Core Viewpoint - The transition of high-energy AI computing to space is moving from theoretical discussions to economically viable validation, with the cost gap between space and terrestrial data centers rapidly narrowing [2][3]. Cost Gap Rapidly Closing - Deutsche Bank's model indicates that while the current cost of deploying a 1 GW space data center is at least 7 times that of terrestrial centers, this ratio is expected to decrease to 4 times by the late 2020s and reach cost parity in the 2030s [3][4]. - The decline in costs is primarily driven by reductions in launch costs and improvements in satellite design and energy efficiency, leading to a significant decrease in the mass required for orbital deployment [3][4]. Projected Cost Data - In the estimated scenario for 2026, the cost of space deployment is projected to be $114 billion, compared to $16 billion for terrestrial deployment, resulting in a difference factor of 7.2 times. By the "optimized scenario" in 2032, space deployment costs are expected to drop to $18 billion, nearly equal to the terrestrial cost of $16 billion, with a difference factor of 1.2 times [4][5]. Key Factors for Economic Reversal - The critical variable for achieving this economic reversal is the dramatic drop in launch costs, which are projected to fall from $1,600 per kg in 2026 to $67 per kg by 2032 [6]. - The report emphasizes the importance of fully reusable rockets and economies of scale in operations, suggesting that launch costs could potentially decrease to as low as $1 million or even below $70 per kg over time [7]. Hardware Optimization - In addition to launch costs, significant advancements in orbital hardware are anticipated. By the 2030s, the cost of a single satellite is expected to drop below $2 million, or just $10,000 per kW, featuring a 150 kW power system and custom chips designed for space AI infrastructure [9]. - However, the model's assumptions are based on the premise that ground capacity costs remain unchanged, and it does not account for the expensive procurement costs of GPU/TPU chips. The report warns that if ground-based energy sources, such as nuclear power, become rapidly available and inexpensive, the assumptions may no longer hold [9][10].

将高能耗的AI算力送入轨道，正从理论探讨走向经济可行性验证阶段。 据追风交易台消息，德银在1月14日的报告中表示，虽然目前的太空部署成本高昂，但这一差距正在以惊人的速度缩小。分析师预测，随着发射成本的暴跌 和卫星设计的优化，在未来十年内，建设一个太空数据中心的成本将与地面建设成本趋于平价。这意味着，困扰地面数据中心的能源和散热瓶颈，或将在太 空中找到终极解决方案。 成本鸿沟的快速弥合 德银的模型显示，目前在太空部署算力仍是昂贵的赌注，但转折点已在视线之内。 根据其测算，建设一个1吉瓦（GW）容量的太空数据中心，目前的成本至少是地面的7倍。然而，这一倍数将在本十年后期迅速收窄至4倍，并最终在2030 年代实现成本平价。报告指出： 发射成本暴跌是关键 实现这一经济性逆转的核心变量在于发射成本的断崖式下跌。模型假设，每千克发射成本将从2026年的1600美元，一路狂泻至2032年的67美元。 德银在报告中强调了完全火箭复用和运营规模化的重要性： 我们假设目前低地球轨道（LEO）的发射市场价格约为7000万美元或4000美元/千克。随着时间的推移，若是实现完全火箭复用和运营规模化， 这一成本可能大幅下降至1000万美 ...