Don't Pay the AI Premium: A Buy-Side Thesis on Restaurant M&A in 2026

I passed on a deal last quarter I’m now glad I lost. A small regional group, deck opening on a slide titled “AI-First Operating Stack,” asking a turn and a half above where I underwrite the category. The seller’s banker pitched it the way I keep hearing it pitched: the embedded AI tooling deserves a premium because it’s generating lift today and will compound. The buyer who won paid the full sticker. I think they were wrong, and I’m writing this so that if you’re a financial buyer in mid-market hospitality, you don’t make the same trade.

The thesis, plainly. Paying a headline AI premium for a restaurant group in May 2026 is a mispricing of optionality. The tooling has a feature half-life under fourteen months, the public hospitality-tech comps have already priced the platform-side optionality (so there is no public-market arbitrage at exit), the vendor lock-in tax inside these stacks is real and rising, and the strategic acquirers who can underwrite an AI premium — DoorDash, Toast itself, Booking via OpenTable, Thoma Bravo — are buying the platform layer, not your operator. You’re paying their valuation for their thesis on an asset that doesn’t accrue to you. Buy the conventional group at the conventional multiple. Do the work yourself.

That’s more aggressive than Marcus drew in his deal roundup last week. Marcus calls the AI premium “real but measured” and tolerates an 18–22% uplift when components are durable. I think he’s half right. The components he flags as durable — data, team — are durable. The component priced into the headline premium is the tooling, and that is a wasting asset.

Four arguments. Each grounded.

Argument 1: The feature half-life is roughly fourteen months

Toast launched its conversational AI assistant publicly as Toast IQ on October 29, 2025, expanding what early users had piloted under the “Sous Chef” name into a full operator interface (Toast newsroom, Oct 29 2025). Roughly six months later, on May 5, 2026, Toast shipped Toast IQ Grow, a bundled marketing-agent product at $499/month that absorbs a meaningful chunk of the “AI marketing” stack restaurants were buying piecemeal a year ago (BusinessWire, May 5 2026; Toast newsroom). Q1 2026 ARR was up 26% to $2.2B and Toast explicitly tied that growth to Toast IQ adoption (Toast Q1 2026 release).

On the reservations side: VOICEplug AI and OpenTable announced a global voice-reservations integration on November 24, 2025, expanding automated phone bookings to twenty countries simultaneously (PR Newswire, Nov 24 2025). Multilingual voice handling, real-time inventory sync, group bookings — exactly the capabilities an “AI-forward” operator pitched as proprietary last spring.

What does this cadence tell a buy-side underwriter? It tells you the specific tooling deployed in any operator you diligence today is, at best, the last cycle’s version. Toast’s release notes describe Toast IQ Grow as bundling features operators were previously buying from “point solutions” — platform-level commoditization on a six-month tempo. Whatever your seller built or selected in 2024–2025, the public platforms shipped a substitute four to fourteen months later.

Judgment, flagged: I underwrite hospitality-tech features the way I underwrite kitchen equipment with a four-year useful life and a two-year economic life. The book asset depreciates over four; the competitive moat depreciates faster. Pay for it accordingly.

Argument 2: The public comps already price the optionality

If the AI-stack thesis were truly undervalued in private markets, you’d expect a yawning gap between hospitality-tech and operator-side public multiples. There isn’t one. The gap goes the other way.

Hospitality-tech comps, current (May 2026):

• Toast (TOST): market cap ~$16.86B, EV ~$14.89B, TTM revenue $6.15B → ~2.4x TTM EV/Sales (stockanalysis.com; Toast Q1 2026 release). EV/EBITDA TTM ~171x reflects a still-thin EBITDA base (Yahoo Finance TOST).
• Olo (OLO): taken private by Thoma Bravo, closed September 11, 2025 at $10.25/share, ~$2.0B equity value — a 65% premium to the unaffected $6.20 close (Thoma Bravo release; Goodwin Procter). A strategic sponsor pulled the asset off the public-comp set. Thoma Bravo’s edge is consolidating restaurant SaaS, not running restaurants.
• Lightspeed (LSPD): EV ~$849.6M against Q4 FY2026 guided revenue $280–284M → ~3.0x annualized EV/Sales (Yahoo Finance LSPD.TO; Lightspeed Q3 FY2026 transcript).
• DoorDash (DASH): market cap $72.4B, P/S ~5.3x on Q1 2026 revenue $4.0B (+33% YoY) (Yahoo Finance DASH; DoorDash Q1 2026 release; CNBC, May 6 2026).

