@Rocky_Bitcoin: Goldman Sachs has released a very interesting report on AI computing power between China and the US! It first mentions that China's investment in AI data centers will reach 2 trillion RMB (about 300 billion USD) over the next five years, with massive CapEx spending, and most computing power will be located in the western regions because electricity is cheap and land is abundant there. Let me interject here—I think young people should consider going to the west, ...

Goldman Sachs released a very interesting report on US-China AI computing power!

First, it mentions that China’s investment in AI data centers will reach 2 trillion RMB (about $300 billion) over the next 5 years, with huge CapEx spending. Most of the computing power will be located in the western regions because electricity is cheap and land is abundant there.

I’ll add a side note here: I think young people should consider going to the western regions instead of grinding in the Pearl River Delta or Yangtze River Delta. The west offers many opportunities in the future AI industry — even if you’re an electrician or installer, there’s a promising path there.

Second, the report points out that while China’s AI computing costs appear cheap, they are actually more expensive. When purchasing domestic chips, the capital expenditure per IT power is 40-50% cheaper than imported chips (commonly known as “cheaper to buy the cards”). However, due to lower computing density and higher power consumption, when converted to “capital expenditure per unit of computing power,” domestic chips are actually 2 to 4 times more expensive than imported ones! Even more critical: the computing power squeezed out per IT power is only 10% to 30% of that of imported chips.

Finally, the report compares Huawei vs. Nvidia’s real-world gap (see Figure 2). It explicitly names Huawei’s 910B and 910C. In actual servers, their average daily token output is only 1/6 to 1/3 of Nvidia’s earlier compliant version H800 (note: not even the original H100/H200).

The conclusion is that the AI computing power battle between China and the US has just begun, and the gap is still very wide. Therefore, short-term CapEx spending on AI will only increase, not decrease. AI hardware investment is far from over!

What do you all think?!

Rocky (@Rocky_Bitcoin): This time, Apple’s leak incident in India seems more serious than we thought!

At least six documents in the leaked archive fully map the dependency relationships of hundreds of core components and their corresponding suppliers for the iPhone 18 Pro. Also included are partial multi-layer motherboard circuit diagrams and technical specifications for the next-generation A20 Pro processor and Apple’s self-developed C2 baseband chip. These are Apple’s “top secrets.”