Sweetgreen's Infinite Kitchen, in Public View: A Case Study

I spent a Saturday morning in early May reading Sweetgreen’s Q1 2026 earnings transcript, the Q4 2025 transcript before it, three years of Infinite Kitchen press releases, and the Spyce-to-Wonder sale documents. There is no other AI deployment in fast-casual that has been narrated this publicly. The numbers — labor savings, location counts, capital structure, the M&A around the technology itself — are all in the public record. The question isn’t what did Sweetgreen do, because Sweetgreen has told us. The question is what does the playbook actually look like when you read it sideways, three years in.

This is the Operator column’s first Case Study format. A note on what that means.

Methodology

This Case Study is based on Sweetgreen’s public investor materials, recent earnings calls, press coverage, and operator-side commentary in public forums — not a first-person interview with the leadership team. Our usual Operator format runs a named operator walking through their playbook; we haven’t done that here. The case study is honest about what it is: a synthesis of what’s publicly known about a publicly-documented AI deployment. Where I quote, I quote from transcripts or releases. Where I infer, I say so. Where I don’t know, I name the gap.

The six core sources are: Sweetgreen’s Q1 2026 earnings transcript and Q4 2025 earnings transcript via The Motley Fool; the November 2025 Spyce-to-Wonder sale announcement via Restaurant Business; the Costa Mesa Sweetlane opening release via Restaurant Dive; the QSR Magazine 2025 step-up coverage; and the Restaurant Business and TechCrunch coverage of the original Spyce acquisition and first Infinite Kitchen opening. Where the post-sale strategy gets specific, I draw on Food on Demand’s coverage of Wonder’s plans for the technology. Everything in this piece traces back to one of those.

Company context: a publicly-traded chain under genuine top-line pressure

Sweetgreen (NYSE: SG) ended Q1 2026 with 285 restaurants, of which 33 are powered by the Infinite Kitchen, per CFO Jamie McConnell on the Q1 call. The chain is a publicly-traded fast-casual operator that has been, by most measures, the loudest restaurant-tech bull in its peer set since IPO.

It’s also a chain under genuine top-line pressure. Q4 2025 same-store sales fell 11.5%, with traffic and mix down 13.3%, partly offset by 1.8% of pricing. Restaurant-level margin dropped nearly 700 basis points year-over-year to 10.4%. Net loss in the quarter was $49.7M. Total Q4 revenue was $155.2M, down 3.5% versus the prior year — per the company’s own disclosures summarised by IndexBox. The company guided fiscal 2026 to negative 2-4% same-store sales, with restaurant-level margin of 14.2-14.7% and adjusted EBITDA of $1M-$6M.

That’s the context for everything that follows. The Infinite Kitchen story is not a story about a fast-growing chain layering automation on top of expanding comps. It is a story about a chain whose comps are negative, whose margins compressed 700bps in a single year, and whose automation thesis is now load-bearing for the equity story.

I’m not making a stock call. I am pointing out that the operator reading this case study is reading it from a particular vantage: a chain that bet enormous capital on automation because the unit-economic envelope was tightening, not in spite of it.

The deployment: four years, $70M, 33 stores, then a divestiture

Here is the timeline of what Sweetgreen actually did, drawn from press coverage and SEC-disclosable events.

August 2021: Sweetgreen acquires Spyce. TechCrunch reported the deal at the time. Spyce was an MIT mechanical-engineering spinout founded in 2015 by Michael Farid, Kale Rogers, Brady Knight, and Luke Schlueter (the four co-founders, named in Restaurant Business’s coverage of the 2025 sale). Spyce had run two automated restaurants in Boston pre-acquisition. Sweetgreen shut Spyce’s own restaurants and absorbed the team and technology to build what would become the Infinite Kitchen. The reported purchase price was approximately $70M.

May 10, 2023: First Infinite Kitchen opens in Naperville, Illinois. Restaurant Business covered the opening. The system is a conveyor-belt makeline: bowls move along a track, ingredients dispense from tubes overhead, and a human “finisher” adds the last items (herbs, avocado, dressing). A new “host” role was created to guide customers through ordering. CEO Jonathan Neman framed it at the time as quality- and convenience-driven: “We believe that automation will enable us to elevate the quality and integrity of our food, while also providing a faster and more convenient experience for our customers.”

