A Cup of Water on a Burning Cartload: Yields Hit a New High and Meta Rose 6.55 Per Cent the Same Day
On Wednesday $S&P 500(.SPX)$ closed 0.48 per cent lower, its third session in a row heading down. $Meta Platforms, Inc.(META)$ rose 6.55 per cent the same day to close at US$653.69. The whole market was backing away while one of the largest companies in it put on six and a half points, and those look like two unrelated things. They are one thing.
What is holding the indexes down is rates: Brent crude moved back above US$100 for the first time in two months, the inflation print lands on Friday, the ten-year Treasury yield was pushed to 4.85 per cent and the thirty-year broke above 5.30 per cent. The Treasury stepped in that day and raised the cap on its buybacks of ten- to twenty-year paper to US$6 billion. Yields did not come down. They set a new high instead. Scott Bessent, the Treasury secretary, has spelled out the limits himself: he cannot change the equilibrium price of Treasuries, he only wants less volatility, and a buyback is not quantitative easing and does nothing about the growth of the debt.
A cup of water on a burning cartload is exactly what that was: the cartload is supply, with the deficit sitting there and the issuance sitting there and oil above US$100 adding another layer to inflation expectations; the cup is US$6 billion of buying. The sizes do not meet. You pour it on and nothing even hisses.
There is a new bundle on that cart, and a record was set the same day: six American companies went to Europe to raise debt in a single session, with the reporting putting it down to AI borrowing crowding out the capacity of the market at home, leaving them to go abroad for it. Where that money is going is no mystery either. OpenAI is projected to spend US$750 billion on compute by 2030, and still says it does not have nearly enough.
So on Wednesday the equity side was buying AI demand and the bond side was carrying AI's borrowing, two faces of the same money. Buyers looking at US$750 billion that has to be spent went and bought the row that collects it; the bond market, looking at the same US$750 billion that has to be raised first, sent yields up.
The names collecting that money did rise on Wednesday. $SK hynix(SKHY)$ closed 7.05 per cent higher at US$198.63, a record high; $Micron Technology(MU)$ rose 2.75 per cent to close at US$1,027.77, back above the thousand mark; $SanDisk Corp.(SNDK)$ rose 1.51 per cent to US$1,764.17. Goldman Sachs's line that the worst may be over came out after Tuesday's close and only reached a price on Wednesday. Marvell rose 4.26 per cent to US$235.01, having raised its growth outlook three times since last December and still trading 28 per cent below its high. $Advanced Micro Devices(AMD)$ rose 3.04 per cent to US$521.10, and the names Bank of America put in its report were Micron and AMD. $Intel(INTC)$ rose 1.69 per cent to US$106.24, having moved High-NA EUV into production with ASML.
$Meta Platforms, Inc.(META)$ is a different case. It shipped something that day, an AI agent called Muse, and the sell side came out positive on it. The move should not be stretched: the stock is still down 0.79 per cent this year and 14.35 per cent over twelve months, behind the S&P 500 on both, and Wednesday only filled part of the hole back in.
That line does not explain the whole board: $NVIDIA(NVDA)$ fell 0.91 per cent and Broadcom fell 1.13 per cent, two of the names standing closest to the front of AI, lagging for a second session with no bad news of their own to be found. Alphabet fell 2.28 per cent with no named reason either, and all that turns up is the capital spending argument — one piece asking whether Amazon and Alphabet's combined US$420 billion of capital expenditure is worth it, and Dan Ives of Wedbush warning that 10 to 15 per cent of American data centre projects could be cut on political resistance. Those arguments press on the whole AI line rather than on Alphabet alone.
Some of it has nothing to do with rates or with AI. $SpaceX(SPCX)$ fell 3.86 per cent to US$147.55 and $Rocket Lab USA, Inc.(RKLB)$ fell 4.25 per cent to US$63.07 on an insider share unlock of about US$47.2 billion, with a further 319 million shares released and AST SpaceMobile down about 4 per cent the same day. None of that touches orders or launch schedules; the number of people who can sell just went up. $Apple(AAPL)$ is blunter still: the folding iPhone Duo was unveiled at US$1,999 to start, against the US$2,400 and US$2,000 circulating before the event, and the stock closed 0.28 per cent lower. Good news delivered on the day, and no money for it either.
Two more tests come in the next two days. Oracle reports after tonight's close, with the market looking for US$1.78 a share on revenue of US$19.53 billion, and the single line being watched is how much of its roughly US$638 billion AI backlog turns into cash — its capital budget already runs at more than twice the cash the business produces, and the stock closed 0.55 per cent lower on Wednesday. A day after that comes the inflation print. With oil sitting above US$100, that print is what decides whether the cup of water was ever going to be enough.
The above is personal analysis, not investment advice.
💬 【Talking Point】
$Oracle(ORCL)$ reports after tonight's close carrying roughly US$638 billion of AI backlog against a capital budget running at more than twice the cash the business produces. What would you need to see in tonight's numbers before you would put money behind a backlog that size?
💰 【Bounty】
Drop your view in the comments and there are Tiger Coins in it for you! 🎁
🔔 Better shared than saved — tag a friend and split the coins!
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Tonight, I would focus on three things: RPO conversion, AI revenue growth, and cash generation. If Oracle can show that major AI contracts are moving into actual revenue faster than expected, while keeping margins under control, the backlog starts to look like a genuine earnings engine rather than a headline number.
The bigger question is capex. Oracle is spending heavily to build AI infrastructure before customers fully pay for it. That creates a dangerous gap if financing costs stay high.
So I would not buy simply because the backlog is huge. I want evidence that AI demand is converting into cash faster than Oracle is converting cash into data centres.
If tonight’s numbers prove that equation is turning positive, I think the market could start valuing Oracle very differently.
@Marktomarket [龇牙]
I would watch cloud growth, AI demand, contract wins and especially free cash flow. If Oracle shows that AI investments are starting to generate stronger cash returns, I would be more comfortable investing behind the backlog. I also want to see whether management can maintain strong growth without continuously increasing its spending burden.
I remain bullish on AI infrastructure long term, but I do not want to chase the story based on backlog alone. I want the numbers to prove it first. If the results are strong, I would rather accumulate gradually than chase a big post-earnings spike.
@Tiger_comments @TigerClub @TigerStars @Marktomarket
具体来说,我会先看 OCI增速、未来12个月可确认的RPO比例、Capex增速、自由现金流。如果OCI继续高增长,同时未来一年可确认收入比例提升,说明这些订单不是“很远的故事”;如果Capex还在猛增,但自由现金流开始改善,那才真正说明AI基础设施投入进入回收期。
反过来,如果RPO继续创新高,但收入转化慢、Capex继续高于经营现金流很多,市场反而会更担心:订单越大,融资缺口是不是也越大。
所以我觉得今晚最理想的组合是:RPO继续增 + OCI高增长 + Capex增速开始低于收入增速 + FCF改善。这四个一起出现,6380亿美元才真正有含金量。
一句话:甲骨文现在不缺订单,缺的是“订单变收入、收入变现金流”的速度;今晚真正要看的是RPO质量,不只是RPO大小。