Snap Lost $40 Million on Sunglasses in a Vending Machine. It Was Tuition.

In late 2016 you could not buy Snap's first piece of hardware in a store. You bought it from a cartoonish yellow vending machine parked near the beach in Venice, California — feed it $130, and a pair of camera sunglasses dropped out.⁴ It was a marketing stunt that worked too well. The lines were real, the hype was real, and the demand signal Snap read off those lines was, it turned out, almost entirely fake. A year later the company quietly told the SEC the bill: $39.9 million in charges, most of it parts ordered for glasses that would never sell.¹ For most companies, that is where a hardware story ends. For Snap, it was barely the first chapter.

The official story is that Spectacles was an expensive flop Snap got embarrassed by and moved on from. The truer story is that Snap was never trying to sell sunglasses. It was buying tuition — a decade of it — on the only question its founder seems to actually care about: who owns the camera you wear on your face when the phone stops being the screen.

A camera company that happens to run a social app

Snap insists it is a camera company. Everyone hears that as a cute slogan, because the cash comes from ads sold against a messaging app. But follow the money the company spends rather than the money it makes, and the slogan reads like a confession. The hardware bet predates Snapchat's IPO: in December 2014 Snap quietly bought Vergence Labs — a maker of camera-equipped eyewear — for $11 million in cash plus $4 million in stock vesting over two years.³ That was the seed. Spectacles was the sprout. And the spending never stopped — in fiscal 2024 alone Snap poured $1,692 million into R&D while posting an $787 million operating loss on $5,361 million of revenue.² You do not run that engine for a decade to garnish a chat app. You run it because you believe the chat app is the bridge, not the destination.

Here is the thesis, plainly: Snap's AR bet is not a side project and not a pivot — it is the company's primary identity, financed by an ad business it treats as a means to an end. The wager is that the next computing surface is a lens, not a slab, and that owning the lens early is worth losing money on it for a very long time. The $39.9 million write-down was not the bet failing. It was the bet learning.

Why the vending machine lied

The mechanism behind that first loss is worth dwelling on, because it explains everything Snap did afterward. Spectacles Gen-1 wasn't sold through normal retail; it was sold through scarcity theater. The Snapbot vending machines created queues, and queues create a demand signal that points the wrong way. Hype is a measure of curiosity, not of intent to keep wearing. Snap read the lines as orders, committed to long-lead-time parts to match, and then watched the curiosity evaporate the way curiosity does. Spiegel said it himself: the company 'ordered a lot of long-lead time parts, and ultimately weren't able to sell as many Spectacles as we thought.'⁵ That is not a manufacturing failure. It is a measurement failure — mistaking a stunt's crowd for a market.

And notice what the write-down actually was. It was not a warehouse of unsold glasses marked to zero. It was a controlled detonation across three lines: $19.5 million in excess inventory reserves, $17.9 million in cancelled purchase commitments, and $2.5 million in asset impairments.¹ The biggest piece — the cancelled commitments — is the sound of a company hitting the brakes mid-order. Snap took the pain, told the SEC the truth, and then did the one thing a learning organization does: it changed the cadence. From that point on, the hardware stopped chasing consumers and started serving developers — building the software platform first and letting the audience come later.

From a $99/month wall to a $2,195 storefront

By 2024 the hardware had become something genuinely ambitious. The fifth-generation Spectacles were standalone AR glasses — no phone tether — running Snap OS on Snapdragon chips, with a 46-degree stereo display projecting digital objects into the room.⁶ But here is the tell that the bet was still in its incubation phase: you couldn't buy them. Snap sold access, not product. The price was a developer-programme commitment of $99 a month for a minimum of one year — at least $1,188, and a deliberate wall against curiosity tourists.⁶ That wall is the entire post-2017 strategy in one number. Snap had learned that you do not ship to consumers until the thing consumers will actually use — the apps, the lenses, the reasons to wear it — already exists. So it rented the hardware to the people who would build those reasons.

Then, in June 2026, the wall came down. At Augmented World Expo, Snap introduced Specs — note the name, deliberately broken off from the old 'Spectacles' brand — as the first AR glasses it would sell to the public, at $2,195, with pre-orders open and fall delivery promised.⁷ Fully standalone, two Snapdragon processors, four hours of mixed-use battery.⁷ After a decade, Snap finally has a storefront instead of a vending machine, and a product instead of a stunt. The bet, at last, has a proof point with a price tag attached.

Isn't this just a small company picking a fight it can't win?

The honest objection writes itself. Snap is losing money — $787 million in operating loss in 2024 — while pouring $1.69 billion into R&D against rivals who can each spend more than Snap's entire revenue on the same category.² Meta, Apple, and Google have the balance sheets to wait Snap out and the distribution to commoditise glasses the moment the market is real. A $2,195 price tag is a developer's price wearing a consumer's badge; it is not a mass-market device. And the organization is visibly strained — the SVP running the Specs effort departed abruptly in early 2026 after a reported clash with Spiegel.⁸ That is not the picture of a smooth march to victory.

All true. But the steelman has a counter. Snap's edge was never capital — it was a ten-year head start on the part nobody can buy quickly: the lens software, the creator culture, the muscle memory of building AR people actually use. While the giants debated whether the category existed, Snap shipped five generations and ate its mistakes in public. The Qualcomm partnership announced in 2026 means it no longer carries the silicon burden alone.⁸ The question is not whether Snap can outspend Meta. It can't. The question is whether owning the developer ecosystem and the creative tooling matters more than the balance sheet when the consumer category finally turns — and whether Snap can convert a decade of loss-making R&D into revenue before someone richer makes the glasses cheap. That is a real bet with a real way to lose. Which is exactly what makes it a bet and not a press release.

Snap spent ten years and an unbroken string of losses to earn the right to ship a pair of glasses people can actually buy. The vending machine was never the strategy — it was the first, costly lesson that a crowd is not a market. Everything since has been Snap paying down the difference between curiosity and intent, one generation at a time. Whether the bet pays is now a question the public, not a developer programme, gets to answer. But the genius — or the folly — was always the same move: deciding that a camera company is one that builds the camera, and being willing to lose money on it for as long as it takes to find out if you were right.