TheNumbers.com collapse signals fragility of open web under AI pressure
Founder Bruce Nash says only 10% of traffic is human as AI crawlers strain infrastructure, prompting industry-wide reckoning on bandwidth costs and security.
TheNumbers.com, a definitive repository for film industry data, suffered a catastrophic server collapse on 5 March 2026, leaving the site offline for more than a week. The incident, which resulted in the permanent loss of historical charts, individual movie pages, and the Report Builder tool, highlights the escalating strain on independent digital infrastructure caused by automated artificial intelligence traffic. The site resumed limited operations on 13 March 2026 on entirely new infrastructure, having abandoned legacy systems that were deemed too vulnerable to restore.
Founder Bruce Nash attributed the outage to unsustainable loads from agentic AI scraping and potential cyberattacks. Nash noted that by early 2026, only 10 per cent of the site’s traffic originated from human users, with the remainder comprising AI crawlers and automated bots. This shift from manageable search engine traffic to industrial-scale scraping began around 2024 and intensified significantly by December 2025, overwhelming the site’s capacity to maintain service.
The collapse appears linked to attempts to gain an advantage in prediction markets. Polymarket uses TheNumbers.com’s daily box office performance figures to resolve weekly betting markets. Nash suggested that some automated probes were not merely scraping data but were actively searching for backdoors to access information before public release, allowing traders to front-run positions. While the exact perpetrators remain unidentified, the motive aligns with the financial incentives present in data-heavy prediction markets.
The incident reflects a broader crisis across the internet, where AI crawlers generate disproportionate bandwidth costs without returning human readers. Cloudflare data indicates that while Google crawls approximately five pages per visitor, Anthropic crawls over 38,000 pages per visitor. Similar strain has been reported by organisations including Read the Docs, iFixit, and the Wikimedia Foundation, prompting infrastructure providers to introduce monetisation models such as pay-per-crawl to address the economic imbalance.
Nash explained that restoring the site from backups was rejected due to the age of the 160,000-file legacy codebase and ongoing security probes. The team instead built a skeleton version on new systems to deliver current data while gradually rebuilding historical content. The Numbers is now optimising its content for large language models to direct them toward licensing information via its OpusData service, aiming to support six distinct traffic audiences rather than the traditional human and search engine focus of the past.
