September CPI Prints +3.0% YoY — Restaurant Pricing Power vs. the BLS Data Crater

I was standing at the espresso machine at 8:31 a.m. ET when the September CPI tape ripped across the desk — the lone BLS release we are getting until the shutdown ends, rushed out specifically so Social Security can compute its 2026 COLA. Headline +0.3% month-over-month, +3.0% year-over-year, per the BLS news release. Core +0.2% MoM, +3.0% YoY. Food away from home up 3.7% YoY. Energy bounced. Shelter cooled a hair. The print was a tenth softer than the +3.1% consensus on the headline, and the two-year breakeven moved about three basis points before lunch.

And then nothing. The BLS reschedule notice is explicit: “no other releases will be rescheduled or produced until the resumption of regular government services.” We get one data point, designed for one regulatory purpose, and then the lights go back off. This is the macro equivalent of a single tape from a closed exchange — directional, but the order book around it is gone.

My contrarian read: restaurant pricing-power narratives are about to get a stress test that the sell-side is not ready for. The CPI print looks benign at the headline. Underneath it, food away from home is still running a full point hotter than headline, shelter disinflation has stalled at a level that does not solve the consumer’s cash-flow problem, and the same week Marriott published a 22,266-respondent EMEA Ticket to Travel report showing 50% of EMEA travelers have now used AI to plan or research a holiday — up from 41% in 2024 and 26% in 2023. That is the demand-side curve that gets pointed at hospitality margins when chain operators step on stage for Q3 calls next week. The bifurcation between operators who have something to say about AI-driven demand and those who do not is now sharper than at any point in 2024.

Let me show you the math.

What the September print actually says

Start with the topline. Headline CPI +0.3% MoM, +3.0% YoY. Core CPI +0.2% MoM, +3.0% YoY. Energy contributed the upside surprise relative to August on the headline. Shelter — the single biggest line in the basket — printed +0.2% MoM, the same cadence we have seen since July. Annualized that is about 2.4%, which is below the 3% headline but above the 2% target that monetary policy is theoretically still aimed at.

Food at home: +0.3% MoM, +2.7% YoY. Food away from home — the line that matters for the operators on my coverage list — +0.3% MoM, +3.7% YoY. That spread between food-at-home and food-away has been the single most consistent feature of this cycle. It compressed for two prints in mid-2024, the bull case I wrote up at the time, and it has since re-widened back to roughly 100 basis points. CNBC’s wire write-up hung its hat on the headline being a tenth below consensus. That is the right one-line summary for rates desks. It is the wrong summary for anyone modeling restaurant comps.

Why the wrong summary? Because food-away-from-home is the operator’s revenue line, and food-at-home is the substitution threat. When the spread is positive and widening, the consumer’s effective price for choosing the restaurant is rising faster than the price of staying home. The spread has now been positive for fourteen consecutive months. The pricing power story rests on the idea that consumers tolerate that spread because the experience is differentiated or the convenience is non-negotiable. That story works at 60 basis points of spread. It is a more fragile story at 100.

The other piece of the print worth dwelling on: services ex-shelter, which is where wage pressure shows up in the basket, ran +0.3% MoM, +3.6% YoY. Wage growth in the latest Atlanta Fed wage tracker has cooled to roughly 4.1%, so services ex-shelter is now running about 50 basis points below wage growth. That gap is the operator’s labor-margin compression in real time. It explains why every public limited-service chain has been emphasizing “managed labor hours” on calls since Q2.

The data crater is the story, not the print

I want to be careful here. The September CPI print is real data, and it is the only real data we are going to get on the consumer until the shutdown ends. Everything else — PPI, retail sales, the employment situation, JOLTS, the BEA personal income and outlays release that would normally land at the end of this month — is in a queue. The reschedule notice does not promise a sequence. It does not promise that any of it gets produced at all in October.

For an M&A desk, this is consequential in three ways.

First, deal financing committees that mark to a forward inflation path now have to mark to a six-week-old reference point and a single September data point. The committee at the bank I had coffee with on Tuesday is using a 2.9% blended forward, anchored on the August CPI plus a model of the September print they were forced to construct internally. The actual print at 3.0% is inside their band, but the absence of PCE — which is what the Fed actually targets — means the path of policy from here is being argued from priors rather than data.

Second, public-comp analysts who run cross-sectional models against CPI components are now running models with a hole in the time series. Most of the comp shops I respect will note the missing data in a footnote and proceed. The dispersion in their estimates for Q4 same-store sales at the chains is going to widen as a result. That dispersion is opportunity for anyone who has a stronger view than the consensus.

Third, and this is the one I want to spend the most time on, private-company sellers who are mid-process are now negotiating against a buyer’s market for macro narrative. Every PE buyer I have talked to this week has used some version of “data uncertainty” as a justification for re-papering price. None of them have used “data uncertainty” as a reason to walk. It is a discount lever, not a deal-killer. Sellers who do not understand that distinction are going to leave 50 to 150 basis points of EV/EBITDA on the table over the next 60 days.

The Marriott EMEA tell and what it means for chain operators

Same week, different release. Marriott’s 2026 EMEA Ticket to Travel Report — conducted by Mortar Research across 22,266 adults — landed with a single statistic that should reshape how analysts model technology adoption in hospitality: 50% of EMEA travelers have now used AI to plan or research a holiday. That number was 41% in 2024 and 26% in 2023.

