August CPI Prints 3.4% and Takes September Hike Odds to 72%. Oracle Grows Cloud 121% and Guides to $95B of Capex.
Friday, September 11, 2026  ·  Before Bell
Morning Note  ·  Equity Research

August CPI Prints 3.4% and Takes September Hike Odds to 72%. Oracle Grows Cloud 121% and Guides to $95B of Capex.

September 11, 2026
Fund X
Steven Yuan
August CPI +0.4% m/m, 3.4% y/y, in line
Core CPI +0.3% m/m against a 0.2% call
Core CPI y/y 2.4%, basically at target
Gasoline +3.9% m/m and +27.4% y/y
September 16 hike odds 72%, from a coin flip
30Y Treasury 5.373%, highest since 2007
Oracle OCI $7.4bn, +121%, RPO $664bn
Oracle FY27 capex $90bn to $95bn against $90bn of sales
Saudi output 6.238m b/d, lowest since 1990
Adobe Q4 guide $6.80bn to $6.85bn vs $6.96bn

Friday, September 11, 2026, written at 9:20 a.m. Eastern, ten minutes before the open and fifty minutes after the August consumer price report. Everything below is a pre-open mark. Oracle and Adobe reported last night, the Fed is in blackout, and the next scheduled data point that can move the September vote is the University of Michigan preliminary at 10:00.

Top Call: The Fed Is Being Asked to Hike Against a Price It Does Not Set

August CPI rose 0.4% on the month and 3.4% over the year, both in line. Core rose 0.3% against a 0.2% call and 2.4% over the year, in line on the annual figure and one tenth hot on the monthly one. That single tenth took September 16 from roughly a coin flip last week to 72% priced this morning, and it is the only number in the release that carries any information.

Look at the wedge. Headline runs 3.4% and core runs 2.4%, so a full hundred basis points of the inflation the committee is responding to is energy. The energy index rose 2.1% on the month and 16.3% over the year, gasoline rose 3.9% on the month and 27.4% over the year, and electricity actually fell 0.2%. Food rose 0.1% with groceries flat. Strip the barrel out and American consumer inflation is sitting within a rounding error of the target.

That is the whole argument, and it cuts against the tape’s conclusion. A core rate at 2.4% while headline runs a point above it describes a relative price shock, not a monetary one, and a rate rise does nothing to an Iranian tanker. Our view is that the September hike is now close to a done deal on the politics of a 27% gasoline print rather than on the arithmetic of a 2.4% core, and that the committee is about to buy option premium it may not need. We would not fight the pricing into Tuesday, but we think the dots and the statement language matter far more than the move itself, because the move is already in.

The honest counterweight is that one tenth. Yesterday we argued that a soft core PPI alongside a 24.1% diesel print meant the pass through had not arrived yet, and the very next print came in hot on core. That is the mechanism working. It arrived one month after we said it would take one to two quarters.

The Same Trade Is Driving Oracle Up 7% and the Thirty Year to a 2007 High

The most important thing on the page is not the CPI. It is that the thirty year Treasury sits at 5.373%, its highest since 2007, the ten year printed as high as 4.954% and is on the cusp of 5%, and Oracle is up 7% in the pre-market. Most desks are reading those as a risk asset story and a rates story. They are one story.

Oracle’s quarter was an unambiguous demand answer. Cloud infrastructure revenue rose 121% to $7.4 billion, total cloud rose 62% to $11.6 billion, revenue rose 30% to $19.3 billion, non-GAAP earnings of $1.92 beat $1.74, and remaining performance obligations reached $664 billion. The company delivered 850 megawatts of capacity and over 300,000 GPUs in the quarter, close to triple the prior one, at 97.9% utilisation. That is conversion, which was exactly the thing missing last quarter.

Now read the funding side. Capital expenditure was $28.5 billion in a single quarter, free cash flow was negative $5 billion despite operating cash flow of $23 billion, and full year guidance puts capital expenditure at $90 billion to $95 billion against revenue of at least $90 billion. A company spending more than its entire revenue on plant, carrying above $100 billion of long term debt and rated on the bottom rung of investment grade does not fund that out of earnings. It funds it in the bond market, and it competes there with the Treasury.

The scale is no longer a footnote. JPMorgan’s Cembalest puts the five large hyperscalers plus Nvidia at roughly $320 billion of debt raised in 2026, roughly $303 billion of it in ten year equivalent terms, which is about 68% of net new long dated Treasury borrowing this year. Goldman has lifted its 2026 investment grade gross supply forecast to $2.3 trillion from $2.1 trillion specifically on AI related supply, with AI issuers close to a quarter of the total. Bank of America counts $75 billion of AI related investment grade issuance across September and October alone, led by Meta at $30 billion and Oracle at $18 billion.

