I saw that 36kr's article provided the following data in the chart.

These temporary miners who switch their hashing power to run AI $BTC are no different from retail investors cutting losses at $BTC to buy gold and silver.
Both are chasing the trend and panic selling without their own judgment. Is the US's computing power shortage due to your miners? No, it's an energy issue. Nvidia's new GPUs are about to double their hash rate again.
Mining machines are not the bottleneck; the generators in mining farms are what AI giants are eyeing.
As everyone knows, tokens in the US are expensive; spending 1 kWh on tokens can earn you the equivalent of 10 kWh. But tokens won't stay expensive forever. Look at Minimax leading the total token consumption—aren't you envious?
The competitors across the Pacific are pricing at only 1/15 of that, and OpenAI and others will eventually have to adjust their prices.
So, the huge profit of earning 10 kWh from 1 kWh is just a trap, like DeFi mining.
Once suppliers are all in place and have invested heavily upfront to build generators and data centers, they will definitely go crazy with competition later.
The outcome might even be running tokens at a loss. You have to mine somehow—leaving idle equipment to depreciate to zero is pointless, and mining at a loss might still recover some of the initial capital.
Or maybe go back to mining BTC?
Mining tokens requires different availability/network latency than mining BTC. It involves additional investments, such as slicing large models, parallel processing, and pipelines that need ultra-low latency and extremely high throughput between GPUs across multiple servers, requiring special cabling.
2C inference also needs a high-speed network environment.
These may sound like minor details, but the costs to implement them are not cheap at all.