The AI graveyard: a running list of projects and startups that didn't make it

The world of Artificial Intelligence moves at an incredible pace, often painted with stories of groundbreaking innovations and billion-dollar valuations. Yet, beneath the surface of success, there's a growing "AI graveyard" – a landscape littered with projects and startups that didn't quite make it. From the ambitious endeavors of tech giants to promising startups, many AI initiatives have struggled to meet expectations, faced technical roadblocks, or simply ran out of steam. This isn't just a tale of failure; it's a critical learning opportunity for the entire industry, revealing the complex challenges and harsh realities of building and deploying AI.

Recent data paints a stark picture: the failure rate for AI startups is estimated to be as high as 85% to 90%, a figure notably higher than that for traditional technology companies. Some research even suggests an overall failure rate of 92% for AI and tech startups. This translates into significant capital destruction, with one analysis revealing $6.6 billion burned across just 34 AI startup failures, and another documenting over $226 billion in capital raised and lost across 562 documented failures. Even after the ChatGPT hype wave, 24 AI startups that launched in 2023-2024 collectively saw $461.7 million in venture capital evaporate. These numbers underscore a crucial disconnect between the immense investment and the often-elusive real-world success of AI projects.

Apple's Siri: A Long Road to Redemption

Apple, a company synonymous with intuitive technology, has faced persistent challenges with its voice assistant, Siri. What was once a pioneering feature has, for many, stagnated compared to rapidly evolving competitors. The feed item highlights Siri AI's "repeatedly delayed" enhancements, a sentiment echoed by numerous reports. Apple's initial plans for a more personalized Siri, expected in 2025, were officially pushed to 2026.

The root of these delays appears to be multi-faceted. Reports indicate that Apple was taken aback by the launch of OpenAI's ChatGPT in 2022, prompting a rushed internal effort to catch up in generative AI. A major technical hurdle emerged when engineers attempted to merge Siri's existing, legacy code with new AI features, leading to significant complications. Internal testing of Apple's own large language model (LLM) chatbot reportedly showed it lagging behind ChatGPT in accuracy by about 25%. Adding to the woes, communication breakdowns between Apple's product development and marketing teams meant some advertised features for products like the iPhone 16 were not even close to being ready.

In an effort to accelerate its AI capabilities, Apple announced a partnership with OpenAI at WWDC 2024, intending to integrate ChatGPT for requests Siri couldn't handle. However, even this integration saw delays, pushed to December 2024. As of September 2026, Apple is releasing a redesigned Siri AI, but it's launching in beta, with potential daily usage caps and future paid access, and users may even need to sign up for a waitlist. While the goal is a Siri capable of natural conversations, answering open-ended questions, and retrieving personal information across apps, the journey has been marked by significant internal struggles and external partnerships to bridge the AI gap.

OpenAI's "Super App" Ambitions: A Messy Launch

OpenAI, the company that ignited the generative AI boom with ChatGPT, also experienced a bumpy road with its ambitious "super app" launch. On July 9, 2026, OpenAI transformed its existing ChatGPT app into a "do-it-all" platform, integrating its standalone coding agent app, Codex, and introducing a "Work" mode alongside "Chat" and "Codex." The original ChatGPT app was rebranded as "ChatGPT Classic."

However, the rollout was met with widespread user confusion and frustration. Many users struggled with the unintuitive navigation and found it difficult to differentiate between the various modes. The backlash was significant enough that OpenAI President Greg Brockman publicly acknowledged the app was "kind of a mess." Tech writer and investor M.G. Siegler critically described it as the "ChatGPT 'Super App' Sort of Super Sucks."

Despite the initial fumbles, the integration did see some positive outcomes, particularly for Codex, which jumped from 5 million to 10 million users in a matter of days. However, critics pointed out that this growth might have been a result of "forcing" the coding tool onto general ChatGPT users. OpenAI's strategic shift towards a "super app" is reportedly driven by intense competition from rivals like Anthropic and Google, and a desire to move towards more agentic AI and enterprise solutions. This strategic pivot was so pronounced that a senior OpenAI employee reportedly declared, "Chat is dead," signaling a move beyond purely conversational interfaces. In response to user feedback, OpenAI has plans for a major redesign to simplify the app's interface. It also quietly wound down its Atlas web browser, consolidating its efforts.

Beyond the Giants: Startup Casualties in the AI Race

While the struggles of tech titans make headlines, the AI startup landscape is where the vast majority of projects fail. The reasons are varied and often interconnected, highlighting the inherent risks and complexities of building innovative AI solutions.

A dominant cause of failure for AI startups is intense competition. Many promising ventures are simply "outgunned" by well-capitalized incumbents and tech giants who possess superior models, massive data advantages, and extensive distribution networks. These larger players can often deploy AI features as loss leaders or bundle them into existing products, making it nearly impossible for smaller startups to compete.

Another critical factor is the sheer capital intensity of AI development. Training and inference costs, data acquisition and labeling, and the demand for specialized machine learning talent commanding premium salaries often lead startups to "run out of cash." Companies like Olive AI, which aimed to automate healthcare transactions, burned through $852 million before collapsing due to unit economics that simply didn't work – the cost of running their AI models exceeded what customers were willing to pay.

Poor data quality and a lack of AI-ready data are cited as primary causes in an astonishing 85% of failed AI projects. An AI model trained on messy, incomplete, or poorly governed data will underperform, hallucinate, and erode user trust. Only a small fraction, about 12%, of organizations possess data of sufficient quality to support AI applications.