Restaurant-operator comps:

• CAVA (CAVA): TTM EV/EBITDA ~60–64x; ~6.9x NTM EV/revenue (Yahoo Finance CAVA; TIKR Q1 2026 analysis).
• Chipotle (CMG): TTM EV/EBITDA ~20.4x, EV ~$47.0B, TTM EBITDA ~$2.31B (Yahoo Finance CMG; valueinvesting.io CMG).
• Sweetgreen (SG): EV/Revenue ~1.3–1.5x on negative TTM EBITDA (gurufocus SG; Multiples.vc SG).
• Wingstop (WING): TTM EV/EBITDA ~21.5x to ~37x depending on EBITDA normalization; EV ~$4.67B (valueinvesting.io WING; Yahoo Finance WING).
• Texas Roadhouse (TXRH): TTM EV/EBITDA ~15.2x (gurufocus TXRH; Yahoo Finance TXRH).

Read the public-market signal. Toast — the AI-tooling platform that is the premium being paid downstream — trades around 2.4x sales. Olo got rescued from the public market because public investors would not pay a software-comp multiple for hospitality SaaS. The best operator comp, CAVA, trades at ~6.9x NTM revenue because public markets pay for unit economics, brand, and growth runway, not “we deployed Toast IQ Grow.”

If you, as a private acquirer, pay a 20% premium for an operating group on the basis of its AI stack, you are paying for an asset the public market values at 2.4x sales inside Toast itself. You cannot mark that premium up at exit. The arbitrage doesn’t exist. The optionality is priced into the platform vendor’s stock, not into the operating group’s purchase price.

Judgment, flagged: the AI-platform layer has compressed onto the operator layer, not vaulted past it. CAVA and Wingstop multiples > Toast multiple. The operator multiple does the work, not the tooling multiple.

Argument 3: The lock-in tax exceeds the productivity gain

LBO models on software-enabled operators routinely skip one line: the switching cost of the embedded vendor stack as a hidden contingent liability. That liability has grown.

The public review record on Toast describes multi-year auto-renewing contracts (typically 2–3 years), proprietary hardware that’s useless if you leave, mandatory payment processing, and early termination fees calculated as the remainder of the subscription term (Capterra Toast POS reviews; summarized in POS USA hands-on review, 2026 and startupowl Toast review, 2026). G2 and Capterra averages sit at 4.1–4.2 across roughly 1,600 combined reviews — this isn’t a hate-read, it’s the median experience. SevenRooms — now a DoorDash subsidiary — has its own thread of reviews flagging auto-renewal “traps” and slow cancellation (SevenRooms Trustpilot, UK).

Two consequences for an acquirer.

First, you are inheriting the unexpired portion of every one of those contracts. If the seller signed a three-year Toast deal in 2024 with a steep ETF, you cannot reasonably switch off Toast until 2027 without writing a six-figure check for a multi-unit group. That removes the re-platforming optionality that justifies buying a conventional group instead.

Second, the productivity gain in the data room is partially captured by the vendor through payment-processing spreads, marketing-as-a-service take rates, and tiered feature gating. DoorDash’s Q1 2026 marketplace GOV was $31.6B against $4.0B revenue (DoorDash Q1 2026 release; DoorDash 10-Q). That blended take is the operator’s cost line. The “AI lift” in the deck is partially a transfer payment from operator P&L to platform revenue.

Judgment: I haircut any seller-reported “AI uplift” by 30–50% for take-rate capture and lock-in over the hold. For platform-side aggregation see Marcus on the DoorDash Commerce Platform and Eitan’s four-margins framework.