2024-2025: Scaled rollout. Sweetgreen targeted half of 2025’s development pipeline (about 40 projected new units) to include the Infinite Kitchen, per QSR Magazine, with at least 25 new Infinite Kitchens scheduled to come online in 2025. The chain also began retrofitting select existing locations rather than building from scratch.

November 13, 2025: First “Sweetlane” drive-thru with Infinite Kitchen opens in Costa Mesa. Per Restaurant Dive, the Costa Mesa unit combined the chain’s two flagship tech bets — the Sweetlane digital-pickup drive-thru and the Infinite Kitchen makeline — in a single prototype. Orders are placed ahead via the app or website and handed off through the drive-thru.

November 2025: Sweetgreen sells Spyce to Wonder for $186.4M. Restaurant Business and The Robot Report reported the structure: $100M cash + $86.4M in Wonder Series C preferred stock. The four Spyce co-founders and 38 engineers/support staff moved to Wonder. Sweetgreen retained a supply and license agreement to keep deploying Infinite Kitchens; per Restaurant Business, the equipment goes to Sweetgreen at cost plus approximately 5%, with installation and service charges held roughly flat versus before.

Q1 2026: 33 Infinite Kitchen units, ~half of 2026 openings will be Infinite Kitchens. McConnell on the Q1 call: about 13 net new restaurants expected in 2026, with nearly half featuring Infinite Kitchen technology. (Q4 2025 commentary had pointed to 15 net new openings; the Q1 update revised that downward.)

So: roughly four years from acquisition to divestiture. $70M in, $186.4M out — a paper gain of $116M on the technology asset, while retaining operational access. 33 deployed units. The arc is unusual. Most operators who buy a technology company don’t sell it four years later, retain the rights, and also lock in their unit-cost structure at cost-plus-5%.

Results in real numbers

What the public record actually says, in numbers I can attribute:

• Labor savings of “more than 700 basis points” at established Infinite Kitchen locations versus classic locations of similar age, per Neman on the Q4 2025 call and re-stated on the Q1 2026 call. This is the headline number. It’s been repeated quarter after quarter, which is itself a tell — companies don’t repeat a number this consistently unless it’s holding up.
• “Nearly 100 basis points of COGS improvement” at Infinite Kitchen locations versus comparable classic units, per Neman, also re-stated in the Costa Mesa Sweetlane release.
• Higher AUVs at Infinite Kitchen locations versus traditional units, per Neman on the Q1 2026 call. The exact dollar AUV uplift was not disclosed.
• Drive-thru comp growth >20% at the Schaumburg, IL Sweetlane location (a Sweetlane unit without the Infinite Kitchen) in Q1 2025, cited as one rationale for combining the two formats. Reported via Restaurant Dive.
• Group-level labor cost 31.4% of revenue in Q1 2026, up 250bps year-over-year, with management projecting Q2 in the “low 29%” range, per McConnell. This is the chain number, not the Infinite Kitchen number. The IK savings are real, but they’re not enough to offset the comp deleverage at the consolidated level.
• Throughput: hundreds of bowls per hour, far above a typical hand-built makeline, per multiple Sweetgreen materials. The exact bowl-per-hour figure varies by source and I’d treat it as directional.

What’s not in the public record, and that I’m not going to invent:

• Per-unit capital cost for an Infinite Kitchen install. Sweetgreen has never disclosed it.
• Retrofit cost vs. new-build cost.
• ROI payback period in dollars.
• AUV in dollars for Infinite Kitchen vs. classic units.
• Whether the “700bps labor savings” is comparing same-day-part labor or all-day labor.

Those gaps matter. The whole investor pitch on automation hinges on payback period, and Sweetgreen has chosen to disclose the operating-margin delta but not the capex denominator. An operator reading this should not assume the math works at their unit volumes; the public number could be reasonable at $4M AUVs and unreasonable at $2M AUVs, and we cannot tell which from outside.

