The Earth really can't fit AI anymore! The next stop can only be space! The ground data center has hit a wall: a 1GW AI cluster consumes as much electricity as a medium-sized city, with millions of liters of fresh water used for cooling every day, and electricity, water, and site selection are all at their limits. Why is space the ultimate destination of computing power? one ️⃣ Energy dimensionality reduction strike Orbital solar energy is not affected by atmospheric refraction or cloud cover, and its utilization efficiency is 8-10 times higher than that of the ground. It also provides 24-hour uninterrupted stable power input. Musk said that we are struggling with micro nuclear fusion on Earth, just like making an ice maker in Antarctica, but ignoring the 4.5 billion year stable operation of the super fusion reactor above our heads. Isn't it good to use ready-made ones directly? two ️⃣ Cooling is almost free The background temperature in space is -270 ° C. In a vacuum environment, heat can be directly discharged into deep space through huge radiation radiators, and the energy efficiency of data centers can infinitely approach 1. 40% of the electricity in ground data centers is wasted on air conditioning, while space can directly feed almost all of the electricity to chips. This is the dimensionality reduction attack of computing power density. three ️⃣ Communication is also faster The speed of light propagation in vacuum is 30% faster than in optical fibers. By directly constructing a global computing power network through satellite laser links, computing nodes become global relay stations with faster response times. It's no longer the PPT stage: -Path A: Starcloud-1, an on orbit edge computing startup, has put a Starcloud-1 system about the size of a small refrigerator into orbit in cooperation with Nvidia. It not only successfully called Google's Gemma model to send a greeting of "Hi Earthlings!" to Earth, but also completed in orbit training using Nano GPT. This mode allows AI to directly process sensor data in space. -Path B: Google's "Suncatcher" concept consists of a cluster of 81 satellites interconnected through laser links, forming a super large-scale computing architecture suspended in space. The Trillium TPU developed by Google has passed radiation testing equivalent to a lifespan of 5 years, aiming to provide continuous computing power supplementation for ground cloud computing. -SpaceX announced the acquisition of AI company xAI, with a combined valuation of up to $1.25 trillion. Musk has made it clear that the core task of the merger is to promote the deployment of space data centers. He expects that within the next 2 to 3 years, utilizing Starship's full recovery capability, space will become the place with the lowest cost of deploying AI computing power globally. Of course, the real roadblocks are also very hardcore: -Cost gap: Building a 1GW ground data center costs about 5 billion RMB, while the cost of a space center of the same scale may reach hundreds of billions of dollars, which is equivalent to several tens or hundreds of times the cost of starting from the ground. -Engineering Extreme Challenge: In order to support 100MW computing power, the solar panel area of the satellite needs to reach dozens of football fields, and it must rely entirely on robots for automated maintenance, which is vastly different from the difficulty of manual operation and maintenance on the ground. -The variable of launch cost: The industry is currently hoping for a cliff like decline in launch costs. Google predicts that the launch cost will drop to $200/kg in the mid-2030s, and if Starship achieves full recycling, the cost may even be reduced to 15- $60/kg. In addition, orbital congestion and debris risk are also hanging over Damocles' head This is a bold gamble of exchanging ultra-high investment costs for long-term near zero operating costs. In the short term, it will not replace the ground, but rather ground air coordination. Ground based high-frequency interaction, specializing in large-scale model training in space, which involves energy consumption, heat dissipation, and offline work. The pace of human civilization's evolution may really be breaking away from the gravitational pull of the Earth. Do you think space computing power is a castle in the air, or the next new track that is about to take off?