Argument 4: The acquirers who can pay are not you

When DoorDash bought SevenRooms for $1.2B in May 2025 (CNBC, May 6 2025; DoorDash IR) and Deliveroo for ~$3.9B the same day (DoorDash IR; CNBC, May 6 2025), the deal logic was coherent. DoorDash was buying a CRM/guest-data layer and a geographic expansion that — plugged into its existing marketplace and commerce platform — would extract additional take rate across a wider merchant footprint. Tony Xu cited a 40-country, 1B-population reach. That math works for DoorDash. It does not work for you.

Same logic on Thoma Bravo’s Olo take-private. Thoma Bravo runs one of the largest restaurant-software portfolios in private equity; Olo plugs into their consolidation thesis. The 65% premium (Goodwin Procter) is justified because they own the consolidation engine. They will cross-sell into and out of Olo. You will not.

The pattern: every recent transaction where an “AI capability” premium was credibly paid was paid by a strategic acquirer whose existing platform converted that capability into incremental take rate or distribution. Operator-side buyers — family offices, regional roll-ups, sponsors without a platform — do not have that conversion engine. When you pay the same premium, you pay for the capability without owning the lever that turns it into cash flow.

You are not DoorDash. You are not Thoma Bravo. You should not be paying their multiples. See also: the Resy/Amex platform-fit piece and the Toast Desk Review — strategic premium accrues to the platform side.

What I’d actually pay: three heuristics

Not a model. Three rules I currently use for indie groups in the 5–25 unit range.

One: anchor the operator multiple to operator comps, not tech comps. TXRH at ~15x EBITDA is the scaled-mature anchor. CMG at ~20x is the growth-with-brand anchor. Indie groups print at a discount to TXRH on quality and to CMG on growth — call it 6–9x trailing EBITDA depending on unit economics, brand, and lease portfolio. Do not slide toward CAVA’s ~60x EBITDA or Toast’s ~2.4x sales because the deck has “AI” in the title.

Two: pay separately for data and team, never for tools. If the seller has 18+ months of clean, integrated POS-plus-reservations data and an operations team that built the tooling and will stay through close-plus-24-months, that’s a durable asset — price it as an additive 3–8% on the operator multiple. The tools themselves get zero premium. Toast just bundled them.

Three: underwrite a 24-month re-platforming reserve. Assume you rip and replace at least one component during the hold. Reserve roughly $5–15K per unit (judgment, not data) against migration, contract buyouts, and retraining. If existing contracts have steep ETFs, double the reserve. If the deal cannot clear with that reserve built in, the deal cannot clear.

I’ll run the framework against a worked example next week — same data, no fabricated specifics.

Where I might be wrong

Three places. I owe the steelman.

One: team-and-data is worth more than I’m pricing. If the operations team has shipped two or three AI integrations end-to-end and credibly knows how to deploy the next three, that meta-capability compounds and is scarce. I’m at 3–8% premium. A reasonable underwriter could price it at 10–12%. I might be light.

Two: the platform-take-rate trend reverses. If antitrust pressure or merchant pushback forces DoorDash, Toast, and the OpenTable/Booking stack to compress take rates over the next 24 months, much of Argument 3’s lock-in tax evaporates. I don’t see the evidence yet, but I’d update fast.

Three: an indie finds a wedge the platforms can’t replicate. If a small group develops proprietary operating data — voice-of-customer at a specific cuisine, labor patterns at a specific format — that the platform vendors cannot ingest from their own customer base, that’s a real asset deserving a real premium. I haven’t seen one yet at the scale I underwrite. I’d love to.

Marcus and I will keep arguing the middle ground in print. The Bottom Line is meant to be a place where the buy-side and the sell-side disagree out loud. If you’re a seller and you think I’m undervaluing your group, write me. Counter-arguments especially welcome.

— Oliver writes the buy-side perspective for The Bottom Line. He sits on three operator boards. Tips, especially counter-arguments: tips@tabletransfers.com.