So the correct way to hold this morning is that the equity market and the bond market are pricing the same fact with opposite signs. Oracle’s beat is the proof that the buildout is real, and the thirty year at a nineteen year high is the price of financing it. Anyone long the AI complex and short duration as separate positions should understand they are running the same exposure twice. We are constructive on the demand leg and we think the discount rate leg is underpriced, which is a narrower position than simply being bullish the theme.

Crude Fell and the Physical Picture Got Worse

Brent gave back nearly four dollars to $103.78 and WTI fell to $99.23 after Brent topped $107 on Thursday for its highest close since May 19. The proximate cause is a technical correction plus another round of diplomatic signalling on the US and Iran track. Both indices remain up close to 10% on the week.

We would not treat the pullback as information. In the same week that the war premium came out, OPEC’s own reporting showed Saudi crude output fell 1.9 million barrels a day in August to 6.238 million, the lowest since 1990, with exports at 3.2 million barrels a day, the lowest in thirteen years, and total OPEC production down 900,000 barrels a day. The detail that matters most is that the kingdom reported market supply of 7.122 million barrels a day against production of 6.238 million, which means roughly 900,000 barrels a day of what customers received in August came out of stored crude rather than out of the ground.

That is a bridge, and bridges have a length. A producer covering a ninth of its supply from inventory can present a normal face to the market for some number of months and then cannot. Pricing a four dollar discount for a diplomatic headline while the largest swing producer quietly draws down its own tank farm is a positioning decision rather than a supply judgement, and we would use the giveback rather than follow it.

Asia Broke and Europe Did Not

Japan was the clean loser. The Nikkei fell 1,259 points, or 1.93%, to 64,011.34, having traded nearly 3% lower below 63,500 intraday, and the selling was concentrated in exactly the names that have led the year: Kioxia down 7%, Advantest down 6.5%, SoftBank down 4%, Tokyo Electron down 2.6%. The yen at 154.32 is off its seven month high, so this was not a currency session. It was the American long end arriving in Tokyo with a Bank of Japan meeting on the 18th behind it, and when memory and equipment names sell off 6% to 7% on a rate move rather than a demand datapoint, the message is that AI exposure in Asia has become a duration asset.

Hong Kong fell 149 points to 24,805 for a third consecutive decline and Shanghai finally moved, down 46 points to 3,888 after weeks of sitting still. Europe held. The Stoxx 600 rose 0.3% to 637.60 after closing at a two month low on Thursday, with the DAX at 25,429.85, the CAC at 8,149.65 and the FTSE at 10,638.53, and the index is still tracking its worst week since April. Technoprobe rose 4.7% on TSMC’s August revenue, which is the second consecutive session in which a European semiconductor name has traded on a Taiwanese monthly print rather than on anything local.

Overnight and Pre-Market Developments

Earnings That Matter

Oracle is the print of the week and it delivered on the half that was in doubt.

Metric Consensus Actual Beat/Miss
Revenue $19.1bn $19.3bn, +30% Beat
Non-GAAP EPS $1.74 $1.92, +30% Beat
Cloud infrastructure n/a $7.4bn, +121% Large beat
RPO n/a $664bn, +$26bn q/q Slower build
FY27 capex n/a $90bn to $95bn Negative

Our take: the composition inverted from last quarter in the most useful way. In June the story was $85 billion of sequential backlog growth against disappointing delivery, and the market sold it. This quarter backlog added only $26 billion while delivery nearly tripled to 850 megawatts and 300,000 GPUs at 97.9% utilisation, and the market is paying 7% for it. That is the right reaction, because a $664 billion backlog against $19 billion of quarterly revenue was never demand constrained, it was megawatt constrained, and the constraint just loosened. The reservation is entirely financial. Guiding capital expenditure to as much as $95 billion against revenue of at least $90 billion, with free cash flow negative in a quarter that produced $23 billion of operating cash flow, makes Oracle a levered infrastructure developer with a software attachment. At a 5.373% thirty year that is a materially more expensive business model than it was in June, and the credit rating is the variable to watch rather than the RPO.

Adobe beat and fell, which is the cleaner read on where the AI premium actually sits.

