Sweetgreen vs. Cava: The Unit-Economics Divergence

I spent the morning of August 13 with two earnings PDFs open side by side and the trading screen behind them, and the experience was almost embarrassing in its clarity. Cava and Sweetgreen — the two fast-casual stocks that the buy side persists in pairing because both sell bowls — had reported within thirty-six hours of each other. One printed comps positive, restaurant-level margins above 26%, and an AUV path toward $3M. The other printed a 7.6% comp decline, restaurant-level margin compressed by 360 basis points, and a stock chart that opened down nearly 25% on the print.

Here is the contrarian framing I want to defend in this piece: the market has not mispriced these two. It has priced Cava as a winning operator and Sweetgreen as a speculative automation bet — and the unit math, examined cleanly, agrees with that framing rather than refutes it. If you came to this column expecting me to argue Sweetgreen is the contrarian long because the multiple compressed, you are going to be disappointed. The multiple compressed because the operating data deserved it.

The two prints, side by side

Cava’s Q2 release put revenue at $278.2M, up 20.3% year over year, on a 2.1% same-restaurant sales gain. Restaurant-level profit margin landed at 26.3%, the highest the company has printed since the IPO window. AUVs continued their climb toward the $3M mark the management team has telegraphed as the steady-state for the format. Sixteen net new openings in the quarter. Adjusted EBITDA up. The release does the unusual thing of being almost boring to read — a sign, in this category, that the operator has the model under control.

Sweetgreen’s Q2 call transcript tells a different story. Revenue $185.6M, up only 0.5%. Same-store sales down 7.6%. Restaurant-level profit margin 18.9%, down from 22.5% in the year-ago quarter — a 360-basis-point compression that the company attributes to deleverage from the comp decline, wage pressure, and the front-loaded cost of the Infinite Kitchen retrofit program. The stock traded down 24.72% on the print, per the AInvest summary — a move that suggests the buy side had been underwriting a comp recovery story that this quarter explicitly disconfirmed.

The temptation, when two comparables print in opposite directions in the same week, is to call one a winner and one a loser and move on. The more useful exercise is to ask whether the unit economics — the only thing that ultimately matters in this category — actually justify the divergence. They do. Let me show the math.

The 26.3% versus 18.9% gap, decomposed

A 740-basis-point gap in restaurant-level margin is not a rounding error. On Cava’s roughly $2.7M trailing AUV, 26.3% restaurant margin equals about $710,000 of unit-level cash contribution before G&A. On Sweetgreen’s roughly $2.9M trailing AUV (which, note, is still slightly higher than Cava’s on the top line), 18.9% restaurant margin equals about $548,000. So Sweetgreen — despite a marginally higher AUV — generates roughly 23% less cash per box per year than Cava. Put that on a 250-unit base versus a 400-unit base and the absolute dollar gap is the difference between a company that can self-fund its build pipeline and one that has to keep apologizing for capex.

Now apply Cava’s incremental open rate. Sixteen units in the quarter, on a base scaling toward 400 by year-end, with an opening AUV the company has guided closer to $2.8M in year one. The cohort math compounds: each new Cava unit added to the system generates roughly $700K of restaurant-level cash on year one, against a build cost the company has indicated runs roughly $1.2M to $1.4M all-in. That is a 50%+ cash-on-cash payback in year one — the kind of unit economics that, in 2018, would have gotten a fast-casual founder onto the cover of a magazine. Cava is printing it quietly, in 2025, with the stock already up.

Sweetgreen’s incremental unit math is harder to defend at current restaurant-level margins. At 18.9% on a $2.9M AUV, a new traditional unit generates about $550K of contribution. On a build cost the company has indicated runs $1.0M to $1.4M for a conventional Sweetgreen, that is a year-one cash-on-cash of roughly 40-50% — still respectable, but only if the comp stops bleeding. With comps down 7.6% in the most recent quarter, the assumption that the next unit’s first-year AUV holds at the system average is itself a load-bearing forecast, not an observation.

Where Infinite Kitchen does and doesn’t help

The bull case for Sweetgreen is, and has been for two years, that Infinite Kitchen — the automated bowl-making line the company began deploying in 2023 and has since pushed into a 33-IK target for 2025 — fundamentally re-rates unit margins by collapsing labor as a percentage of sales. We sketched the operating theory in a forthcoming May piece on what IK actually changes inside the four walls. Management’s framing on the Q2 call held the line on that thesis: IK retrofits and greenfields are tracking to 7-point labor-margin improvements at maturity, and the 33-unit fleet milestone for 2025 is intact.

