3BQRIIINO. INDEPENDENT TECH JOURNAL
← All stories

AI Industry / AI Startups / Tech Analysis · November 4, 2026

The AI Graveyard: Four Ways AI Products Die — and What Survives Them

Beyond the shutdown lists: a decade of failed AI products falls into four patterns — compute costs, trust failures, demos that were not real, and the platform squeeze.

The AI Graveyard: Four Ways AI Products Die — and What Survives Them

Every few months, a new list of dead AI products makes the rounds. TechCrunch now keeps a running one, and the image is backed by a bleak statistic: in S&P Global Market Intelligence's 2025 Voice of the Enterprise survey of 1,006 IT and business professionals at companies actively investing in AI, 42% said their company had abandoned the majority of its AI initiatives before production, up from 17% a year earlier. The average organisation scrapped 46% of its projects between proof of concept and broad adoption.

But a list of names teaches less than the pattern underneath them. Look closely at a decade of failures — from Microsoft's Tay in 2016 to Character.AI's bankruptcy filing in October 2026 — and the deaths cluster into four distinct causes. Just as important: some entries in the graveyard are not dead at all, and telling the difference matters before drawing lessons from them.

1. Death by compute bill

AI products have a cost structure most software never faces: every active user consumes expensive computation, every day. Character.AI is the clearest case study we have, because its numbers are now public. In a first-day affidavit filed with its Chapter 11 petition in Delaware on 12 October 2026, CEO Colin Adams disclosed that compute spending ran at roughly $13.8 million a month in late 2024 and about $10 million a month through 2025 — and still totalled $56.1 million in the first eight months of 2026 — while monthly revenue only reached about $5.3 million by August 2026. Demand was never the problem; the filings put current and past users at around 163 million. Each user simply cost more than they paid.

Two precision notes matter here. Character.AI is in bankruptcy, not shut down: the platform is still operating during a court-supervised sale, and Disney's $20 million stalking-horse bid is for a perpetual, non-exclusive licence to its intellectual property — not for the app or its users. And even OpenAI is not immune to this arithmetic: it discontinued the Sora web and app experiences on 26 April 2026 and the Sora API on 24 September 2026, folding video generation into ChatGPT rather than carry a standalone product's costs.

2. Death by trust

The fastest deaths in the graveyard were not financial. Microsoft launched Tay, a chatbot designed to learn from Twitter conversations, on 23 March 2016; coordinated manipulation turned it abusive, and it was offline within about 16 hours. Meta's science model Galactica lasted three days in November 2022: released as a public demo on 15 November, it was confidently fabricating citations and producing biased text, and Meta pulled the demo on 18 November. In both cases the technology "worked" — what failed was the assumption that a model safe in the lab stays safe in public.

Trust also compounds the first cause of death. Character.AI's own filing cites product-safety litigation and a probe by 46 states and territories among the factors that thwarted its attempts to raise capital or sell itself. Guardrails are not a compliance luxury; in this industry they are a financing condition.

3. Death by demo

Builder.ai sold the dream of building an app "as easily as ordering a pizza," raised roughly $445 million from backers including Microsoft and the Qatar Investment Authority, and was valued at around $1.5 billion. It announced insolvency on 20 May 2025. Reporting by the Wall Street Journal as early as 2019, and later by the Financial Times and Bloomberg, described a company whose "AI" output relied heavily on human engineers, and whose reported 2024 revenue was revised down from about $220 million to roughly $55 million after a change of leadership; lenders seized some $37 million shortly before the collapse.

Accuracy requires two caveats. The company described its model as "human-assisted AI," and the viral version of the story — hundreds of engineers secretly pretending to be an AI — has itself been challenged by former employees and by the Financial Times, which traced the most repeated version of the claim to a social media post rather than reporting. The documented core is narrower and still damning: a wide, years-long gap between what customers and lenders were told and what the numbers later showed. A demo that outruns the product is a debt that eventually comes due.

4. Death by platform squeeze — and the giant that dodged it

Relay, a five-year-old AI workflow-automation tool built as an alternative to Zapier, announced its shutdown on 16 July 2026. As OpenAI, Google, and other platforms built equivalent automation into their own products, a standalone tool lost its reason to exist. Its closure came in stages — free accounts ended on 15 August 2026 and the service switched off for paying customers on 14 September — and its founder and much of the team moved to Google's Chrome team.

Apple's Siri belongs in this story as the counter-case, which is why it does not belong in a graveyard at all. After its promised personalised Siri slipped repeatedly, Apple did not shut the project down: on 12 January 2026 Apple and Google announced a multi-year collaboration making the next generation of Apple Foundation Models based on Google's Gemini technology, and the rebuilt Siri began rolling out in beta with iOS 27 on 14 September 2026. (The often-cited figure of about $1 billion a year is Bloomberg-reported, not confirmed by either company.) When a platform giant cannot ship alone, it partners. When a startup gets squeezed by the platforms, it usually cannot.

What survived the graveyard

A striking share of the "dead" were absorbed rather than erased. HP acquired Humane's AI platform, team, and 300-plus patents for $116 million in February 2025 — while the AI Pin itself, excluded from the deal, lost its server connection ten days later and became a paperweight: the product died, the assets did not. Sora's video generation lives on inside ChatGPT, as do the useful parts of OpenAI's Atlas browser, retired standalone in August 2026. Relay's people now build automation inside Chrome. The technology, the teams, and the ideas mostly survived; the standalone businesses did not.

A checklist for creators and small teams

The graveyard will keep growing — Gartner reported in January 2026 that at least 50% of generative-AI projects were abandoned after proof of concept. Read it as a taxonomy, not an obituary page: most of these products were killed by one of four causes, and all four are visible early, if you are willing to look.

Sources and classification notes

Checked 4 November 2026. Events described are confirmed news unless labelled otherwise in the text; Character.AI is an in-progress bankruptcy (platform operating), Apple/Siri is a pivot, and Builder.ai details are presented as reported by the cited outlets, with the company's position noted.

Prefer reading in Arabic? You can translate this article into Arabic with Google Translate.