Oil Back Above US$100, Diesel Cracks at a Record: What Do You Do With Energy Here?
On Thursday $Oracle(ORCL)$ closed 5.38 per cent lower at US$152.94, and only then reported. Its remaining performance obligations stand at US$664 billion, up from US$638 billion at the end of the previous quarter, a record. $S&P 500(.SPX)$ closed 0.58 per cent lower the same day, its fourth session in a row heading down and the longest such run since June. The order book set a record. The fall came before it.
That US$664 billion is work signed, not money collected. To get the work done, Oracle's capital expenditure ran to about US$28.5 billion in the quarter against US$8.5 billion a year earlier; free cash flow came in at minus US$5.4 billion; and in the same fiscal quarter it completed US$20 billion of at-the-market share sales. For fiscal 2027 it plans to spend roughly US$70 billion on data centres alone, against US$55.7 billion in the fiscal year before. The work has to be paid for up front, out of its own pocket.
Money got more expensive on Thursday, which is the one thing anyone fronting money cannot afford: August PPI accelerated to 5.4 per cent year on year, above expectations, the ten-year Treasury yield rose 11 basis points on the day and touched 4.96 per cent, a near-three-year high, and it closed at the session high; the two-year rose 13 basis points. Traders' odds of at least a 25 basis point hike at next week's 15–16 September meeting went from 64 per cent before the PPI release to 75 per cent after it.
What pushed the PPI up was oil. WTI jumped 7 per cent on the day to settle at US$102.78 a barrel, above US$100 for the first time since late May; Brent rose more than 6 per cent to US$107. The diesel crack spread broke US$110, a record, and retail diesel is approaching US$6 a gallon. What is going up is the cost of moving things from one place to another, and nobody gets out of paying that.
The Treasury was buying the same day. Its buyback of ten- to twenty-year paper accepted US$5.19 billion in the end against an announced ceiling of US$6 billion, and that ceiling was already triple the previous US$2 billion limit for a single operation. Bids came in at about US$10 billion against a usual US$20 billion, meaning far fewer holders offered to sell. Scott Bessent, the Treasury secretary, said this was not weak demand but prices that were not cheap enough, and that he was deliberately pacing the purchases; holders of long bonds declining to sell them back is, to his reading, confidence. His verdict on the market was that conditions are good. Peter Boockvar of OnePoint BFG Wealth Partners said some investors had looked for US$7 billion to US$8 billion, and that the rise in yields reflected the disappointment.
Bessent is not simply talking tough: the thirty-year auction that day stopped 3 basis points through the when-issued level, and the primary dealer take-down was the lowest on record, meaning end investors took the paper themselves. Gold did not run up either, falling 1.7 per cent to US$4,324.15 an ounce — if anyone were worried about whether the debt gets paid, gold would not be going that way. Core PPI was 4.6 per cent year on year, and what rose was mostly energy. What was being repriced on the day was the rate, not America's ability to pay.
So on Thursday nobody got paid for good news that needed capacity built out of its own pocket before any of it could be collected. Jensen Huang was talking about AGI having arrived and US$4 trillion of AI infrastructure to come; $NVIDIA(NVDA)$ closed 2.26 per cent lower at US$218.36. Michael Burry closed out his Nvidia and Palantir put positions, so even the short side stepped away, and $CoreWeave, Inc.(CRWV)$ still fell about 5 per cent, with $NEBIUS(NBIS)$ closing 5.09 per cent lower at US$228.11. The Philadelphia Semiconductor Index dropped 2.66 per cent.
Piper Sandler initiated coverage on two names that day, one at Neutral and one at Overweight, and both fell. It started Intel at Neutral with a US$110 target, its analyst David O'Connor liking the CPU business but wanting a better entry point at this valuation and looking for less than 10 per cent upside over a year; Intel closed 5.57 per cent lower at US$100.32, giving up the US$100 mark. It started AMD at Overweight the same day, and $Advanced Micro Devices(AMD)$ closed 3.36 per cent lower at US$503.60. Marvell closed 3.43 per cent lower at US$226.96, a company that has long called itself the Switzerland of AI. What was being sold that day was not any particular company. It was exposure.
Memory gave back everything it had gained the day before. $SK hynix(SKHY)$ closed 5.20 per cent lower at US$188.30, one session after setting a record high; $Micron Technology(MU)$ fell 4.90 per cent to close at US$977.41, losing the thousand mark; $SanDisk Corp.(SNDK)$ fell 4.06 per cent to US$1,692.59. Not one part of the pricing story behind that row was contradicted. They just have to pay for their own expansion too.
Only one order book got paid for on the day, and it was Apple's. $Apple(AAPL)$ closed 3.56 per cent higher at US$326.57, one of the few large caps to rise on a day this heavy, on the sales forecasts the sell side put out after its event: Goldman Sachs's central case for the iPhone Duo in 2026 is 14 million units, its bull case 35 million; the Duo starts at US$1,999, tops out at US$3,199 and goes on sale on 23 October. Apple's good news needs nothing fronted — from October it collects from consumers, rather than building a data centre and then waiting for a customer to pay. IDC sits on the other side of it: foldables will account for less than 3 per cent of global smartphone sales in 2026.
That line does not explain the whole board either: $Alphabet(GOOG)$closed 0.61 per cent higher at US$330.39, and its capital spending is anything but small, yet it did not fall with the rest, with no good news of its own to be found. $SpaceX(SPCX)$ closed 0.43 per cent higher at US$148.18, its chief financial officer saying a very strong alliance with Nvidia could support a 10 gigawatt AI push and that annual recurring revenue could reach US$100 billion by year end — capacity that has to be built before anyone pays for it, and the stock was not sold off. $Meta Platforms, Inc.(META)$ closed 1.42 per cent lower at US$644.38 having just removed a large risk with an US$18 billion settlement, and fell anyway.
Oracle rose as much as almost 8 per cent after the close before paring to a 4.38 per cent overnight gain at US$159.64 — revenue of US$19.3 billion was up 30 per cent year on year, adjusted earnings came in at US$1.92 a share, both ahead of expectations, and guidance for the coming quarter is revenue growth of 30 to 34 per cent with cloud revenue growth of 65 to 71 per cent. The CPI print lands tonight. What this row of front-payers ends up paying in interest is not answered on that US$664 billion order book. It is answered in tonight's data.
The above is personal analysis, not investment advice.
💬 【Talking Point】
WTI settled at US$102.78, up 7 per cent on the day and back above US$100 for the first time since late May, with the diesel crack spread through US$110 and retail diesel near US$6 a gallon. How are you handling energy exposure from here?
💰 【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.

WTI重新站上100美元、柴油裂解价差冲高,短期当然对能源股有利,但现在油价里已经包含了很高的 地缘政治溢价。这种行情最危险的地方是:一旦冲突缓和、运输恢复,原油价格回撤会非常快,所以我不会在100美元上方去重仓追期货或高杠杆油气标的。
如果一定要做,我会更偏 XOM、CVX这类综合能源,或者现金成本低、资产负债表好的上游公司,而不是纯粹赌油价继续冲。炼油股则要看 裂解价差能不能维持,因为现在柴油利润很强,但这同样是高波动变量。
我最关注三个信号:霍尔木兹/中东实际供应有没有中断、商业库存有没有持续下降、裂解价差能不能高位维持。只有这三个一起出现,100美元以上的油价才更像基本面,而不是单纯战争溢价。
一句话:现在原油涨的是供应风险,不是需求繁荣;我宁愿赚能源公司的现金流,也不愿追最贵的那桶油。