Whether Microsoft or Amazon, both prove that computing power leasing is to some extent a simple problem, because demand is real.

English summary

① Open source catching up with closed source does not reduce demand. After open source models catch up, millions of developers, SMEs, and vertical agents flock in to run inference, expanding computing power from 'a few giants for self-use' to 'the whole market's shared rigid demand'. Microsoft Azure's 90% of revenue comes from long-tail users, Amazon maintains a net profit margin of nearly 40% with rapid growth. Computing power is rigid demand, and renting computing power is the best positioning. ② The dispute between software and hardware is not important to computing power leasing. On the software side (models, agents, applications), it's hard to pick winners, trial and error costs are high, and narratives switch quickly; on the hardware side (chips, servers), it's either stuck in supply and technology iteration, or you have to bet on the path. Computing power leasing sits in the most comfortable middle position: the upstream helps you digest 'getting cards, networking, funding' and other dirty and tiring work, while the downstream standardizes computing power into a rentable service. Customers don't have to invest in a bunch of machines with their own CapEx, sign 5-year long-term contracts, and cash flow is predictable (it's rare to have a sector where 5-year profits can be calculated). ③ Additional gains beyond fixed income are already visible...