The AI Product Graveyard Keeps Growing in 2025

Dozens of AI products and startups have been quietly discontinued, acqui-hired, or abandoned since the generative AI boom began in late 2022. The growing 'AI product graveyard' now includes high-profile casualties from Google, Microsoft, Meta, and countless venture-backed startups — exposing a brutal reality beneath the industry's trillion-dollar valuations.

What was once a gold rush has become a reckoning. As the market matures, the gap between AI hype and sustainable business models has claimed products that once commanded millions of users and billions in investment.

Key Takeaways: The AI Shutdown Trend

• ChatGPT Plugins, once heralded as the 'App Store moment' for AI, were discontinued by OpenAI in favor of GPTs and the newer Actions framework
• Microsoft Cortana was officially killed in 2023 after years of declining relevance, replaced by Copilot branding
• Inflection AI was effectively gutted when Microsoft hired its co-founders and most of its staff in a $650 million deal
• Artifact, the AI-powered news app built by Instagram's co-founders, shut down in early 2024 despite strong user engagement
• Neeva, the AI-powered search engine, closed its consumer product and was acquired by Snowflake for roughly $150 million
• Stability AI underwent massive leadership upheaval, layoffs, and near-collapse before restructuring in 2024

The Acqui-Hire Graveyard: Talent Over Products

Big Tech's preferred acquisition strategy in the AI era isn't buying products — it's absorbing talent while letting the original product die. This pattern has become so common that regulators have started scrutinizing these deals as disguised acquisitions.

The most dramatic example remains Inflection AI. Founded by DeepMind co-founder Mustafa Suleyman, the company raised $1.5 billion and launched Pi, a conversational AI assistant. Then Microsoft hired Suleyman as head of its consumer AI division, along with most of Inflection's engineering team. The $650 million Microsoft paid was framed as a 'licensing fee,' but the result was clear: Pi was effectively orphaned.

Character.AI followed a similar trajectory. Despite having one of the most engaged user bases in consumer AI — with users spending an average of 2 hours per session — the company's founders and key researchers were absorbed by Google in a deal worth approximately $2.7 billion. The product technically continues operating, but its long-term trajectory without its core team remains uncertain.

Adyen co-founder Pieter van der Does once noted that the best companies are bought, not sold. In AI, the twist is darker: the best teams are bought, and the products are left behind.

Consumer AI Products That Didn't Survive

The consumer-facing AI space has proven especially unforgiving. Products that generated enormous initial excitement have struggled to retain users or find revenue models that justify their compute costs.

Artifact stands as perhaps the most poignant example. Built by Kevin Systrom and Mike Krieger — the duo behind Instagram — the AI news aggregation app launched to significant fanfare in early 2023. It used large language models to summarize articles, personalize feeds, and even rewrite headlines. Despite strong engagement metrics and critical praise, Systrom announced the shutdown in January 2024, citing an insufficient market size to justify continued investment.

Neeva attempted to challenge Google Search with an ad-free, AI-enhanced search engine that charged users $5.99 per month. Founded by former Google executive Srinivas Laxman, Neeva was one of the first search products to integrate LLM-generated answers directly into results. But consumer willingness to pay for search proved virtually nonexistent. Snowflake acquired the company in 2023, primarily for its AI talent and technology rather than its consumer product.

Other notable consumer casualties include:

• Lensa AI — the AI avatar app that went viral in late 2022, generating $30 million in a single month before downloads cratered by 95%
• Replika's companion features — repeatedly rolled back due to safety concerns, alienating its most dedicated users
• Google Bard — not technically dead, but fully rebranded to Gemini after persistent perception problems
• Microsoft's Bing Chat — similarly rebranded to Copilot after the 'Sydney' incident and failure to meaningfully capture search market share
• Humane AI Pin — launched at $699 plus a $24/month subscription, received devastating reviews, and the company reportedly sought a buyer at $1 billion

The Plugin and Platform Casualties

OpenAI's ChatGPT Plugin ecosystem deserves special attention as a cautionary tale about premature platform ambitions. When plugins launched in March 2023, the tech press declared it AI's 'iPhone moment.' Companies rushed to build integrations. Venture capital flowed into 'ChatGPT plugin startups.'

Less than a year later, OpenAI quietly deprecated the entire plugin system. Usage data reportedly showed that fewer than 5% of ChatGPT Plus subscribers regularly used plugins, and the experience was clunky compared to native capabilities. OpenAI replaced plugins with custom GPTs, then with the GPT Store — which itself has struggled with discoverability and monetization challenges.

