硬AI·2025-11-03 09:20Core Insights - A significant infrastructure race driven by artificial intelligence is unfolding in the United States, with planned data center projects exceeding 45 GW and attracting over $2.5 trillion in investments [1][2] Group 1: Major Players - The expansion is primarily led by major tech companies such as OpenAI, Amazon, Meta, Microsoft, and xAI, which are rapidly planning and constructing computing clusters to train and run increasingly complex AI models [1][2][4] - OpenAI's Stargate project aims for a capacity of 10 GW and an investment of $500 billion by the end of 2025, with approximately 7 GW already committed [6][10] - Meta is advancing multiple Titan clusters, including a 1 GW project in Ohio and a planned 5 GW expansion in Louisiana [6] - Amazon has added 3.8 GW of capacity globally in the past year and is expected to double its capacity by 2027, potentially increasing by about 13 GW in the U.S. between 2026 and 2027 [6] - Microsoft is constructing a 900 MW AI facility in Wisconsin and has plans for several similar projects across the U.S. [6] - xAI is expanding its data center in Memphis, Tennessee, to 1.4 GW for training its Grok model [6] Group 2: Investment and Costs - The construction cost of data centers has surpassed $17 million per MW, with OpenAI's Stargate project reflecting a cost of $57 million per MW, highlighting the high capital density of AI infrastructure [7][11] - The financing structure for these massive investments is complex, involving not only the tech giants' capital expenditures but also private equity firms and specialized infrastructure funds [11][12] - The "Energy as a Service" (EaaS) model is emerging, with energy companies like Williams signing long-term power purchase agreements to invest billions in dedicated power facilities for data center operators [12] Group 3: Power Supply Challenges - The existing power grid poses significant challenges for data center construction, leading companies to adopt on-site power generation strategies to ensure reliability and accelerate project timelines [9][10] - For instance, the Stargate 1 project plans to deploy approximately 350 MW of on-site natural gas generation despite having grid access approval for 1.2 GW [9] Group 4: Supply Chain Issues - The explosive demand for power equipment has strained the supply chain, with heavy gas turbine prices rising by 50% in less than two years and extended delivery times [14] - Companies are resorting to acquiring second-hand or "off-the-shelf" new equipment to circumvent long order queues, as seen with Fermi America acquiring a Siemens gas turbine from an unused LNG project [14]