McDonald's Spent $6 Billion Going Digital. COVID Saved Them by Accident.

Four bets. One company. Twenty-five years.

In 1998, McDonald’s spent roughly $181 million rebuilding every kitchen in America. Service times got worse. The CEO was out within two years.

In 2014, McDonald’s spent more than $6 billion on store remodels, self-service kiosks, and a mobile ordering app. The franchisees revolted and formed the first independent advocacy group in the chain’s history.

In 2019, McDonald’s spent roughly $300 million on an Israeli AI company called Dynamic Yield — the chain’s largest acquisition in two decades. Inside three years, they sold it to Mastercard.

In 2021, McDonald’s partnered with IBM to put automated voice ordering in the drive-thru. Customers filmed the AI adding 260 chicken nuggets to single orders and refusing to remove bacon. By June 2024, the contract was cancelled.

Four cycles. The same five-step shape every time: a vision, a massive spend, a long quiet underperformance phase, a fork in the road, and a “we will revisit this” press release.

McDonald’s survived all four. Right now, every Fortune 500 board in America is signing off on a fifth — agentic AI — using the same justification McDonald’s used in 1998, 2014, 2019, and 2021. The interesting question is not whether the technology works. McDonald’s bets mostly worked. The question is whether the company placing the bet has the runway to live through what comes after the press release.

This is the architect’s view of a 25-year tech saga, and why it is the cleanest analog we have for the AI capex cycle the rest of the economy is currently inside.

Act 1 — “Made for You” (1998)

McDonald’s in the late 1990s had a problem. Customers wanted hotter, fresher food. The chain’s batch-cook-and-hold-under-heat-lamps model produced consistency at the cost of quality. The fix was an operational rebuild called Made for You: a build-to-order kitchen system that toasted bread on demand, cooked meat just in time, and assembled the burger only after the customer ordered it.

The promise was real. Fresher food. Less waste. Better margins on a longer time horizon.

The execution was a disaster.

McDonald’s spent approximately $181 million of corporate capital outfitting around 12,500 U.S. restaurants. Franchisees were charged an additional $25,000 to $40,000 per store for their share of the hardware. The new system was slower than the heat-lamp model it replaced — exactly when Wendy’s and Burger King were optimizing their drive-thrus for raw speed. Same-store sales decelerated. Franchisee morale collapsed for the first time in the chain’s history. McDonald’s stock halved between 1999 and 2002.

CEO Jack Greenberg departed in late 2002. Made for You was one of several stated reasons — alongside the Boston Market acquisition and broader strategic disagreements — but it set the template for every cycle that followed.

The kitchen system itself was never explicitly killed. It was absorbed, modified, renamed in subsequent investor presentations, and quietly continued. This is the first thing to notice about McDonald’s tech bets: they rarely die. They get absorbed into the next vision.

The pattern, stated for the first time: large infrastructure spend → quiet underperformance phase → leadership turnover → strategy quietly continues under a new name.

Act 2 — Experience of the Future (2014)

By the mid-2010s, the chain’s stores looked dated next to Chipotle, Shake Shack, and Panera. McDonald’s launched its most ambitious physical-and-digital overhaul ever, branded Experience of the Future (later “Velocity Growth Plan,” later just absorbed). The scope was enormous: a unified store design language, self-service kiosks in every restaurant, a mobile order-and-pay app, table service for digital orders, and a complete drive-thru lane rebuild.

The reported total spend across the U.S. system over five-plus years was north of $6 billion, combining corporate capital, ongoing tech, and franchisee co-investment. Per-store remodel costs ranged from $160,000 to $700,000 depending on tier — and franchisees were on the hook for a significant share, on top of royalty and rent obligations.

Two things broke.

Franchisees revolted. The remodel mandate landed in the middle of a same-store sales slump. Operators were being asked to pay six figures per location for a remodel they were not convinced would pay back, on a corporate-mandated timeline they could not negotiate. In 2018, U.S. franchisees formed the National Owners Association — the first independent franchisee advocacy group in McDonald’s 60-year history. It exists today.

The app underperformed. Through 2019, mobile order adoption was a fraction of what corporate had modeled. Kiosks worked in some demographics, were ignored in others. The drive-thru remodels were popular but expensive. The infrastructure was being built faster than anyone could prove it was worth building.

