The Eight-Story Glass Tower That Was Always the Problem

In the early 2010s, an entrepreneur named Ernest Garcia III built a used car dealership without dealers. The interface was a website. The pickup mechanism, deployed with theatrical confidence, was a giant glass tower — eight stories tall, transparent, lit at night — into which freshly purchased cars were lowered by an automated mechanism for customer pickup. The customer would arrive at the tower, insert a giant souvenir coin into a kiosk, and watch their car descend through the glass column like a vending-machine candy bar.

The company was Carvana. The market loved it. At peak in August 2021, the stock traded above three hundred and seventy dollars, valuing the company at roughly sixty billion dollars. Eighteen months later, in late 2022, the stock traded below four dollars. The peak-to-trough decline was approximately ninety-nine percent. The vending-machine glass towers, by then numbering more than thirty, were still standing across the United States, mostly idle, mostly unused.

The Cash-Burn Math. The model the market had originally rewarded was a particular kind of e-commerce assumption: that buying a used car online would scale with the same unit economics as buying a sofa or a pair of shoes — except with a much larger ticket size, which would generate operating leverage as the company grew. The reality, as the company's financials slowly revealed, was different. Used cars depreciate while sitting in inventory. Carvana's average inventory turn was slower than traditional dealerships. The cost of acquiring each customer through digital advertising was meaningfully higher than projected, partly because used-car purchases are infrequent (every five to seven years) and therefore the lifetime customer value calculation was less favorable than the marketing models had assumed.

The vending-machine towers, themselves, cost between three and six million dollars apiece to build. They were never the primary mechanism by which Carvana delivered cars to customers — most cars were delivered by truck, like any other dealer's delivery service. The towers were marketing. The marketing was expensive. The cars were depreciating. The customer acquisition cost was rising. The unit economics, when finally pieced together, did not work.

The Debt Spiral. By 2022, Carvana was carrying approximately seven billion dollars of debt against a market capitalization that was rapidly shrinking. The company's interest expense alone was running at several hundred million dollars annually. Cash burn from operations was material. The company narrowly avoided bankruptcy through a debt restructuring negotiated under duress, in which lenders accepted partial concessions in exchange for new collateral terms. The Garcia family, which had personally guaranteed substantial holdings, lost most of their on-paper wealth. The stock recovered partially in 2023 and 2024 as the business slowly stabilized at a much smaller scale — but the period from peak to trough remains one of the most consequential single-company corrections of the post-pandemic era.

The Vending Machine Was Always the Problem. The lesson, available with hindsight but not with foresight, was that the glass tower was a brand metaphor mistaken for an operational mechanism. The tower told a story — that buying a used car could be as simple as a vending machine transaction, that the friction of the traditional dealership was being engineered out of the process, that this was a technology company rather than a car company. The story was compelling. The story attracted capital. The story did not, however, change the underlying physics of the used-car industry, which is a low-margin, inventory-intensive, geographically distributed business with very limited operating leverage.

E-commerce, at its most successful, is a logistics business in a vertical that happens to be reachable through a website. Carvana, in retrospect, was a used-car business that had built a website and a marketing tower. The website did not transform the unit economics. The tower did not transform the unit economics. The peak valuation, sixty billion dollars in 2021, was a story that the market briefly believed and then, with unusual speed, stopped believing. The towers remain. They are, in their way, the most expensive marketing collateral ever erected by an American startup — eight-story monuments to a particular kind of capital-markets enthusiasm that could not survive contact with the depreciation curve.