Sweetgreen's Infinite Kitchen Calculus After a Soft Q1

I was halfway through Sweetgreen’s Q1 2025 transcript on a Tuesday morning in late May — coffee on the desk, one finger on the capex line — when the number that should have been the headline showed up not in the financials but in a footnote-shaped remark about Hingham. Thirty percent restaurant-level margin. Month one. A brand-new build outside Boston, the third Infinite Kitchen in the system, opening into a Q1 where the rest of the comp base was running negative.

That juxtaposition is the case. Q1 same-store sales were down 3.1%. Restaurant-level margin compressed. Revenue grew, but on the back of unit count, not throughput. And the response from management was not to slow the automation rollout — it was to accelerate it. The 2025 plan calls for 40 new restaurants. Twenty will open with an Infinite Kitchen. That is not a pilot ratio. That is the new-build standard, asserted in the middle of a soft quarter.

The contrarian thesis is this. The wobble is real. Q1 was not a one-off and the macro pressure on premium fast-casual lunch traffic is not going to evaporate by Q3. And the response — doubling down on Infinite Kitchen, with Hingham as the proof point and a capex line moving from $3.4M to $19.1M — is the right call. It is also a call the equity market is not yet reading correctly, because the public narrative is still focused on the comp number and not on the unit-economics math the IK rollout is actually solving for.

This piece is about that math. What Hingham appears to be telling us. What the capex jump is buying. What the labor and COGS deltas look like when you stack them. And where the failure modes live.

The Q1 wobble, sized honestly

Let’s put the quarter on the table. Sweetgreen reported Q1 2025 revenue of $166.3M. Same-store sales were down 3.1%. That is the negative print that drove the post-earnings reaction. Restaurant-level margin compressed against a difficult prior-year comparison, and management acknowledged on the call that traffic was the principal drag, partially offset by menu mix.

The temptation reading a quarter like that is to do one of two things. Either decide the model is broken and the automation story is a distraction from a demand problem. Or decide the comp is noise and the automation thesis carries through. Both are wrong, because they collapse two different questions.

The first: is Q1 a demand signal that should change the build plan? Partly. Premium fast-casual is sensitive to white-collar lunch traffic, and the office-return curve has plateaued in several of Sweetgreen’s denser MSAs. The chain’s exposure to weekday lunch in urban cores is real, as is the price ladder it sits on relative to Chipotle, Cava, and the better regional bowls.

The second: given that demand signal, what is the right capital response? Here the math gets interesting. If the problem is not enough revenue per restaurant, you have three levers. Drop price (which is what most operators in a wobble do, and what compresses RLM further). Cut SG&A or close stores (which Sweetgreen is not doing because the unit count is the equity story). Or shift the cost structure of the box itself — labor as a percentage of sales, COGS as a percentage of sales, throughput per hour — so the same revenue line produces a better margin line.

That third lever is what Infinite Kitchen is for. Q1 made the case for that lever, paradoxically, stronger rather than weaker. A chain growing comps at +5% can amortise labor inflation through pricing. A chain running -3.1% has to take cost out of the box. The IK rollout is not a story the company is telling against the Q1 print. It is the operational answer to the Q1 print.

The AInvest write-up of the quarter called it a quarter of “contradictions” — soft comps alongside accelerated automation. The framing is right. The contradiction is the strategy. Management is choosing the structural lever over the cyclical one, loudly enough that the deck is meant to be the answer to anyone asking why the comp print is being absorbed rather than fought.

Hingham as the case study, read carefully

Hingham, Massachusetts, opened earlier this year as the third Infinite Kitchen build in the network. It is a brand-new restaurant — not a retrofit of an existing site — and it is the unit management has been pointing to most consistently as the proof of where the IK thesis lands when the deployment is clean.

The number that got my attention is the thirty percent restaurant-level margin in month one. RLM at thirty percent, in the first month of a new opening, is exceptional. The chain’s system-wide RLM in Q1 was nowhere close. New openings are usually drag-on-margin events for at least the first quarter because labor schedules over-staff while the team learns the box and waste runs high while production rates calibrate. To open at thirty percent suggests three things are happening simultaneously.

