💸 Infrastructure Costs & Inverted Unit Economics

If scaling users drives server bills higher than ad or subscription revenue can cover, high growth will simply bankrupt the company.

Anywhere.FM (YC S07)

• What they built: A pure software, cloud-based web music player. Long before Spotify dominated the streaming market, Anywhere.fm allowed users to upload their entire local MP3 libraries (which were overwhelmingly pirated audio files) to the cloud so they could stream their personal playlists from any web browser in the world.
• The Failure: The product was an instant viral hit, but it suffered from a catastrophic underlying financial model. Hosting, transcoding, and streaming massive volumes of heavy audio files for millions of free users required astronomical cloud server bandwidth. Because they couldn't legally monetize a platform built largely on unlicensed MP3s, every new viral signup simply accelerated their cash burn.
• The Outcome: Suffocating under the immense infrastructure costs of their own popularity and unable to raise further venture capital without a legal revenue model, they sold the company in 2008 to Imeem (a larger, well-funded social music network). Imeem strictly wanted their active user base and engineering talent, and the standalone Anywhere.fm site was quickly absorbed and shut down.

Kifi (YC S14)

• What they built: A pure software knowledge management tool. It was a highly sophisticated browser extension that analyzed everything you read online, created deep-searchable bookmarks, and allowed remote teams to share complex knowledge graphs.
• The Failure: They fell into the classic "vitamins vs. painkillers" trap in B2B software. Users genuinely loved the tool, but they didn't love it enough to pay enterprise SaaS rates for it. Bookmarking and link-sharing are utility features that employees expect to get for free (via browser defaults or Slack). Meanwhile, the server costs for deep-indexing millions of web pages for a predominantly free user base completely outpaced their revenue.
• The Outcome: After struggling to find a viable, venture-scale business model for a standalone bookmarking tool, they were acqui-hired by Google in 2016. Google bought the team to work on a messaging app called Google Spaces, and the Kifi software—along with all user knowledge graphs—was deleted.

Locket (YC W14)

• What they built: An Android-exclusive mobile application that essentially rented out a user's smartphone lock screen. Every time a user woke up their phone, they were shown a highly targeted, full-screen advertisement. In exchange for looking at the ad, the user was paid real cash rewards.
• The Failure: They suffered from brutal, upside-down unit economics. The fundamental flaw of the business model was that paying users cash to view ads completely destroyed the advertiser's Return on Investment (ROI). Advertisers quickly realized the clicks were "hollow"—users were interacting with the ads purely to get the cash payout, not because they actually wanted to buy the underlying product. As advertisers realized the traffic was worthless and fled the platform, the cash rewards dried up, and user retention instantly cratered.
• The Outcome: Unable to make the math work on a financially subsidized ad network, the founders desperately tried to pivot the lock screen into a general news aggregator. It failed to gain traction. Eventually, they were acquired by the e-commerce giant Wish in 2015 strictly to help build mobile growth features. The standalone Locket app was immediately shut down.

Spool (YC S11)

• What they built: A profoundly useful, pure software mobile application and browser extension. It allowed users to bookmark, cache, and automatically download heavy web content (specifically text and streaming web videos) directly to their smartphones so they could consume it later without an active internet connection.
• The Failure: They built a highly viral utility, but they ran into two fatal walls. First, passively downloading and caching third-party web videos for offline viewing triggered massive, existential copyright threats from major media companies. Second, building a standalone company simply to "save links for later" was a fragile business model, as the exact feature was rapidly being cloned and absorbed natively into mobile operating systems (like Apple's Safari Reading List).
• The Outcome: Realizing they were caught between impending copyright lawsuits and native OS obsolescence, the founders executed a strategic exit. In 2012, they were acquired by Facebook. Facebook had absolutely zero interest in maintaining a standalone bookmarking app; they bought Spool strictly to absorb the elite mobile engineering team to help fix Facebook's notoriously clunky early iOS application. The Spool service was instantly shut down upon acquisition.

Vidme (YC W14)

• What they built: A pure software video-hosting platform designed to be a creator-first hybrid between YouTube and Reddit. It allowed creators to easily upload user-generated video content and relied on a community upvote system for organic discovery.
• The Failure: They built a highly successful community, attracting millions of users and billions of views, but they ran into the fatal reality of building a consumer video startup: hosting, transcoding, and streaming massive video files is astronomically expensive. YouTube only survives because it is heavily subsidized by Google's infinite server infrastructure and proprietary advertising engine. Vidme couldn't build a standalone ad network fast enough to monetize their traffic, meaning every new viral video they hosted actively drained their bank account via AWS bills.
• The Outcome: Crushed under the immense cloud infrastructure costs of their own popularity, they simply ran out of runway. The founders suspended operations and completely shut down the video platform in 2017, proving that in consumer video, massive web traffic without a native ad engine is a financial death sentence.

💡 Key Takeaway

For startups in this category, the core challenge is not the code but the surrounding market dynamics. Ensure you validate this bottleneck before scaling.