CEO Steve Easterbrook, who had taken the role in March 2015, accelerated the program. He left McDonald’s in November 2019 under unrelated personal-conduct circumstances. Chris Kempczinski inherited a half-finished, half-paid-for, half-working digital infrastructure.

Same pattern. Same shape. Bigger spend.

What changed everything was a thing nobody at McDonald’s planned for.

Act 3 — Dynamic Yield (March 2019)

The most literal AI parallel in the saga sits here, four months before the rest of the world started using the term “generative AI” in earnest.

In March 2019, McDonald’s acquired Dynamic Yield, an Israeli AI-personalization startup, for a reported ~$300 million. The acquisition was, by McDonald’s own press release, the largest in 20 years — bigger than any deal since the 1999 Boston Market purchase.

The technology was credible. Dynamic Yield’s models drove drive-thru menu boards that changed in real time based on weather (cold day, more coffee), time of day (morning skews breakfast longer), current traffic at the location, and prior order patterns. The promise: bigger average tickets and faster decision times at the most important channel in the chain — the drive-thru.

It mostly worked. Internal tests showed ticket lifts. The technology deployed across thousands of locations through 2019 and 2020.

The strategic case did not survive 24 months.

In December 2021, McDonald’s sold Dynamic Yield to Mastercard for an undisclosed price, taking a reported book gain of approximately $271 million. McDonald’s still uses the technology under license. But the strategic bet — we will own the AI personalization stack — was abandoned in less than three years.

What killed it was not that the AI failed. It was that the AI tooling market commoditized faster than McDonald’s could justify being an in-house operator of one. The strategy that was a clear win in 2019 (“we will own this capability”) was a clear loss in 2021 (“why are we operating an Israeli AI startup instead of paying a vendor?”). Same technology. Same numbers. The market context shifted underneath the bet.

This is the bet every Fortune 500 board is making in 2026 with their first agentic AI buildouts. Build it in-house, lock in the talent, own the stack. The Dynamic Yield case says: the technology will probably work, the operating cost will be real, and the strategic case may not survive the duration of the underperformance phase.

Act 4 — The accident (March 2020)

In the first quarter of 2020, every assumption in the prior three acts inverted.

Restaurant lobbies closed worldwide. The drive-thru — the channel Dynamic Yield had been built to optimize — became 70% or more of system revenue in many markets. The mobile app — which had been underperforming through 2019 — became one of the most-downloaded retail apps in North America inside a single quarter.

McDonald’s reported that 2020 digital sales exceeded $10 billion across the company’s top six markets (U.S., Australia, Canada, France, Germany, U.K.). The infrastructure that had been treated as a cost center for five years was suddenly the entire business.

Read carefully: McDonald’s did not win the pandemic because the strategy was right. They won the pandemic because they had already paid for the infrastructure, had already survived the franchisee revolt, and had already absorbed the Made for You and Dynamic Yield losses. The bet was not vindicated by execution. It was vindicated by an external catalyst nobody at McDonald’s could have planned for.

This is the architectural point that matters more than any other in the saga: every infrastructure bet has an underperformance phase. The question is not whether the bet pays off. The question is whether you survive the underperformance phase long enough for a catalyst to arrive.

You cannot plan the catalyst. You can only plan the survival.

McDonald’s, with $20+ billion in annual revenue and a global franchise base that absorbs most of the cost, can absorb a five-year underperformance phase. Most companies cannot. The mid-cap retailer, the regional bank, the manufacturer betting on AI in 2026 — most of them do not have the balance sheet to survive five years of “the data does not yet show ROI” before an external event vindicates the spend.

That is the actual bet being placed in every Fortune 500 boardroom right now. Not “is AI real.” Real or not, the underperformance phase is structural. The question is whether the company has the runway, and whether the executive who placed the bet still has a job when the runway ends.

Act 5 — IBM voice AI (2021–2024)

The fourth and most recent cycle is also the most explicit AI parallel.

In 2021, McDonald’s announced a partnership with IBM, in which IBM acquired McDonald’s existing automated drive-thru voice ordering technology (originally built via the 2019 acquisition of a startup called Apprente) and re-deployed it at scale across pilot restaurants. The system was deployed at over 100 McDonald’s locations in the United States.

It mostly didn’t work.

