wesley教授
wesley教授
Founder of Block Infinity, Poker player, Trader, Chinese whale, @drhashclub
868Following
4.3Kfollowers
Feed
Feed
A piece of data that got buried is much more interesting than those hype trading reports. The National Energy Administration just released electricity usage for September: Internet data services electricity consumption increased by 41.6% year-on-year, charging and battery swapping services by 55.4%, while electricity consumption for urban and rural residents is declining.
To translate: electricity is being crazily poured into data centers and computing power. Many will take this as ironclad evidence of an AI demand explosion—the electricity consumption is real, no doubt about that. But electricity consumption only proves money is being burned; it doesn't prove that money can be earned back. Oracle, Microsoft, and others are frantically building data centers and consuming electricity, while their free cash flow turns negative and they rely on issuing debt to fill the gaps. A steep upward electricity consumption curve and whether the return curve can keep up are two different matters. AI is not a scam, but the current valuation bets on the speed at which "electricity turns into profit," and historically, this speed has never been as fast as the PPT slides depict.
Citigroup raised SpaceX's target price from 200 to 240 tonight, with the bullish logic mainly being the strong launch pace and Starlink's cash flow starting to materialize. I won't argue the fundamentals; this company is indeed capable.
But for trading, you have to distinguish two things: a good company and a good price. For $SPCX, a pre-IPO perpetual in the secondary market, the real short-term variable isn't Citigroup's target price but the unlocking schedule—batches of shares will be unlocked over the next few months, with almost all unlocked by the end of the year. Early investors holding the chips and whether they can resist selling at high levels is what really determines the price. The analyst's price target increase adds emotional value, but the unlocking flood is a real supply factor. Consider both together; don't just focus on that attractive 240.
The cracks in AI infrastructure have this time appeared in the physical world. Australia's veteran data center operator NextDC plummeted 44% intraday today, triggered by something far from glamorous: a new project site in Melbourne faced strong opposition from local residents, power upgrade works delayed until 2027, and planning approvals postponed until after the state election. The company directly cut its profit forecast by one-third.
Half a year ago, its stock price was above 16 AUD, now just over 5 AUD. Everyone is talking about booming computing power demand and chip shortages, but no one reminds you: for data centers to actually be built, you need power, land, and approvals—each of which can hold you up for two years. The AI story looks infinitely big on PPT slides but is gradually worn down by the power grid and residents' votes in reality. This is not a demand falsification but a serious underestimation of costs and timelines. Those in $BTC riding the AI sentiment should also take a careful measure.
Tonight, the screen is full of storage research reports, all so uniformly worded it gives me chills: Nomura says the market is overly worried, supply tightness will continue until 2028; Citi maintains a buy on SK Hynix and says the upcycle is not fully priced in; Dell's CEO warns of even tighter memory in 2027; TrendForce straightforwardly calls for a 20% quarter-on-quarter increase in Q4 DRAM contract prices. All bullish, not a single dissenting voice.
Having traded for so many years, I believe one thing: when sell-side analysts stand so uniformly, not even bothering to feign disagreement, prices have often already fully priced in this narrative. The super cycle might be real, but "real" doesn't mean the current price is cheap. The reason I’m still holding this short position in storage is that I’m betting on the "consensus" containing some exaggeration. The more perfect the narrative, the more you need to watch for volume-price divergence. $MU this round, are you getting on board or catching the falling knife?
A news this morning: Nvidia is negotiating to invest hundreds of millions of dollars in Australian AI cloud service provider Firmus.
At first glance, it's just an investment. But when you connect the dots from the past few months—Nvidia investing in CoreWeave, investing in OpenAI, and now investing in AI cloud—you see a familiar loop: the chip manufacturer invests money in customers, and the customers use that money to buy the chip manufacturer's cards, which counts as revenue.
