先進推理能力的模型，如OpenAI的o1，需要顯著更高的推理成本。例如，與該公司的非推理型GPT-4o相比，OpenAI的o1的推理成本高出六倍。為降低推理成本，領先的AI企業正通過稀疏啟動、知識蒸餾等技術優化模型架構。這些方案能夠減少AI模型產生回應時所需的算力，提升營運效率。Solutions.The escalating computational demands of large-scale AI models are driving up the costs to train them, particularly regarding inference, or the process where trained AI models apply their learned knowledge to newsion, sdics dataly their learns formly their learndeen to newsion, sdics dataorv. capabilities, such as OpenAI's o1, require significantly higher inference costs. For example, it costs six times more for inference on OpenAI's o1 compared with the company's nonreasoning GPT-4. optimizing their model architectures by using techniques like sparse activations and distillation. These solutions reduce the computational power needed when an AI model generates a response, making operations more icient.人工智慧基礎設施發展的關鍵考量因素Critical considerations for AI infrastructure growth企業在規劃人工智慧基礎設施投資時，需應對多種潛在結果。在需求受限的場景下，受供應鏈限制、技術變革及地緣政治不確定性影響，人工智慧相關資料中心的產能建設可能需要 3.7 億美元資本支出。而在需求加速成長的場景下，這些阻礙因素得到緩解，投資規模或高達 7.9 億美元。緊跟不斷變化的行業格局，對於制定明智且具戰略性的投資決策至關重要。投資者必須考慮的不確定性因素包括：As companies plan their AI infrastructure investments, they will have to navigate a wide range of potential outcomes. In a constrained-demand scenario, AI-related data centercapacityilliondisruptions, and geopolitical uncertainty. These barriers are mitigated, however, in an accelerated-demand scenario, leading to investments as high as $7.9 trillion. Staying on top of the evolving landscape is instical landscapeking informed. uncertainties investors must consider include:技術變革。模型架構的突破，包括計算利用效率的提高，可能降低對硬體和能源的預期需求。Technological disruptions. Breakthroughs in model architectures, including efficiency gains in compute utilization, could reduce expected hardware and energy demand.供應鏈限制。勞動力短缺、供應鏈瓶頸和監管障礙可能會延遲電網連接、晶片供應和數據中心擴展——減緩整體人工智慧的採用和創新。為瞭解決關鍵晶片的供應鏈瓶頸，半導體公司正在投入大量資本建設新的製造設施，但由於監管限制和上游裝置供應商的長前置時間，這種建設可能會停滯。Supply chain constraints. Labor shortages, supply chain bottlenecks, and regulatory hurdles could delay grid connections, chip availability, and data center expansion—slowing overall 過程investing significant capital to construct new fabrication facilities, but this construction could stall due to regulatory constraints and long lead times from upstream equipment suppliers.地緣政治緊張局勢。關稅波動與技術出口管制可能為算力需求帶來不確定性，進而潛在影響基礎設施投資與人工智慧的發展。Geopolitical tensions. Fluctuating tariffs and technology export controls could introduce uncertainty in compute power demand, potentially impacting infrastructure investments and AI growth.競爭優勢的角逐The race for competitive advantage在人工智慧驅動的計算時代，贏家將是那些能夠預判算力需求並進行相應投資的企業。算力價值鏈上的各類企業，若能主動掌握關鍵資源（土地、材料、能源容量和計算能力）將可望獲得顯著的競爭優勢。為實現有把握的投資，它們可採取三管齊下的策略。The winners of the AI-driven computing era will be the companies that anticipate compute power demand and invest accordingly. Companies across the compute power value chain that proactively secure critical resources — land, erialscity, gy as enerftd enerure critical resources—land, aterialscity, cat enerle enerftd enerft, enery, part, enerityenertercomenerftdedge. To invest with confidence, they can take a three-pronged approach.首先，投資者需要在不確定性中理解需求預測。企業應儘早評估人工智慧計算需求，預判需求的潛在變化，並制定具備可擴展性的投資策略，以適應人工智慧模型及應用場景的演進。其次，投資人需探索提升計算效率的創新路徑。具體而言，可優先投資於兼具成本效益與能源效率的計算技術，在最佳化性能的同時，管控能耗與基礎設施成本。第三，投資者可建構供應端韌性，確保人工智慧基礎設施在不過度投入資本的前提下持續成長。這將需要投資者確保關鍵投入（如能源和晶片），最佳化選址，並為供應鏈注入靈活性。First, investors will need to understand demand projections amid uncertainty. Companies should assess AI computing needs early, anticipate pot ential shifts in demand, and design scalable investment strategies that can ential shifts in demand, and design scalable investment strategies that can ential shifts in demand, and design scalable investment strategies that can ential shifts in deals of designal mod . innovate on compute efficiency. To do so, they can prioritize investments in cost- and energy-efficient computing technologies, optimizing performance while managing power consumption and infrastructure motside Third, the can punion sem. without overextending capital. This will require investors to secure critical inputs such as energy and chips, optimize site selection, and build flexibility into their supply chains.在成長與資本效率之間取得平衡至關重要。戰略投資不只是一場擴巨量資料基礎設施規模的競賽，更是一場塑造人工智慧未來的競爭。Striking the right balance between growth and capital efficiency will be critical. Investing strategically is not just a race to scale data infrastructure—it'sa race to shape the future of AI itself. （DeepKnowledge）