Let me flag the interpretation, because the headline number is doing more work than it should. “Have used AI” includes every consumer who has typed a destination question into ChatGPT, Gemini, or any of the embedded copilots that have shipped in the booking flow itself over the last 18 months. It does not mean 50% of travelers are booking through AI agents. It does not mean 50% of revenue is now AI-mediated. It means the top of the funnel has flipped — from search engines and curated content to conversational interfaces — for half of the addressable customer base in EMEA in the span of roughly 24 months.

For chain operators, this is the demand-side curve that gets pointed at margins on next week’s earnings calls. The mechanism is straightforward: when consumers are using conversational interfaces to plan, the brand-level moat shifts from SEO and paid search rank to whatever the model retrieves when asked. The properties that surface in those retrieval flows capture demand at near-zero CAC. The properties that do not surface lose share quietly, and the loss does not show up in OTA dashboards.

I have written about why I am skeptical of operators paying for an “AI premium” on the cost side in a forthcoming May piece on The Case Against an AI Premium. The Marriott data does not change that view. It does, however, change the conversation about the demand side. The right framing for next week’s calls is not “are operators investing in AI.” It is “are operators capturing AI-mediated demand at zero incremental marketing cost.” Those are very different questions, and the second one has a very different valuation implication.

The bifurcation, to put numbers on it: if 50% of EMEA top-of-funnel is now AI-mediated, and if AI retrieval favors operators with strong structured data, high-quality content, and clean review signal, then the spread between best-in-class and median operators on net customer acquisition cost is widening at a rate that has not yet been priced into multiples. My back-of-envelope math: a 200 basis point CAC advantage on a chain doing $5B in room revenue at 15% blended marketing intensity is $150 million of annual margin that does not appear in the comp set. That compounds.

What I expect to hear on Q3 calls next week

Here is my framework. I am sorting the chain coverage into three buckets going into earnings.

Bucket one — operators who will lean into pricing power and ignore the AI demand-side story. These are the chains whose Q3 prepared remarks will include some variant of “we continue to take low-single-digit price” and whose Q&A will deflect tech questions toward labor savings rather than demand capture. They will trade in line on the day, and they will quietly underperform over the next four quarters as the food-away spread continues to compress their traffic. Watch for any operator that pre-announces softer traffic in the back half of October — I am especially attentive to limited-service operators where management has historically guided down on traffic when the consumer environment turns. The first chain that breaks the pricing-power frame and starts talking about traffic instead of average check is the canary in this cycle.

Bucket two — operators who will frame the AI conversation around cost. Drive-thru voice ordering. Back-of-house labor scheduling. Inventory forecasting. All real, all margin-accretive, all already priced in. The risk here is that management’s enthusiasm about cost savings distracts from the demand-side question. If you are an analyst, the right question to ask in Q&A is not “what is the ROI on your voice AI pilot.” It is “what percentage of your top-of-funnel reservations are now sourced from conversational interfaces, and what is your retrieval visibility versus your peer set.” Nobody is going to answer that question well next week. The operators who can answer it at all are the ones I want to be long.

Bucket three — operators who get the demand-side mechanic. This is a small bucket, and most of the members are full-service hotel chains rather than restaurant chains. The thesis here is that they have years of structured property data, clean review signal, and content infrastructure that was originally built for SEO but happens to be exactly what retrieval-augmented generation surfaces. Their CAC trajectory over the next four quarters is going to look very different from the median.

The Four Margins framework I am putting out in an upcoming May framework piece on Four Margins of AI in Hospitality is the lens I use to keep the conversation honest. The relevant margin for next week’s calls is the demand margin — the cost of acquiring a customer relative to the lifetime value of that customer. That is the margin where the Marriott data lands hardest. The other three margins matter, but they are second-order to the question of whether the operator can defend top-of-funnel economics in a conversational-search world.

How I am positioning around the data crater

A few practical notes for the desk.

On the CPI print itself: I am not changing my forward inflation path. 3.0% headline is consistent with the trajectory I have been running, and I do not have enough information to update on the September data alone. I am marking down my confidence interval on Q4 estimates by roughly 30%, which is a way of saying I am sizing positions smaller until PPI and PCE catch up.

On restaurant chain comps: I am underweight operators in Bucket One above. The food-away-from-home spread at 100 basis points is a slow squeeze, not an acute one, but the calls next week are where the squeeze starts to show up in language. Listen for the word “value” used more than three times in prepared remarks. That is the tell.

On hospitality: I am overweight operators in Bucket Three. The Marriott EMEA data is a tell for the global category, not a Marriott-specific datapoint. The mechanic — top-of-funnel migrating from search to conversation — is platform-agnostic and chain-agnostic. The winners are the operators with the cleanest structured data going into the migration.

On the data crater itself: I am long volatility on rates through year-end. The single CPI print plus the absence of PCE is the worst possible setup for a Fed that wants to cut into a soft landing. The path-of-policy distribution has widened, and option-implied vol has not adjusted to it yet.

The bottom line is this. We got one data point this morning, and it was specifically engineered to make Social Security’s COLA calc work. Everything else is in a queue that may or may not clear. The narrative vacuum is going to be filled by Q3 prints next week, and the operators who can speak fluently about AI-mediated demand are going to separate from the ones who cannot. That separation is mispriced in the multiples today. It will not be mispriced for long.

— Marcus edits The Bottom Line for TableTransfers. Tips: ma@tabletransfers.com.