Many people praise CoreWeave's GPU leasing business.

As an old miner, I just poured cold water on mining companies; now I’ll pour cold water on the computing power leasing industry.
Previously, X mentioned that the cycle development of the mining era is different in each period:
Mining machines reign supreme
Electricity is king
Capital is king
The computational power for AI is essentially also mining, mining tokens;
AI training achieves miracles through massive compute, fundamentally similar to Bitcoin mining,
AI inference and Ethereum mining are both storage and computation integrated, even using the same cards like the 5090,
It's just that the software ecosystem differs, but overall, AI mining’s cycle won’t fluctuate as wildly as Bitcoin halving cycles,
However, the industry will have its natural cycles and development patterns.
The era of mining machines being king is the evolution through CPU-GPU-ASIC stages.
GPU leasing business just happens before ASIC and before electricity becomes king.
Once the following conditions are met, leasing business will take a big hit:
1. Excess GPU supply
2. Mass production of ASIC machines
3. Balanced or even redundant power supply
4. Slowing growth in AI paid user base
If any two of these happen simultaneously, GPU depreciation will plummet, and leasing prices will sharply decline.
A single 8-card 1660s, 240M
A single ASIC A11, 2G
Not only is the computing power crushed, but also the power efficiency, completely crushing GPUs—ASICs are a full generation ahead of GPUs.
In the future, an AI compute arms race will emerge, with ASIC chips blooming everywhere; big companies will have their own custom chips, focusing on higher efficiency, lower costs, and greater compute power…
Google’s TPU is already a model and trend.
The era dominated by ASIC-led mining machines is imminent, likely as soon as next year.
The trend for GPUs is accelerated depreciation.
The most comfortable period for leasing businesses is these two years, which is also the peak.
If you only look at financial reports, it’s very misleading.
As a long-term investment, it’s bound to be a trap.
All mature industries should return to reasonable profits from their core business.
That’s also why capital is疯狂追逐 $CBRS 超20倍认购, the largest tech IPO this year.
Those who understand are willing to pay a premium early on, but then sell at high points. $CRWV
Comparing to Bitcoin mining, you can clearly see the development rhythm and cycle of the AI industry—this is an investment compass, not a map.
A map can lead you into industry internal churn and emotional swings.
If you see legendary AI investor Leopold Aschenbrenner heavily invested in $CRWV last quarter and blindly follow, you don’t know why he invested, when he sold, or his investment style.
You’re definitely buried.
So, you see, the stock god bought puts to short/hedge the overheated chip sector in Q1, with a total position of 13.7 billion, hedging with 8.5 billion in puts, while in April and May everyone was rushing into chips…
He might have already sold out and reversed into options shorts.
As retail investors, giving up independent thinking and sensitivity, what advantages do you have over institutions?
His investment style is that of a manager with deep industry chain knowledge and expertise in financial leverage.
He is very sensitive to the rhythm of investments, such as the shifts in chips, storage, power, etc.;
He is highly attuned to market sentiment, using both long and short options simultaneously, with faster opening and closing speeds than many investors.
All these are based on a clear understanding of industry development, which is also related to his access to information in the AI hub.
If you don’t understand the industry’s direction and follow his delayed disclosures to build positions, you’re definitely the one taking the bait and paying the bill.
Disclaimer: I do not hold CRWV.
Open to opposing opinions—what do you think about the compute leasing business?