This play has a name, called vendor financing. The last time this was done so extravagantly was by telecom equipment vendors in 2000. Griffin's comment today that "the AI bubble is like the eve of the internet" refers to this cycle of creating demand for oneself. Whether the demand is real or just recycled will be clear once financing stops. Those watching $BTC shouldn't just look at the market; take a look at where this money is flowing.
Tonight, those watching the coins shouldn't be swayed by the war narrative around oil prices. Let's sort out the contradictory news from the Middle East today.
The Iranian Revolutionary Guard said two oil tankers hit mines and caught fire in the Strait of Hormuz; the U.S. Central Command directly denied this, saying the mines in the channel had long been cleared and that Iran was attacking with missiles and drones. Secretary of State Rubio added: the volume of oil transported through Hormuz is almost the same as before the war.
On one hand, they shout that the strait should be closed, but on the other hand, the oil transport volume hasn't dropped. That's how the war premium is created—it's based on headlines, not on an actual decrease in oil tankers. If a real war breaks out, oil prices will of course soar, but before that, the part of the premium inflated by panic headlines will sooner or later have to be paid back. Don't chase longs based on a mine news story that hasn't been confirmed yet.
Anthropic's IPO filing has been disclosed: revenue of $16.3 billion in the first half of the year, sounds impressive. But in the same document, there's another figure — a future computing power commitment that could reach as high as $750 billion.
$16.3 billion in revenue, signing a $750 billion computing power contract. This isn't business accounting; this is betting on the fate of the nation: betting that demand will always rise, betting that financing will always be available, betting that no one will suddenly pull the rug out.
I don't doubt that AI is useful. What I doubt is, when a company's expansion speed outpaces cash flow by two orders of magnitude, where the money to take over will come from, and when it will suddenly stop coming. The moment the bubble bursts is never because the technology fails, but because the money stops coming first. Those watching $BTC, remember this.
Griffin from Castle Securities dropped a harsh statement on Tuesday: It is now almost certain that we are in an AI bubble, a scenario very much like the eve of the internet bubble burst. This guy is the top Republican donor, who earlier this year called AI "trash" at Davos.
These days, the screen is full of Wood's calls for storage bottoming out and Bank of America calling the cycle bottom. Interestingly, those calling for a bottom are all sell-side and fund managers, while those calling a bubble are actually putting real money on the line.
I haven't moved a single short position in this round. I'm not betting on any particular news, but on the entire AI financing chain starting to settle accounts. $BTC sentiment can rebound first, but the final bill will fall on capacity and inventory. Which side are you on?
Another big influencer is out saying storage has hit the bottom. Bank of America says valuations have reached the cycle bottom, and Wood Sister says South Korean semiconductors have entered a buy zone. It sounds lively, but Hynix still closed down 2% today, and Micron and SanDisk followed downwards.
I've been trading for many years, and I have one experience: the more bloodletting the drop causes and the louder the bottom-fishing voices get, the more it means no one really dares to catch the fall. When it truly hits bottom, no one talks, volume dries up, and no one cares; it’s not a screen full of the words "buy zone" flooding everywhere.
I heavily bet on the storage leg in this AI deleveraging short, without reducing a single share. It’s not stubbornness on my part; the price votes for me every day. Shouting doesn’t matter, the closing price does.
Do you trust the big influencers’ "buy zone," or do you trust the K-line that keeps closing in the red?
Let's talk about the most frustrating thing during a rebound: holding a short position.
Today the market rebounded sharply, and the floating profit on my short position gave back a bit. Immediately, that voice in my head popped up—"Maybe I should close it first, then buy back when it drops again." This is the classic death trap for retail traders: the direction is clearly right, but they get shaken off by a single rebound, then watch helplessly as it continues down according to their original plan.
Card players understand this. When you hold a strong hand and your opponent bluffs occasionally, you don't fold just because they furrowed their brow. The real time to fold is when the logic changes, not when the price wobbles a bit.
My rule for trend trades is simple: if the invalidation level isn't broken, rebounds are just noise. Only by holding through the rebounds can you earn the full gains of the trend.