TechCrunch has published a running inventory of artificial intelligence projects and companies that have either shut down entirely or fallen well short of the ambitions attached to them at launch, with notable entries including Apple's repeatedly stalled Siri overhaul and the turbulent rollout of OpenAI's so-called super app.
The list is, in one sense, unremarkable. Every technology cycle produces its casualties, and the reporters and investors who breathlessly promoted those casualties on the way up rarely volunteer to compile the postmortem. What makes TechCrunch's exercise worth attention is the timing. The AI investment wave that began in earnest after the public release of ChatGPT has now been running long enough that a graveyard has become possible — and crowded enough that maintaining a running list is a coherent editorial project rather than a short blog post.
To understand why that matters, it helps to remember how the current cycle was framed when it began. The argument made by the largest technology companies, and accepted with varying degrees of credulity by financial markets, was that large language models and generative AI tools represented something categorically different from previous software cycles. The implication was that the usual pattern — hype, investment, shakeout, consolidation around a handful of survivors — might not apply, or might apply more gently. What TechCrunch's list documents is that the pattern is applying in the ordinary way, and perhaps more swiftly than the optimists projected.
The Apple entry is particularly instructive because it does not describe a startup that ran out of runway. It describes one of the most capitalised companies in the world repeatedly failing to ship a promised product on schedule. Apple's difficulties with Siri's AI capabilities have been reported extensively, and the core problem appears to involve the tension between the company's historically strict approach to on-device privacy and the infrastructure demands of competitive large language model performance. That tension is not unique to Apple, but Apple's public commitments made its struggles more visible. When a company of that scale struggles to deliver, it is a signal that the engineering challenges in this space are more durable than the product announcement calendars suggested.
The OpenAI super app reference points to a different species of problem. OpenAI occupies a structurally unusual position in the industry: it is simultaneously a research organisation, a product company, a platform on which other products are built, and a focal point for an enormous amount of speculative attention. Launching what would effectively be a consumer interface to compete with the devices and operating systems of its own biggest partners and investors is a legitimately complicated manoeuvre. The suggestion that the launch was messy, as TechCrunch characterises it, fits a broader pattern of the organisation moving quickly into product territory that its internal structure and external relationships may not have been ready to support.
Stepping back, the likely consequences of this shakeout period divide along company size. For well-funded incumbents, failed AI projects are expensive but survivable. Microsoft has absorbed the cost of AI features that did not find traction. Google has navigated product stumbles that would have been existential for a smaller organisation. The damage to them is reputational and measured in opportunity cost rather than solvency.
For the startup layer, the consequences are more severe and the sorting is likely to accelerate. Venture capital deployment into AI has been enormous, but the return horizon for investors is not infinite, and the presence of a credible graveyard list changes the psychology of follow-on funding conversations. A startup that cannot clearly differentiate its product from a capability that a foundation model provider might absorb natively into their platform faces a difficult argument. This suggests the next twelve to eighteen months will see meaningful consolidation, with acqui-hires, quiet shutdowns, and a narrowing of the field to companies that have found defensible niches or genuine distribution advantages.
The consumer end of the market is where the consequences are perhaps most underappreciated. Products that were marketed aggressively to ordinary users and then discontinued or substantially scaled back leave behind a residue of scepticism that is difficult to recover. Each entry in a graveyard list is also an entry in the memory of a user who was asked to change their behaviour, integrate a tool into their life, and then found the tool withdrawn or quietly diminished. That scepticism compounds, and it eventually shows up in adoption curves for the next wave of products.
What to watch for next is whether the pace of entries in lists like TechCrunch's accelerates or plateaus. If it accelerates through the back half of this year, it will be a meaningful indicator that the correction phase of the cycle is deepening. Also worth watching is whether any of the major foundation model companies begin explicitly acquiring the talent and intellectual property of shuttered startups at scale — that kind of consolidation would signal that the industry is transitioning from its expansionary phase into something more structurally settled.