First, the labor model at IK Hingham is meaningfully different from a non-IK new opening. Management has spoken on prior calls about labor savings on the order of seven-plus percentage points in IK stores versus the traditional makeline. I’ll come back to this number in the next section because it is the load-bearing claim of the entire rollout, but for now take it at face value: an IK store opens with structurally less labor than a non-IK store, and the savings show up immediately in the period-one P&L because there is no pre-IK staffing pattern to unwind.

Second, throughput at peak is higher. The makeline is the bottleneck in a Sweetgreen kitchen during the noon-to-one rush. The bowl assembly is sequential, the bowls are customised, and the variance per bowl in the manual model is significant. Automate the assembly and the bottleneck moves elsewhere — to ingredient prep, to order capture, to the handoff window. You can serve more bowls per hour at the same staffing level, which means peak revenue is captured rather than walked away from.

Third — and this is where I want to slow down — a brand-new IK build avoids the retrofit penalty. Retrofitting an existing store with an IK is expensive, disruptive, and produces a P&L drag for the quarters during and after the conversion. A new build sites the IK from day one. The floor plan is right, the back-of-house flow is right, the order capture pattern is right. The Hingham margin is not “what Infinite Kitchen will do everywhere.” It is what Infinite Kitchen does when you build the store around it from a blank floor plan.

That is the case Sweetgreen wants to make, and it is why the 2025 build mix is being weighted so heavily toward new IK builds rather than retrofits. The unit economics on a new IK build are much cleaner than on a retrofit conversion, and the deck is being written around the cleaner number.

The risk of leaning on Hingham specifically is that it is one store, one month, in a Boston suburb that may or may not be representative of the dense urban cores where Sweetgreen’s traffic problem actually lives. The thirty percent first-month RLM is real. It is also a data point the bear case can dismantle by pointing out that Hingham’s site selection, demographics, and competitive landscape may have flattered the print. I think the bear case is too clever — a thirty percent RLM in month one is hard to explain away with site selection alone — but it is worth marking.

The seven points of labor and the point of COGS

The two operational deltas management has been most willing to quantify are seven-plus percentage points of labor savings and roughly one percentage point of COGS improvement, both measured against a comparable non-IK store.

Take labor first. Seven percentage points is a very large number. If a Sweetgreen store runs labor at twenty-eight percent of sales pre-IK and the IK box drops it to twenty-one percent, you have moved seven dollars of every hundred from labor to flow-through. On a $3M AUV store — a reasonable mid-range unit — that is $210,000 of annual RLM uplift from labor alone. Annualised across twenty IK builds in 2025, the delta is non-trivial even if you haircut the seven points materially.

What does the saved labor look like operationally? The makeline is the position most directly automated. In a non-IK store, the makeline runs with multiple team members assembling bowls in parallel during peak, plus expediters managing the handoff window and a runner replenishing bins. In an IK store, the bowl assembly is robotic, team members move to roles around the IK — ingredient replenishment, order capture, hospitality at the pickup window, light prep — and head count at peak compresses. The savings are not “fewer people in the store” one-to-one; they are “the same people producing more bowls per hour” plus “fewer hours to cover the peak shoulder.”

That nuance matters because the seven-plus point number is a structural claim about labor efficiency, not a head-count cut. The labor model is being rewritten, not just trimmed. The rewrite is what makes the saving durable rather than a one-time event that gets eroded by minimum-wage step-ups.

COGS at one percentage point is a smaller, less obvious win — and a more interesting one. The story management tells about COGS in IK stores is that the robotic portioning is more consistent than manual portioning. A makeline employee assembling 400 bowls in a peak hour will, despite training and bin-line discipline, over-portion proteins, premium add-ins, and dressings on a significant minority of bowls. The robotic line over-portions less. Over-portioning at scale is real money, particularly on proteins where the unit cost per ounce is high.

One percentage point of COGS, again on a $3M unit, is $30,000 a year. Stack it on the labor delta and you are clear of a quarter-million dollars of annual RLM uplift from the cost structure of the box, before any throughput gain at peak. Throughput gain at peak is the third leg of the stool — more bowls served per hour means more revenue captured during the busiest two hours of the day — but the seven points and the one point are the parts management is willing to put a number to in public.

The Restaurant Dive coverage of the rollout acceleration sized the operational deltas in similar terms, with the seven-plus-point labor figure and the one-point COGS figure surfacing in management commentary around the build plan. The numbers are not new claims invented for Q1. They are the same numbers the company has been telling for several quarters, now being deployed as the rationale for an accelerated rollout rather than a more cautious one.

