Wendy's Doubles Down on FreshAI at Investor Day

I am watching the Wendy’s Investor Day stream from a back office in a Columbus-area franchisee group’s HQ, a half-eaten Baconator on a paper tray next to my notebook. The operator hosting me keeps the volume low so the line cooks don’t hear the capex slide. When the number lands — $100 to $110 million for 2025, more than double the prior year — he taps the screen with the back of a pen and says one word: “Google.” Then he goes back to expediting.

That is the trade, in miniature. Yesterday the Wall Street Journal put McDonald’s edge-AI build on the front page of its tech section — Nvidia silicon at the store, IBM as integrator, hundreds of US restaurants in flight (see Tuesday’s note). Twenty-four hours later, the second-largest US burger chain by domestic sales walks into its Investor Day and answers with a different stack entirely: FreshAI, built with Google Cloud, a 100-restaurant pilot scaling to 500 to 600 stores by year-end, and a CEO using the phrase “lot of confidence” on an open line to the sell side.

For the first time, the QSR AI map has two public-market reference designs sitting next to each other. That is not a small thing.

The number behind the number

Capex of $100 to $110 million is the headline, but the interesting line is what it is not. It is not a remodel program. Wendy’s already runs its Global Next-Gen design separately. The 2025 capex step-up is concentrated in technology — drive-thru hardware, digital menu boards, the back-end plumbing that lets a Google-hosted model actually take an order in 4.2 seconds and route it cleanly to the kitchen display system without a human re-key.

The franchisee next to me does the math out loud. Wendy’s has roughly 6,000 US restaurants. A 500-to-600-store FreshAI footprint by December is not a pilot anymore; it is ten percent of the system, deployed inside a single fiscal year, on a stack the company does not own end-to-end. That is the part the analysts on the call did not press hard enough. When you are leasing the model, the scaling question is not “can we build it.” It is “what does our cloud bill look like at 6,000 stores at peak Friday-night drive-thru volume.” Nobody on the call asked. They will, eventually.

Matt Spessard, the SVP and global CTO, dropped the one operator-grade number that mattered: the Columbus, Ohio test restaurant ran 22 seconds faster than the Columbus market average on drive-thru service time. Twenty-two seconds, in a business where the industry benchmark moves a tenth of a second per year, is the kind of figure that ends franchisee debates. It is also the kind of figure that gets re-tested the minute the rollout hits a market with a different accent, a different menu mix, and a 4G backhaul instead of fiber. I would not extrapolate the Columbus number to 600 stores. Wendy’s, to its credit, did not try to on stage.

Nvidia vs Google, in two parking lots

Here is the contrarian read, and it is the read I think the buy-side is going to be slow to catch.

McDonald’s and Wendy’s have now publicly split on the two questions that actually define drive-thru AI: where the inference runs, and who owns the model. McDonald’s is putting Nvidia hardware inside the restaurant — edge inference, IBM glue, latency budget measured in tens of milliseconds because the silicon is twenty feet from the speaker post. Wendy’s is leaning into Google Cloud — model in the cloud, the restaurant is a thin client with good microphones, latency budget measured by what the WAN gives you on a Tuesday night.

Neither is wrong. They are bets on different five-year cost curves. Edge gets cheaper if Nvidia keeps shipping and the chains get good at managing fleets of small boxes. Cloud gets cheaper if Google keeps undercutting on inference pricing and the models keep shrinking. The chain that picks correctly compounds an order-time advantage; the chain that picks wrong eats a stranded-asset write-down in 2028.

What is new, as of this afternoon, is that the bet is now legible to public-market investors. You can short one and long the other. You can ask the CFOs about it on the next earnings call. You can — and franchisees will — start pricing the technology line item differently in your unit-economics model. The Yum stack question I flagged in January just got sharper too: KFC, Taco Bell, Pizza Hut are still on a private AI roadmap, and they now have two public comps to be benchmarked against, not one.

What I am watching from here

Three things. First, the comp on labor hours per transaction in the FreshAI markets versus the non-FreshAI markets — Wendy’s will not break this out voluntarily, but the franchisee P&Ls will leak it inside ninety days. Second, the cloud-cost disclosure: if Wendy’s starts reporting technology-of-revenues as a separate line, that is the tell that Google’s bill is becoming material. Third, GTC in Pittsburgh on the 18th — Nvidia’s keynote will name QSR customers, and the absence of Wendy’s from that list, more than the presence of McDonald’s, is the data point.

The Baconator is cold. The operator is back to expediting. The Investor Day slide is still up on the screen behind him, $100 to $110M glowing softly. The bet is placed.

— Luca covers restaurant operators. Tips: tips@tabletransfers.com.