What worked

Three things, by my read.

The automation delivered the labor-line outcome it was supposed to deliver, and the delivery was repeatable. The 700bps labor savings number has held across multiple quarters of reporting. That is more impressive than the number itself. Most automation pilots in foodservice produce a great number in quarter one and a worse number in quarter four when you account for downtime, maintenance, and the part-time human roles that creep back in to fill gaps. Sweetgreen’s number hasn’t decayed in public disclosure. Either it’s holding up genuinely or Sweetgreen has gotten very good at framing it. Both are interesting.

The format work — Sweetlane plus Infinite Kitchen — looks like real operating-model innovation. Most QSR automation stories are about replacing labor in existing formats. The Costa Mesa Sweetlane is something different: a digital-only ordering channel that hands off to a digital-only production channel, with humans relegated to finishing and handoff. This is closer to what McDonald’s drive-thru AI experiments have been gesturing at, but Sweetgreen has actually shipped a unit that combines them. Whether the comp-growth thesis holds is a 2027 question. The architecture decision was 2025.

The Spyce-to-Wonder transaction was, financially, an extremely good outcome for an operator. Read it cleanly: Sweetgreen paid $70M for a technology asset, used it operationally for four years, sold it for $186.4M (with $100M in cash hitting the balance sheet at exactly the moment the chain needed liquidity given comp pressure), and negotiated a cost-plus-5% supply agreement for future Infinite Kitchen builds. That last clause is the one I’d underline. Sweetgreen no longer carries the R&D burden, no longer has to support a 38-person engineering team, and gets the equipment at near-cost while a buyer with deeper pockets — Wonder, backed by Marc Lore — funds the next generation. That is a very operator-friendly deal structure, and most operator-acquires-tech-company deals do not end this well.

What didn’t work — or hasn’t yet

Two things, both important.

The automation did not save the comp. Same-store sales declined 11.5% in Q4 2025 and are guided to decline 2-4% in 2026. The chain’s restaurant-level margin compressed nearly 700bps in a single year — the same number as the Infinite Kitchen labor savings, going in the opposite direction at the consolidated level. The 700bps of unit-level labor savings at Infinite Kitchen units are real, but they are not enough to offset the deleverage from negative comps at classic units, which still represent 252 of 285 stores. Automation is a margin-protection lever. It is not, by itself, a demand lever.

The capital intensity story is unresolved. Sweetgreen has never disclosed a per-unit Infinite Kitchen capex figure. Press reporting on the original 2022 commentary suggested Sweetgreen believed the system could reduce labor “by approximately fifty percent” — a much bigger number than 700bps of total labor cost. The gap between “50% labor reduction” and “700bps labor savings” is the gap between the pitch and the operating reality. Both can be true with careful definition. A skeptical reader notes that the labor number disclosed is the smaller of the two framings.

The 2026 plan walks back the development pipeline too. Q4 2025 commentary pointed to 15 net new openings; the Q1 2026 update was 13. The 18 gross openings haven’t changed materially, but lease-related closures are eating more of the net number. The Infinite Kitchen rollout is still scaled — about half of openings — but the denominator is shrinking. This is not a chain accelerating into automation. It is a chain holding the line on automation while everything else gets harder.

What’s still open

A few questions I’d be putting to the IR team if I were on the next call.

What is the per-unit capex on an Infinite Kitchen, retrofit vs. new build? Without it, the ROI math is unaudited. The cost-plus-5% Wonder agreement implies Sweetgreen knows the number precisely; they’ve just chosen not to disclose it.

What is the AUV in dollars at Infinite Kitchen units versus classic? “Higher AUVs” is qualitative. Investors are increasingly willing to accept it on faith; operators reading this should not.

What share of labor savings comes from headcount reduction vs. hours reduction vs. labor-mix shift to lower-wage roles? The system kept humans on the floor (the host, the finisher). The composition of the staffing model — and where in the wage band the cuts landed — is the part of the playbook other operators most need.

