AI is everywhere—but how much of what you hear is real, and how much is just hype? In this post, we cut through the buzzwords to explore what AI can actually do today. Get ready for a grounded look at the promises, pitfalls, and possibilities of artificial intelligence.

🧠 AI POC & Project Failure Rates

  • 42% of companies are abandoning most AI initiatives in 2025, up sharply from 17% in 2024, per S\&P Global Market Intelligence. On average, 46% of AI POCs are dropped before production ([zebra.com](https://www.zebra.com/ap/en/blog/posts/2023/what-a-mobile-device-will-really-cost-your-organization.html)).
  • 88% of AI pilots never reach production—only about 1-in-8 are deployed, according to IDC/LENOVO .
  • 70–90% POC failure rates are reported by Gartner and others for custom-built AI, vendor-built AI POCs struggle too .
  • Gartner predicts ≥30% of GenAI projects will be abandoned after PoC by end of 2025 . ➡️ Bottom line: AI POCs regularly fail at rates between 40–90%, depending on definitions (abandoned mid-project vs never entering production).

🌐 Previous Waves: Internet, Mobile, Cloud

Internet Era & Dot‑com Boom

  • The dot‑com crash (2000–2002) wiped out a majority of internet startups. Some estimates suggest 70–80% of dot‑coms failed, but survivors laid the groundwork for modern tech ([en.wikipedia.org](https://en.wikipedia.org/wiki/Dot-com_bubble)).
  • Classic projects like CRMs had failure rates as high as 70% in early 2000s, later dropping to ~40% by 2003 .

Mobile & IoT

  • IoT suffered 70–75% of deployments stuck at pilot stage, unable to scale ([en.wikipedia.org](https://en.wikipedia.org/wiki/Internet_of_things)).
  • Mobile hardware saw nearly 50% device failure by year 4 (cracked screens, etc.) .
  • Many early mobile tech launches (e.g., Google Glass, Microsoft Kin) were commercial failures—though exact PoC failure rates are hard to find .

Cloud

📊 Comparison Summary

Wave Reported Failure / Abandonment
Early Internet ~70–80% of startups failed; CRM POCs ~70% → ~40% over time
Mobile/IoT 70–75% IoT stuck in pilot; mobile hardware ~50% fail by Y4
Cloud High implementation issues (cost, performance), but less clear % for POC failures
AI/GenAI 40–90% POCs don’t go live; ~42% AI initiatives scrapped; 30% GenAI abandoned mid-2025

✅ Insights & Takeaways

  • Every major tech wave begins with high experimentation, broad hype, and high failure/discard rates—often 40–80%.
  • AI’s 40–90% POC failure is normal, especially given its complexity: messy data, immature infrastructure, unclear outcomes.
  • As with internet/mobile/cloud, failure rates tend to decline over time as best practices, tooling, and maturity increase.
  • Failures are part of innovation, not a sign of collapse—learning from them builds future success.

⚙️ Why so many AI POCs fail

  1. Unclear value proposition or objectives
  2. Poor data readiness (quality, integration, privacy issues)
  3. Wall‑street pressure & hype—leading to rushed or under-resourced trials ([en.wikipedia.org](https://en.wikipedia.org/wiki/Internet_of_things))
  4. Lack of measurement tools for tracking success
  5. Compounding error rates in multi-step AI flows (20% error per step leads to 63% total failure by 100 steps) ([en.wikipedia.org](https://en.wikipedia.org/wiki/ECRM))

✔️ Bottom Line

  • Yes, AI POC failures (~40–90%) are currently high—but not unexpected, given similar disruption in past tech revolutions.
  • Over time, failure declines as understanding, standards, data quality, tooling, and domain readiness improve—just as it did with internet, mobile, and cloud.

The tech industry often experiences waves of hype cycles. Each of these domains you mentioned—mobile apps, blockchain, IoT, data science, and generative AI—came with legitimate potential, but the market often overestimates short-term returns while underestimating the long-term impact.

Here’s why that happens:

  1. Speculative Investment Culture: Venture capital tends to chase trends. This inflates expectations and fuels the creation of countless startups, many of which lack solid business models or meaningful differentiation.

  2. Misaligned Incentives: Media, influencers, and tech evangelists often amplify the hype to drive engagement, investments, or personal branding, regardless of maturity or scalability.

  3. Corporate FOMO: Enterprises jump on trends to avoid appearing outdated, sometimes without a clear plan for integration or sustainability.

  4. Skill Shortages and Misapplication: The talent gap can lead to poorly implemented solutions. For example, applying blockchain where a database would do or using generative AI without understanding its limitations.

Despite this, over time, the core technologies often settle into valuable niches. Think of how mobile matured into a staple or how data science is now embedded in business intelligence rather than standalone hype.

The key difference with generative AI is that it’s infiltrating productivity tools, consumer applications, and software development itself—giving it broader and potentially more enduring reach. But even here, we’re seeing inflated RoI expectations and underwhelming deployments in many enterprises.

So yes, real value exists—but it’s often drowned in noise.