Metric Consensus Actual Beat/Miss
Revenue $6.70bn $6.76bn Beat
Non-GAAP EPS $6.09 $6.13 Slight beat
AI-first ending ARR n/a above $650m, +150% Strong
Q4 revenue guide $6.96bn $6.80bn to $6.85bn Miss

Our take: AI annual recurring revenue above $650 million growing more than 150% is a real number, and the stock fell about 2% after hours anyway because the fourth quarter guide came in roughly $120 million light. The reason is that Adobe’s AI revenue is additive while its core seat revenue is under pressure from the same technology, so the net is a guide down. Put Adobe next to Oracle and the market’s rule for this cycle is explicit: it pays for the layer that sells compute and discounts the layer that consumes it. We think that rule is currently too crude, but it is the rule, and Adobe does not get re-rated until AI revenue is large enough to visibly outrun the seat erosion.

Two smaller prints matter more than their size because they sit on the optical and space legs of this market. LightPath Technologies reported fourth quarter revenue up 73.8% with backlog up 197% to $110.9 million, and Frequency Electronics earned $0.41 against a $0.22 estimate on a record $129 million funded backlog. Infrared optics and precision timing are both supply constrained inputs to defence and datacentre interconnect, and two record backlogs in one morning from companies this small is a better read on order flow in the photonics and positioning chain than any large cap guide. On the other side, Zumiez guided the third quarter to $222 million to $226 million against $238 million expected, which is the same middle income consumer that Macy’s namesake brand described at 1.1% yesterday.

Key Events Today

When Event What matters
8:30am August CPI Done: 3.4% headline, 0.3% core m/m, 2.4% core y/y
10:00am UMich preliminary, est 51.0 Inflation expectations, the last pre-FOMC read
11:00am Cleveland Fed CPI nowcast September tracking with gasoline still elevated
1:00pm Baker Hughes rig count US supply response to a third week above $100
2:00pm August federal budget Issuance math behind a 5.373% thirty year
All day Fed blackout No official can respond to this print
Sep 15 to 16 FOMC with dot plot 72% priced, the dots carry the information
Sep 18 Bank of Japan 1.25% would be a 31-year high
Sep 30 Micron The memory cycle marked to a hiking Fed

The 10:00 number is the one to watch. With gasoline up 27.4% over the year, the Michigan inflation expectations series is the only pre-meeting datapoint that can still tell the committee whether a relative price shock is becoming an expectations problem. A one year expectation that jumps hardens the case for Tuesday considerably. One that holds leaves the hike resting entirely on a single tenth of core.

Geopolitical Risks

The diplomatic track has reappeared and crude has discounted it by nearly four dollars, which by our count is at least the fourth time this year that a signal short of an actual agreement has taken meaningful premium out of the barrel. The base rate on those signals is poor. Tehran has repeatedly denied direct negotiations while Omani intermediation has been described as days away since early August, and the Al-Azraq missile strike on Jordan earlier this week is not the behaviour of a party about to sign a shipping corridor framework.

What has genuinely changed is on the supply side rather than the diplomatic one. Saudi production at 6.238 million barrels a day is the lowest in thirty six years, exports are at a thirteen year low, and the gap between reported production and reported market supply says the kingdom is running its own storage down to hold service levels. Three things are worth tracking from here. Whether Aramco ever quantifies the Jizan damage, since continued silence past this week reads as a larger number. Whether Saudi inventory coverage becomes visible in export data next month, which is the point at which the bridge ends. And whether the tanker exchanges resume after this pause, because a pause that ends is worse for pricing than one that never began.

Away from the Gulf, the American and Canadian tariff exchange remains live at up to 50%, the Bank of Japan meets on the 18th into a yen that has already round tripped, and the Xi and Trump summit later this month still sits over unresolved rare earth and optical transceiver disputes.

Accountability

Yesterday we said the soft core PPI print had a one quarter shelf life, that the tape was pricing a measurement as a forecast, and that we would fade the relief. Core CPI came in at 0.3% against 0.2% this morning and the relief was faded within twenty four hours. We were right about the direction and wrong about the lag: we explicitly said pass through runs one to two quarters rather than one to two weeks, and it showed up in the next print. Being early on the conclusion and wrong on the timing is not the same as being right, and a fund that had positioned for a one quarter wait would have missed the move.

We also said the Oracle funding question was harder than the demand question. The demand answer was better than we expected and the funding answer was worse, and the stock is up 7%, so the market has told us it weights those two differently than we do over this horizon. We are keeping the view and marking the disagreement.

Everything above is a pre-open mark. The Michigan expectations print at 10:00 and the close on the thirty year are the two things that can change how this note reads by Monday.