This pattern of rapid platform launches followed by quiet abandonment has eroded developer trust. Companies that invested significant engineering resources into building ChatGPT plugins found themselves stranded. The same risk now looms over developers building custom GPTs, MCP integrations, and other AI platform extensions.

The lesson is stark: building on top of AI platforms carries enormous platform risk, arguably more than traditional app store ecosystems. The underlying technology and product strategy are evolving so rapidly that today's integration point may not exist in 6 months.

Enterprise AI: Quieter Deaths, Bigger Stakes

While consumer AI failures make headlines, the enterprise AI graveyard is arguably more consequential. IBM Watson — once the poster child for enterprise AI — underwent such a dramatic repositioning that the Watson brand has been largely retired. IBM sold Watson Health in 2022 and has since reoriented entirely around its Granite models and watsonx platform.

Multiple enterprise AI startups that raised significant rounds in 2023 have since quietly shut down, pivoted, or been absorbed:

• Jasper AI raised $125 million at a $1.5 billion valuation for AI content generation, then laid off staff and pivoted toward enterprise marketing workflows as ChatGPT commoditized its core offering
• Writer and Copy.ai have survived but significantly repositioned from 'AI writing tools' to broader enterprise workflow platforms
• Tome, the AI presentation tool, pivoted away from consumers entirely after finding that free AI alternatives decimated willingness to pay
• Several AI coding assistant startups have been squeezed out as GitHub Copilot, Cursor, and free alternatives from Google and Amazon dominate the space

The enterprise AI shakeout reflects a fundamental challenge: when foundation model providers keep expanding their capabilities, vertical AI startups built on narrow use cases face constant commoditization pressure.

Why AI Products Fail: The Common Patterns

Analyzing the AI product graveyard reveals several recurring failure modes that transcend individual companies.

The 'thin wrapper' problem remains the most common killer. Products that provide a user interface layer on top of foundation model APIs without significant proprietary value are extremely vulnerable. When OpenAI, Google, or Anthropic add a feature natively, wrapper products lose their reason to exist overnight. This dynamic has been compared to building a restaurant on rented land — the landlord can always open their own restaurant next door.

Unsustainable unit economics claim others. AI inference costs, while declining, remain substantial. Consumer products offering free or cheap AI access often face the dilemma of either charging prices users won't pay or subsidizing usage at rates investors eventually won't fund. Lensa AI's viral spike and subsequent collapse perfectly illustrates this — massive user acquisition without a path to retention or profitability.

Regulatory and safety whiplash has forced several products to gut their most popular features. Replika's removal of romantic companion features — then partial restoration — alienated users in both directions. Character.AI has faced intense scrutiny over youth safety concerns, potentially threatening the product's core appeal.

What This Means for Developers and Investors

The growing AI graveyard carries practical implications for anyone building or investing in the space. Defensibility has become the single most important evaluation criterion for AI products, surpassing growth metrics and technical sophistication.

Products with the best survival odds share common characteristics: proprietary data moats, deep workflow integration that creates switching costs, network effects that improve with scale, and pricing models aligned with measurable customer value. Products lacking these qualities — no matter how impressive their demos — face existential risk.

For developers, the message is clear: avoid building on any single AI platform's ephemeral features. Design for portability. Build proprietary value in the layers you control — data pipelines, user experience, domain expertise — rather than in proximity to a foundation model API.

Looking Ahead: The Next Wave of Shutdowns

The AI product graveyard will likely grow significantly through 2025 and 2026. Several categories appear particularly vulnerable.

AI search startups face an increasingly hostile competitive environment as Google integrates AI Overviews, OpenAI scales ChatGPT Search, and Perplexity fights for market share with limited resources compared to its Big Tech competitors. Not all can survive.

AI hardware products — including smart glasses, AI pins, and dedicated AI devices — face the fundamental challenge that smartphones already exist. The Humane AI Pin's struggles and the tepid reception of the Rabbit R1 suggest this category may produce more casualties than winners.

Single-purpose AI agents built for narrow tasks risk being subsumed by general-purpose agent frameworks from OpenAI, Google, and Anthropic. As these platforms build increasingly capable agent systems, standalone AI agent startups will need to demonstrate clear superiority in specific domains to justify their existence.

The AI industry's creative destruction is not inherently negative — it is a sign of rapid maturation. But for founders, developers, and investors, the graveyard serves as an essential reminder: in AI, building something impressive is easy. Building something durable is the real challenge.