Editor’s Note
The AI infrastructure boom became easier to see this week, and so did its cost.
Oracle has enormous demand, but negative free cash flow and heavy data center spending. At the same time, oil moved above $100, inflation stayed uncomfortable, and higher rates put more pressure on valuations and leveraged balance sheets.
Oil Put Rate Hikes Back in Play
The rate story changed quickly.
August payrolls rose 162,000, compared with expected growth in the low-to-mid 50,000s. Unemployment held at 4.1%, while wage growth cooled to 3.1% year over year.
Strong employment gave the Fed room to focus on inflation. Oil then made that job harder.
Brent began the period around $96, climbed near $97, and later moved above $100 as the US-Iran conflict disrupted shipping. One report put Brent just over $100 after a 2.1 per cent rise. Other reports said it briefly moved above $105 to $107.
Inflation was already running above the Fed’s target:
- August PPI: 0.4% month over month
- August PPI: 5.4% year over year
- Headline CPI: 3.4% year over year
- Core CPI: 2.4% year over year
The estimated chance of a September Fed hike moved from roughly the low-50s to about 58-65% after payrolls. It reached roughly 70% after PPI, while another brief put it near 90% after CPI.
Bond investors pushed in the same direction. The 10-year yield rose to just below 4.86 per cent, and the 30-year yield reached 5.35 per cent. Treasury’s $6bn buyback of longer-dated debt was bigger than its $4bn pledge but below Wall Street estimates of $8bn and $10bn.
Higher oil feeds inflation. Sticky inflation strengthens the case for higher rates. Higher rates then make expensive stocks and indebted companies harder to defend.
AI Demand Is Huge, but So Is the Bill
Oracle gave the clearest view of both sides.
The company reported:
- Revenue of $19.3B
- Adjusted EPS of $1.92
- RPO of $664B, up $209B year over year
- Cloud Infrastructure revenue of $7.4B, up 121% year over year
- Free cash flow of -$5.4B
- CapEx of $28.5B
Oracle also booked more than $30B of AI cloud contracts in Q1. Leif | Investing reported that Oracle expects 50% of its backlog to convert within the next 36 months.
Demand is not the uncertain part. The open question is how much profitable revenue arrives after all the spending.
Oracle previously told investors it would invest $70bn during the coming fiscal year, up from $55.7bn in the year ended May 31. It also completed a $20bn share sale as part of a $50bn financing package. S&P downgraded Oracle’s credit in July to one notch above junk status, citing reliance on a few customers and an uncertain path to profitability in its AI data center business.
Capacity Became the Bottleneck
Oracle was not alone.
Nebius said customers were asking for tens of thousands of GPUs, with “much more ask than we can physically serve.” It said AI compute demand was stretching into 2028 and that it could sell all of 2027 today.
Microsoft plans to 3x data center capacity, according to Kalshi. Google plans to invest €13B in Finland AI infrastructure over the next 2 years, according to Bloomberg.
Jensen Huang described the AI infrastructure buildout as “not tens of billions, but tens of trillions.” He said neoclouds are becoming critical because they secure “land, power, and shell” after hyperscalers have exhausted much of that capacity.
The pattern is clear: demand is pushing companies to secure chips, power, land, buildings, and financing. Investors now have to follow those constraints as closely as customer orders.
Nvidia’s size makes this more important for the wider market. It represents approximately 8% of the S&P 500’s market value. At $5.3 trillion, Nvidia is worth the equivalent of 16.3% of US GDP, and its market value is larger than 5 of the index’s 11 sectors.
Higher Rates Exposed Weak Valuation and Balance Sheets
The same interest-rate pressure showed up in very different businesses.
McDonald’s dividend yield was 2% at a stock price of 300, while the 10-year Treasury was at 4.77. The speaker called McDonald’s expensive and far from offering a margin of safety.
Vonovia was a balance-sheet version of the same problem. It had 42 billion in debt, an average debt cost of 2%, an average maturity of six years, and leverage of about 14 times against a target of 12 times.
The business itself was described as stable, with 99.6% collection rates and growing rents. But the speaker said refinancing at higher rates could break covenants and leave shareholders with zero, while acknowledging that nobody knows whether this will happen.
Higher rates do not need to break the underlying business to hurt the stock. They can reduce the price investors accept for slow growth or make existing debt much more dangerous.
Apple Changed the Product, Adobe Did Not Change the Reaction
Apple introduced its first foldable smartphone, the iPhone Duo, starting at $1,999. New CEO John Ternus called the iPhone the best device for AI.
The memory requirements also increased. The new A20 Pro has 50% more memory bandwidth, and one source said Apple is using Micron LPDDR5X mobile DRAM with 12GB of RAM.
Adobe delivered revenue of $6.8B, up 13% year over year, and adjusted EPS of $6.13, up 15% year over year. AI-first ARR growth was 150%+, yet another brief reported that the shares fell 1.7%.
Good operating numbers were not automatically enough for the stock.
Counter-Thesis and Risk Watch
George Gammon said the market is in a “massive AI bubble” and is due for another crash.
Wall Street Millennial raised a more specific concern: Nvidia circular financing could create “massive reported revenue far in excess of end demand for the product.” DeepSeek created another risk when it said its latest model required less high-bandwidth memory, after which SK Hynix and Samsung Electronics fell.
The same source warned about renewed SPAC activity. Citing Jay Ritter data, it said average first-year SPAC returns from 2012 through 2025 were -46%, average three-year returns were -58%, and the average 2021 cohort SPAC fell 73% after three years.
China’s humanoid robot market offered another warning about demand that may be less developed than valuations imply. @wallstreetmillennial said Unitree was valued at about 220 billion CNY, equivalent to about 33 billion US dollars, after generating approximately $250 million of revenue in 2025. Unitree CEO Wang Xingxing said humanoid robots perform well in tests but can quickly fail when objects or surroundings change slightly.
Looking Ahead
Can Oracle turn its $664B backlog into profitable cash flow fast enough to justify its spending?
Can AI suppliers add enough power, land, buildings, and GPUs to serve demand already stretching into 2028?
And if oil stays above $100, how much harder does the rate environment become for expensive stocks and leveraged businesses?