Many startups also fall into the trap of lacking product-market fit. They build impressive technology without adequately identifying a genuine market need or customers willing to pay for the solution. For example, Utrip, an AI-powered trip planning service, flopped because consumers weren't interested in paying for AI-generated itineraries, despite the advanced technology. Similarly, Forward Health's AI-powered health kiosks struggled with infrequent usage compared to their subscription price, missing a crucial signal about customer engagement. The "AI-washing" phenomenon, where companies claim AI capabilities that are actually human-powered, also led to failures like Builder.ai, which eventually filed for insolvency after losing $445 million.

Over-ambitious goals and technical hurdles have also claimed significant casualties, particularly in complex domains. Autonomous vehicle companies like Argo AI (which burned $3.6 billion) and TuSimple (which lost $1 billion) faced immense technical challenges and a perpetually distant path to profitability. Volkswagen's Cariad project, aiming to create a unified AI-driven operating system for 12 brands, resulted in $7.5 billion in losses and severe product delays by attempting a "big bang" transformation instead of iterative integration.

Other notable AI project failures include:

• Amazon Recruiting AI (2018): Scrapped due to gender bias in its automated resume screening.
• IBM Watson for Oncology (2017-2019): Failed to deliver on its promise due to the complexity of medical data and reliance on synthetic training data that didn't reflect real-world scenarios.
• Zillow Offers (2021): Suffered significant losses because its AI models couldn't adapt to market volatility in real estate.
• Microsoft Tay Chatbot (2016): Shut down rapidly after it began posting racist and offensive tweets due to malicious user interaction.
• Google AI Overviews (2025): Became infamous for "confident hallucinations," providing factual errors like suggesting adding glue to pizza or eating rocks for health, prioritizing fluency over factual accuracy.
• McDonald's AI drive-thru (IBM partnership): Ended due to customer frustration and the AI's inability to accurately process orders, leading to humorous but problematic errors like adding 260 Chicken McNuggets.
• Air Canada's chatbot (2024): Ordered to pay damages after providing a customer with incorrect information about bereavement fares.
• Jasper (2023): Saw its valuation plummet by 80% as OpenAI's ChatGPT began to commoditize the copywriting market.
• Inflection AI (2024): Its talent was absorbed by Microsoft, and its product wound down despite raising $1.5 billion.
• Adept AI (2024): Acquired by Amazon, and its agent product was discontinued after raising $415 million.
• Character.AI (2025): Its core team moved to Google, and the consumer product was de-prioritized after raising over $1 billion.

Common Pitfalls and Lessons Learned

The growing AI graveyard offers crucial insights into the common pitfalls that lead to project and startup failures:

• Lack of Clear Business Objectives: Many projects begin without a well-defined problem to solve, leading to solutions in search of a problem.
• Data Quality and Readiness: The "lack of data" is rarely the issue; it's the lack of "AI-ready data" that cripples projects, as messy or incomplete data leads to unreliable outcomes.
• Underestimating Costs and Complexity: AI development is inherently experimental and capital-intensive, with compute costs, data labeling, and specialized talent often underestimated. The transition from pilot to production is particularly challenging, with 85% of AI projects failing to move beyond the pilot stage.
• Over-reliance on Hype: Founders often assume that simply adding "AI" to a product guarantees instant product-market fit, but users ultimately pay for value, not just an "AI-powered" label.
• Strategic Misalignment and Lack of Oversight: Projects often start in silos without proper executive sponsorship or alignment with broader business strategy, making scaling impossible.
• Ethical Blindness and Bias: AI systems trained on biased historical data can perpetuate and even amplify discrimination, leading to reputational damage and legal issues.
• Competition from Incumbents: Startups face an uphill battle against tech giants who can leverage vast resources, existing user bases, and data advantages.

The lessons are clear: successful AI implementation requires a rigorous focus on defining real-world problems, ensuring high-quality and relevant data, building sustainable business models, maintaining human oversight, and adopting an iterative, adaptable approach. The AI graveyard, while a testament to failed ambitions, is also a valuable repository of knowledge, guiding the next generation of AI innovators toward more sustainable and impactful solutions.

Frequently Asked Questions

What is the "AI graveyard"?

The "AI graveyard" refers to the growing number of Artificial Intelligence projects and startups that have failed, shut down, or significantly missed expectations. This includes ventures from both large tech companies and smaller startups that couldn't achieve sustainable success.

Why do so many AI projects and startups fail?

Many factors contribute to AI project failures, including intense competition from tech giants, running out of funding due to high development costs, poor data quality, a lack of clear product-market fit, over-ambitious goals, and strategic misalignment within organizations.

What are some prominent examples of AI projects that didn't meet expectations?

High-profile examples include Apple's Siri, which has faced repeated delays in significant AI enhancements, and OpenAI's "super app" launch, which was criticized for its confusing user experience. Major startup failures include Olive AI due to unsustainable unit economics, and projects like Amazon's recruiting AI that were scrapped due to bias.

What can be learned from these AI failures?

These failures highlight the importance of clearly defining business problems, ensuring high-quality and "AI-ready" data, developing sustainable business models, maintaining realistic expectations for AI capabilities, and prioritizing ethical considerations and human oversight throughout the development process.