Deutsche Bank AG(US:DB) 华尔街见闻·2026-01-15 07:56Core 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].