Herman Jin

Herman Jin

253 Twitter followers

Follow

Ex-Goldman Sachs Asia FICC Exec & Co-founder of Zen Family Office. Now diving into crypto, exploring blockchain's potential.

Posts

OAI and Anthropic already have a 90% gross margin in inference. Jensen mentioned yesterday that AI has passed its inflection point, which means this: because selling tokens is becoming less profitable, infrastructure development is booming. Taking OAI as an example, its revenue this year is 50 billion. Revenue, gross profit, and growth rate are all accelerating. Based on the budget, including the cost of training and preparation, the expenditure in 2030 will only be 100 billion. But this round of financing will raise 150 billion. Why can't capex be sustained?

$nvda 也走不远，一旦市场确信高CAPEX不可持续，估值就会自动下调很简单道理：四大已经把2027年接近100%的自由现金流用来CAPEX，请问2028年怎么办？2030年呢？ $nvda的需求是CAPEX的二阶导投资绝对值不变，对$nvda来说就是需求灾难性下降。哪怕投资增速从80%下调到50%，那需求增速也是断崖式下降。 x.com/yiantony/statu…

Qwen's idea of sending milk tea is good; I guess he'll be trading stocks on Monday. Actually, the model is outdated; figuring out how the agent works first is also a viable approach. It's very likely that all businesses will need to provide APIs in the future, and product showcase dialogues will all be AI-driven. The asset-heavy model is like a much higher-margin telecom operator. So what about the TMT companies that are currently providing showcases? Are they just going to provide APIs and become AI suppliers?

The reason why Capex wasn't punished by the market is that no one backed down. The model itself is like the Five Hegemons of the Spring and Autumn Period and the Seven Warring States Period; behind it is a "land grab," a Monopoly game. In the early stages of AI, buying GPUs is like buying land; stopping investment is like being out of the game. A market worth trillions—replacing 20% of 700 million white-collar workers—even losing this little bit of money as an entry ticket is worthwhile.

orcl好比是玩大富翁他开了N个乌龟卡，把前面最好的地都买了但是其他人发现家伙现金不够了

The current market narrative is not working. The current "split narrative" suggests that AI will devour software, and that excessive AI adoption through Capex will lead to bankruptcy. Therefore, MSFT is both software and involved in Capex. We need to wait for a narrative reversal, until everyone realizes that AI isn't just devouring software SaaS, but rather all jobs. Then it will be a win-win situation. Financial agents continuously spend 10,000 tokens per hour, at 0.1 USD. Later, the SaaS will be sold to clients at a rate equivalent to a 20% increase in workload. Where will the money come from?

但反过来想一下，不管哪边赢微软都有办法活下去甚至活得更好

MSFT was unfairly punished for its software pricing. The problem is: no one can be sure what the average PE ratio for software is. In fact, everyone is worried that any multiple they assign is too high. On the one hand, they don't believe in AI; on the other hand, they feel AI will slaughter the software industry. In reality, software will become like "airlines" rather than "pharmaceutical companies" in the future. The days of selling high margins with copyright are over. Every company needs heavy asset investment, operation and management, and to sell tokens.

Buy when prices drop, hold on after buying! Why are hyperscalers like Orcl still continuing despite CapEX's severe penalties? Because they know AI won't replace SaaS, but rather 30% of the 700 million mental laborers. Today's hundreds of billions are like "buying land" in Monopoly—borrowing from banks, selling credit cards, they'll still participate. No one's stupid enough to ignore an NV credit card quota. PS: Just bought BTC again at 66,306.

@JavierR19950216

Nvda我拿着都慌了，看那些分析师一个个很笃定把目标价调到250以上，但是看最近几个月，尤其最近几天的表现，真是焦虑😟

Bro, what happened to you? I've been giving bad jinxes lately, so I've decided to stick to macroeconomics and not comment on individual stocks for now. P.S. When your gross profit margin is lower than NV's, your growth is lower than NV's, you're losing market share, and your products are a generation behind, and you're still telling a catch-up story... that's what you get. twitter.com/ShanghaoJin/status...

Now it's all summary execution. It's not just software; how many AI companies have failed? Is the software guilty, and are OAI orders guilty?

JPM on Software: “We are now in an environment where the sector isn’t just guilty until proven innocent but is now being sentenced before trial,” said JPMorgan analyst Toby Ogg. Ogg met more than 50 investors across Europe and the US over two weeks and said he found that they had

You can take a look at the MOUs of NV and OAI. Don't spread misinformation; they say buying cards provides financing, but they never mentioned a PE round investment of 100 billion. You probably got a lot of the logic from Twitter. Read more of what I write and less of what's circulating in the crypto, haha.

OAI's B2B revenue increased by 1 billion in January, and Claude is also performing exceptionally well. The market is worried that they will eat up all of the company's software spending. Actually, they are not here to replace software. AI's competition is precisely with "intelligence," and only secondarily with "assistive intelligent tools" (software). The "containers" (such as Excel) within software remain essential. Some companies have been unfairly punished in this round. But intelligence—is ultimately about people. AI B2B revenue is eating up your jobs.

软件、主件的大科技逢跌买起来。走过这段，让常识回归。

Loading..