A series of viral TikTok and Instagram videos through 2023 showed the AI in failure mode. One customer reported it added 260 Chicken McNuggets to a single order. Another video showed it processing a request for “one cup of water” as “nine sweet teas, no straws.” A third showed it repeatedly refusing to remove bacon from a Quarter Pounder order. The videos collectively reached tens of millions of views — orders of magnitude more reach than McDonald’s own announcements of the program.

In June 2024, McDonald’s confirmed the IBM partnership for automated voice ordering would end. The official statement included the phrase “continue to evaluate long-term, scalable voice ordering solutions” — almost identical in tone to the post-Made-for-You and post-Dynamic-Yield language from prior cycles. The infrastructure was not killed. It was paused, repackaged, and queued for the next vendor.

Fourth cycle. Same shape. Same vocabulary.

The pattern, stated plainly

Across 25 years and roughly $6 billion of disclosed spend, every McDonald’s tech bet has followed the same five-step shape:

1. The vision. “If we are not on this technology, we will be left behind.”
2. The spend. Material enough to matter, small enough that the company can survive losing it.
3. The underperformance. Twelve to thirty-six months of “the data does not yet show ROI.”
4. The fork. Either an external event vindicates the spend (COVID validated Experience of the Future), or an executive change kills it (Greenberg out after Made for You, Kempczinski killing the IBM voice AI deal in 2024).
5. The next vision. Often by the same company. Often with the same justification.

Notice what is missing from this list: a step where the company carefully measures ROI and proceeds rationally. That step does not exist in any of the four cycles. The decisions are made on a longer time horizon than any single quarter’s data, and the consequences land long after the executive who signed off has moved on.

Notice what else is missing: a step where the technology turns out to be fake. Made for You worked. Kiosks work. Mobile ordering works. Dynamic Yield’s personalization worked. IBM’s voice AI worked some of the time in a deployment too thin to make it work the rest of the time. The technology is almost always real. The bet that fails is the strategic bet around the technology — that the company should own it, that the timing is right, that the franchise base will absorb the cost, that the underperformance phase is shorter than the company’s runway.

What this means for enterprise AI in 2026

The 2026 enterprise AI capex cycle is the largest infrastructure bet ever placed by the corporate sector. Estimates of global enterprise AI spend in 2026 range from $200 billion to north of $400 billion depending on what is included. Whatever the right number, it is at least an order of magnitude larger than the entire 25-year McDonald’s saga compressed into a single year of spending.

The shape will not be different.

Most of those 2026 deployments are already in the underperformance phase. Pilots that ship but do not measurably increase revenue. Internal copilots that get used by a quarter of the workforce. Customer-facing agents that work 95% of the time and produce viral failure videos on the 5% (compare: 260 chicken nuggets). The pattern is not anomalous — it is identical to where McDonald’s was on Made for You in 1999 and on Dynamic Yield in 2020.

Three questions follow from the McDonald’s case, and they are the questions every architect, every founder, and every operator should be asking inside any company that is currently mid-AI-deployment.

1. Is this Made for You, or is this Dynamic Yield? That is, is the bet on owning a capability we will not need to own in three years (Dynamic Yield), or on a foundational operational rebuild that takes five years to pay back (Made for You)? Different shapes. Different exit conditions.

2. What is our COVID? What external event would retroactively justify this spend? If the answer is “we cannot name one,” the runway has to be longer than the underperformance phase, or the bet is not viable.

3. Whose career ends if this gets killed in 2028? Every prior cycle had a named executive whose career trajectory bent on the outcome. The bet was not really on the technology. The bet was on a person, and the person’s tenure.

McDonald’s has survived four of these cycles and is large enough to bet again. Most companies are not McDonald’s. The interesting question is not whether agentic AI is the future. The interesting question is which of your competitors has the runway to survive long enough to find out, and which of them does not.

Three things to take from this

Every infrastructure bet has an underperformance phase. The bet is not whether the technology is real. The bet is whether you survive the phase before the catalyst arrives — or, more honestly, whether a catalyst arrives at all.

External catalysts validate strategies the strategy itself could not. COVID did not vindicate Experience of the Future on the merits. COVID vindicated it by changing the merits. You cannot plan the catalyst. You can plan the runway.

McDonald’s has done this four times. Most companies do not get four chances. The 2026 enterprise AI bet is your company’s first. Place it accordingly.