The capex jump no one wants on the slide

Here is the part of the deck that does not photograph well. Capex is moving from $3.4M to $19.1M.

That is a roughly 5.6x increase in the capital expenditure run-rate. The increase is doing exactly what you would expect — it is paying for the twenty IK builds inside the 40-store 2025 plan, plus the equipment depreciation pattern that comes with shifting more new builds onto IK. An Infinite Kitchen build costs more than a traditional makeline build. The unit economics promise of IK is that the higher capex pays back through the labor and COGS deltas over the unit’s life, but the up-front cash demand is materially larger.

Five point six times. The number is uncomfortable for two reasons.

The first is cash. Sweetgreen has been carrying a meaningful cash balance and the IK rollout is being funded from existing resources. But the capex jump compresses the cash runway, particularly in a quarter where same-store sales are running negative and operating cash flow is under pressure. A chain with positive comps and expanding margin can absorb a capex step-up easily. A chain with negative comps and compressing margin is betting that the capex deployment hits the P&L fast enough to outrun the cash burn.

The second is the optics with the equity. The market reads capex jumps in soft quarters with skepticism. The narrative the chain has to sell is we are spending on the cost-structure fix because the cost-structure fix is what gets us through the wobble. The narrative the market sometimes hears is they are doubling down on a tech bet at the exact moment the core business is showing strain. Same numbers, different read. The interpretive question is whether the labor and COGS deltas are real enough, and the IK build cadence executable enough, to make the doubling-down read as discipline rather than panic.

I think it reads as discipline, because the alternative is worse. Cutting capex to defend the cash balance would mean fewer IK builds in 2025, which would mean less labor and COGS improvement reaching the consolidated P&L by 2026, which would mean the structural margin path looks weaker exactly when premium fast-casual is going to need a structural margin story. The capex jump is the price of staying on the curve. Not paying it does not get you the curve.

What it does mean is that 2025 is a transition year on the cash statement and 2026 is the year the IK-weighted unit base starts to show up meaningfully in the consolidated margin number. That timing is not flattering for anyone managing on quarterly comps. It is, however, the timing the operational thesis requires.

What twenty IK builds inside forty actually means

The number that I think is under-discussed is twenty out of forty. Half the new builds in 2025 are IK builds. That is not a pilot ratio.

A pilot ratio looks like one or two stores. An evaluation ratio looks like five. Twenty stores in a single year, paired with a stated path toward IK as the default new-build standard, is a deployment ratio. The company has crossed a line from “we are testing this” to “this is how we open stores now, with exceptions.”

Why does the ratio matter operationally? Three reasons.

The first is supplier muscle. Building twenty IKs in a year requires the equipment supply chain, install crews, and training pipeline to work at a fundamentally different cadence than building two or three. The chain is implicitly investing in standing capacity around the equipment partner, which has follow-on effects — better unit pricing, faster install times, deeper bench of trained openers.

The second is field-team adoption. Twenty IK openings means twenty general managers, twenty assistant manager teams, and several hundred new hires being trained on the IK model. The training pipeline becomes the bottleneck if not built ahead of demand. The Hingham first-month margin is partly a vote of confidence in the training infrastructure — you cannot open at thirty percent RLM with a poorly-trained crew.

The third is data. Twenty openings produces twenty times the operating data a one-store-per-year pilot would. That data informs the next generation of the IK design — where the bottleneck has moved in the box, which ingredients are causing portioning misses, which order-capture flows are throttling throughput at peak. The IK is an iteratively-improving platform, and the cadence of improvement scales with the deployment count.

What the ratio does not tell us is how the retrofit-versus-new-build mix evolves. A chain that opens fifty percent IK on new builds in 2025 is still left with the question of what to do with the several hundred legacy non-IK boxes already operating. Retrofitting is expensive and disruptive. Leaving them un-retrofitted means the IK margin profile only reaches the consolidated number as the mix shifts over years rather than quarters. The 2025 plan addresses the new-build cadence. It does not yet articulate a retrofit plan at scale. That is a gap I’ll be watching for on the next two earnings calls.

What I’d be testing if I ran the operator side

If I were on the operator side of a chain this size, with this kind of automation thesis, the questions I would be pushing my own deck on right now look like this.