Does the supply-and-license agreement with Wonder include exclusivity? Wonder has stated publicly it plans to open its own Infinite Kitchen location in Manhattan in 2026 and to install the technology in “half of our new kitchens starting in 2027.” Whether that extends to direct competitors of Sweetgreen — other salad/bowl concepts — is the question I’d most want answered. The deal structure suggests Sweetgreen does not have category exclusivity, but the licensing fine print is private.

Operator takeaways

If you are a multi-unit operator reading this and thinking about what Sweetgreen’s playbook means for you, here is the honest synthesis.

You probably can’t replicate the deal structure. The Spyce acquisition was a $70M check that only a venture-backed, soon-to-IPO chain could write in 2021. The Wonder sale was an $186.4M exit that required a credible technology buyer with $100M of cash. Most multi-unit operators do not have either side of that transaction available. The outcome — owning operational rights at cost-plus-5% — is unusual and not transferable.

The labor math is plausible but jurisdictionally sensitive. 700bps of labor savings at units running urban-coastal wage stacks is a different number from 700bps in low-wage geographies. Sweetgreen’s footprint skews coastal and metro. An operator in lower-wage markets should expect the absolute-dollar payback to be smaller, possibly meaningfully smaller. Eitan’s four-margins framework is the cleanest lens to use here: the labor savings are real on the contribution line, but the capex shows up on the operating line, and the bridge between the two is where most automation business cases die.

The format innovation is more interesting than the automation itself. What Sweetgreen did in Costa Mesa — combining a digital-only ordering channel with a digital-only production channel and a near-zero in-store ordering surface — is a more replicable lesson than “buy a robot.” Most QSR operators already have the digital ordering channel. Most also have an existing makeline. The question to take from Sweetgreen’s playbook is whether your format can be redesigned around the digital order, not whether you can afford the robot. Chipotle’s automation experiments sit at a similar inflection but have not yet committed to a format-level redesign of the kind Sweetgreen has shipped.

You should treat the per-unit ROI as unaudited until disclosed. I cannot stress this enough. The 700bps labor savings number is the only metric Sweetgreen has put on the record with consistency. Everything else — capex, payback, AUV uplift in dollars — is asserted but not numerated. A peer operator considering automation should demand from their own internal team the same numbers Sweetgreen hasn’t disclosed, and should be suspicious of any pitch that handwaves them.

The voice-agent and ordering side is a separate question. Sweetgreen’s automation is on the production side. Sweetlane is a digital ordering format, not an AI ordering agent. The drive-thru question — whether AI takes voice orders — is genuinely separate, and the Voice Agent Curve framework we ran in March applies to that separately. Don’t conflate the two when reading Sweetgreen’s playbook.

The valuation premium is real but uneven. The Bottom Line’s coverage of AI premia in restaurant valuations and QSR M&A multiples is the relevant context: Sweetgreen’s equity story now rests heavily on the automation thesis, and the Wonder transaction monetised the technology layer at a clean $116M gain on cost. Whether operating with the technology generates a corresponding premium to operating cash flow remains the open question. Comp performance suggests not yet.

Close

The Infinite Kitchen is the most transparent AI deployment in fast-casual today. That doesn’t make it the most successful. It makes it the most readable. Sweetgreen has, quarter after quarter, given the trade press and the analyst community the metrics they asked for — 700bps labor savings, 100bps COGS improvement, higher AUVs, share of new openings — and has done so even as the consolidated comp story has deteriorated.

The fact that the chain divested the underlying technology asset four years in, kept the operating license at cost-plus-5%, and reported the same 700bps labor figure in the quarter after the sale tells you most of what you need to know about how Sweetgreen now thinks about automation. It’s an operating capability, not a moat. The moat — if there is one — was always going to be in format, brand, and unit-volume durability. The Infinite Kitchen is a margin lever. It is not the company.

For an operator at the desk-review stage on automation vendors, Sweetgreen’s case is the cleanest public reference point you have. Use the numbers it has disclosed. Discount the numbers it hasn’t. And remember that the chain that built and then sold the technology is, by its own guidance, a chain whose same-store sales will decline in 2026.

That is the case study. The next quarter’s earnings call will be the next data point. I’ll be reading it.

— Priya covers the Operator beat. Tips: [email protected].