The problem is the capex line. Sweetgreen has indicated Infinite Kitchen retrofit cost runs $200,000 to $300,000 per unit on top of base build. Take the midpoint: $250K of incremental capex to capture roughly 7 points of restaurant-level margin on a $2.9M AUV. That is $203K of annual restaurant-level cash gained, against $250K of capex — an 81% one-year cash-on-cash on the automation increment if the labor math holds and the comp doesn’t keep declining. Defensible. Almost good.

But — and this is the unforgiving part — that automation increment is being deployed while the underlying same-store sales line is down 7.6%. The IK math assumes a stable revenue base on which to capture the labor saving. If comps continue to deteriorate, the absolute dollar saving from IK shrinks even as the capex is sunk. The bet, in other words, is that automation savings outrun comp deterioration. The Q2 print did not provide evidence that they are.

This is what I mean by “speculative automation bet.” It is not a slur. It is a description of the geometry: Sweetgreen’s equity story now requires the IK rollout to fix the unit economics faster than the demand environment can erode them, and the controlling variable — comp — is moving the wrong way.

The market’s pricing, examined

I want to be precise about what the 24.72% one-day drawdown in Sweetgreen’s stock actually said. It did not say “this company is going to zero.” It said the buy side had been valuing the equity on a forward path that assumed comp recovery in the back half of 2025, and Q2 disconfirmed that path. The repricing was a recalibration of the comp curve, not a verdict on the IK thesis. The two things are conflated in most of the trade-press coverage, including the otherwise useful CNBC recap of the Cava print, which pairs the two stocks without distinguishing the variables.

Cava, by contrast, is being priced as an operator that has hit its model. The stock didn’t pop on the print — it didn’t need to. The forward multiple is already reflecting the AUV ramp and the 26.3% restaurant-level margin as the operating reality, not the upside case. That is the difference between an operator the market trusts and an operator the market is waiting on. Cava is in the first column. Sweetgreen, as of August 13, is in the second.

If you want the analogous Chipotle frame — the comparison the buy side actually runs internally — see an upcoming May piece on what the Chipotle AI stack is supposed to do for labor and throughput. The short version: the operator that gets automation to amplify a working unit economic base wins; the operator that gets automation to rescue a deteriorating base is making a different, harder bet.

What I would underwrite, and what I wouldn’t

If I were sizing a fast-casual position today on August 22, with this Q2 data in hand and no other catalyst between now and Q3 earnings, the trade is obvious and unromantic. Long Cava on the unit-economic certainty; the multiple is full but the operating data justifies it and the AUV ramp is the kind of compounding variable that re-rates equity quietly over twelve to eighteen months. The downside risk on Cava is a comp slowdown the company has not yet flagged.

Sweetgreen is the harder call. I do not think it is a short here — the stock has already absorbed the comp news, the IK fleet milestones are intact, and the company has the balance sheet to finish the 2025 rollout without raising. But I would not be long either, because the controlling variable — comp — has to inflect before the IK margin story matters in the equity. The trade is to wait for one of two signals: (1) a single quarter of stable or positive comp, which would re-engage the IK thesis as additive rather than rescue, or (2) IK-fleet restaurant-level margin disclosures that show the automated cohort outperforming the traditional cohort by enough basis points to make the rollout self-funding regardless of comp.

I called the underlying framework — restaurant operators having four margins, not one, and automation operating on different margins differently — in a forthcoming May framework piece. The Sweetgreen-versus-Cava print is the cleanest live example of that framework I have seen this year. Cava is winning the operating margin. Sweetgreen is betting on winning the automation-adjusted labor margin. Those are not the same fight, and the market is correctly pricing them differently.

The bottom line

Two companies. Same week. Same category. One generates $710K of unit-level cash on a winning model. The other generates $550K on a model that needs automation to bail out a deteriorating comp. The 740-basis-point margin gap is real, the AUV gap is closing the wrong way for Sweetgreen, and the equity reaction tracked the unit math. The market did not get this one wrong. It got it efficient.

The interesting question for the next two quarters is whether Sweetgreen’s IK rollout produces a public, auditable cohort margin disclosure — not the “tracking to 7 points” framing, but actual comparable restaurant-level margin for the IK fleet versus the traditional fleet. If that disclosure shows up on the Q3 call and the gap is real, the equity story re-rates fast. If it doesn’t, the speculative automation bet stays speculative, and Cava keeps getting paid for being the operator.

— Oliver writes The Bottom Line for TableTransfers. Tips: [email protected].