What is Hingham’s RLM in month six? Month-one prints are a function of opening-team energy, novelty traffic, and the absence of long-tail operational decay. Thirty percent first-month is meaningful. The six-month margin is more meaningful. If Hingham holds at twenty-five-plus, the case is materially stronger. If it drifts toward the system average by month six, the case needs revisiting.

What does the IK new-build P&L look like in non-Boston-suburb demographics? Hingham’s site profile may not be representative. The honest test of the playbook is a build in a dense urban core, in a secondary-city suburban site, or in a Tier-2 MSA where Sweetgreen’s brand pull is less reliable. Twenty builds in 2025 will produce heterogeneity. I want to see the spread.

How do the labor savings hold up at maturity? Seven-plus percentage points at a new-opening IK is plausible. Seven-plus percentage points sustained at twenty-four months, after the local labor market has compressed and the team has had time to expand staffing back toward comfort, is the real number. The most common failure mode in automation deployments is that savings erode as field operators add headcount back to handle exceptions, training, and shoulder shifts. Holding the labor delta is its own operating skill, separate from the technology.

What is the throughput delta at peak, specifically? Labor and COGS are the two numbers management talks about. Throughput is the one I keep waiting for in a more disclosed form. If IK is producing fifteen percent more bowls per hour in the lunch rush, that is a revenue-side claim worth quantifying. The company has been more comfortable putting numbers on the cost side than the revenue side, which is reasonable — cost claims are easier to verify — but throughput is the lever that could turn this from a margin story into a same-store-sales story.

And finally, what is the retrofit playbook? At some point the chain has to either commit to retrofitting the legacy fleet, accept that the IK margin profile only shows up over a long mix-shift horizon, or — the option no one talks about — close legacy boxes faster than they can be retrofitted, accelerating the mix shift via attrition. I don’t have a view on which is right. I have a strong view the chain will need to articulate one within eighteen months.

Reading this alongside the rest of the operator stack

A note on context. Sweetgreen is not alone in pushing automation into the makeline. Chipotle has been running its own stack — Chippy, Autocado, the Augmented Makeline, and the kitchen display layer — through a series of public pilots, and I’ll be walking through that operator’s playbook in a forthcoming May piece on Chipotle’s AI stack (post 12). The comparison is instructive. Chipotle is automating from a position of comparative strength on comps; Sweetgreen is automating into a wobble. The mechanics of the deployment look similar from the outside. The capital-allocation posture is very different.

Marriott is doing something orthogonal on the lodging side — agentic AI deployed into commercial workflow rather than back-of-house production — which I’ll cover in an upcoming May piece on the Marriott deployment (post 13). Different industry, different problem set, similar through-line. Operators with scale are starting to commit capital to AI-and-automation deployments not as growth bets but as structural cost-and-throughput bets in mature P&Ls.

The reason Sweetgreen is the most-watched of these deployments is not that it is the largest. It is that the chain has chosen to do the work in the most public way of any operator in the segment. The press releases, the earnings-call detail, the willingness to put labor-savings and COGS percentages on a slide — all of it gives the rest of the operator world a real data set to read. The opacity in restaurant tech right now is bad for operators trying to build their own decks. Sweetgreen is a counter-example.

The mark

Mark me here. I think the Q1 wobble is a real demand signal, not a noise event, and I expect Q2 to print soft on comps as well. I also think the IK rollout is the right answer to that wobble, the capex jump is the price of the answer, and Hingham is a defensible — if not by itself sufficient — proof point.

The piece of the deck I want to see by year-end is a six-month RLM print from Hingham, a first-month print from a non-Hingham 2025 IK new build in a different demographic, and the beginning of a retrofit framework for the legacy fleet. If we get those three things, the 2025 calculus stops being a contrarian read on a soft quarter and becomes the consensus reading of a structurally-improving margin story. If we don’t, the bear case writes itself: capex up, comps down, automation thesis under-delivered.

I am long the deployment. I am cautious on the timing. The equity market will not properly price the labor and COGS deltas until the IK-weighted store count crosses thirty or thirty-five percent of the new-build cadence on a trailing basis, which is sometime in late 2026. Between now and then, the chain is doing the operationally correct thing in a quarter where the comp number tempts management toward the operationally easy thing. That is worth marking, even on a wobble.

— Priya covers operators for TableTransfers. Tips: [email protected].