Post

Recently, AI infrastructure earnings reports have been coming in like a market fair, one after another. After reading them, all I want to say is: they are spending money faster than I lose money. AMD reported on August 4th, with data center revenue doubling year-over-year to 6.7 billion, EPYC setting records for five consecutive quarters, yet the stock price fell. Capital expenditure was 808 million, expected to be less than 300 million, tripled the forecast, and free cash flow was only 784 million left. Lisa Su said it's still early days for AI—don’t ask about break-even, just know it’s early stage. $SPCX SpaceX’s first earnings report after going public on the same day showed revenue of 7.8 billion, up 92%, with the AI segment soaring 247% to 2.56 billion. But looking at capex is shocking: total spending was 18.4 billion, with AI burning 15.8 billion. Starlink users doubled to 12 million, but ARPU dropped from $85 to $66. They’re also collaborating with Nvidia on Starmind AI, moving a data center to space because there’s not enough power on the ground, so they burn it in space. The big boss Nvidia reported on August 26th. Last quarter revenue was 81.6 billion, with data center revenue at 75.2 billion accounting for 92%, and networking business up 199%. This quarter’s guidance is 91 billion, with Bank of America raising estimates to 107-108 billion, and Vera Rubin starting shipments. Google Cloud’s Q2 revenue was 24.8 billion, up 82%, backlog at 514 billion, CFO said "demand still exceeds supply"—like telling my wife the wardrobe never has enough clothes. But Jensen Huang hasn’t been idle; on August 10th, he teamed up with Blackstone, Goldman Sachs, and four others to create a 500 billion AI financing platform, about the size of Slovakia’s annual GDP. He also invested 3 billion in Texas energy company Lancium, renting a 1GW data center. Selling shovels isn’t enough; he wants to build the power grid himself. But one thing makes me increasingly uneasy: two Nvidia customers account for 40% of revenue, three for 54%, namely Microsoft, Meta, and Amazon placing massive orders, funded by debt issuance. Now Nvidia is also lending money to AI factories—just like Cisco lending money to startups to buy their equipment in 2000. You might say this time is different, Google Cloud has a backlog of 514 billion, Copilot is generating revenue. But everyone is burning cash; when will the books balance? AMD’s capex was triple expectations, stock fell; SpaceX spent 15.8 billion on AI in one quarter, stock fell after hours. The market is voting with its feet—the story sounds good, but can I see the money come back? August 26th is a key date; if guidance exceeds 107 billion, it can still surge. If Jensen Huang can’t explain capex and customer concentration, the good news is fully priced in. For the crypto world, it boils down to two words: liquidity. Massive capex is backed by massive debt issuance, US Treasury yields are rising, putting pressure on risk assets. But conversely, AI computing power and crypto mining compete for electricity and data centers, and many mining companies have already transitioned to AI hosting. Don’t just watch the K-line; also watch how others spend money. Every penny they spend will sooner or later affect your position. #财报观察员：AI基建财报接力登场