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

Relay, a five-year-old AI workflow-automation startup positioned against Zapier, shut down as OpenAI, Google and other large platforms embedded similar capabilities into their own products. S&P Global Market Intelligence estimates that roughly 42% of corporate AI initiatives are ultimately abandoned, reflecting funding constraints, technical hurdles, scaling difficulties, competition and weak demand. The article highlights a broad AI-product washout, including Humane’s AI Pin closure after raising $230 million and its asset sale to HP for $116 million, alongside shutdowns or retrenchments at Huxe, Yupp, Notion Mail and several OpenAI products. Privacy, security and execution remain material risks, illustrated by Microsoft Recall’s continued scrutiny and Apple’s $250 million Siri-marketing settlement.

Analysis

The investable implication is AI application-layer consolidation, not a broad read-through on AI demand. Standalone workflow, agent, and content-generation vendors face structurally higher customer-acquisition costs and weakening pricing power when model owners bundle equivalent functionality into existing productivity suites. This favors distribution-rich platforms—GOOG through Workspace/Gemini and MSFT through Microsoft 365/Copilot—while raising impairment and down-round risk across private AI SaaS portfolios over the next 6-18 months.

MSFT carries a nearer-term asymmetric risk: privacy-sensitive endpoint AI can turn an otherwise incremental product feature into enterprise procurement friction, security-review delays, and potentially regulatory scrutiny. The key question is whether security objections remain confined to opt-in consumer adoption or produce measurable Copilot-PC and Windows commercial attach-rate weakness over the next two earnings cycles; the latter would challenge premium assumptions around AI monetization. AAPL's risk is different: execution slippage reduces the differentiation of the installed-base upgrade story and leaves the company more dependent on hardware refresh mechanics rather than a higher-margin AI-services narrative.

Contrarian view: product closures are not necessarily bearish for hyperscalers. They may improve monetization by eliminating free or subsidized intermediaries that were absorbing inference costs without durable willingness to pay. The important negative second-order effect is on AI infrastructure expectations: if enterprise pilots continue to be abandoned rather than scaled, cloud GPU demand can remain strong while revenue conversion lags, creating a multiple-risk event for AI-exposed software and private-market marks before it becomes a capex-risk event for GOOG or MSFT.

Key Decisions for Investors

• Maintain a 3-6 month long GOOG / short basket of unprofitable AI application software or private-market proxy exposure: platform distribution should capture feature commoditization, but size modestly because GOOG's AI monetization disclosure remains limited. Reassess if Workspace net retention or cloud growth decelerates materially.
• Treat MSFT as a watch-to-hedge rather than an outright short: buy 3-6 month downside put spreads around earnings only if enterprise security incidents, regulatory inquiries, or Copilot-PC sell-through data show adoption friction. Falsification is sustained commercial Copilot seat growth and no incremental privacy-related remediation costs.
• Underweight AAPL versus GOOG over the next two quarters if AI functionality remains geographically or linguistically constrained; delayed feature parity weakens the case for AI-driven upgrade acceleration. Cover the relative short if iPhone replacement-cycle data or services growth reaccelerates despite limited AI availability.
• Avoid allocating fresh capital to standalone agent/workflow and AI-hardware venture exposures absent verified net revenue retention, gross-margin improvement after inference expense, and customer concentration data. Product demos and user counts are insufficient evidence of durable unit economics.