/ Blog

Why Do Most AI Projects Get Shelved Within 6 Months?

AI & Automation 8 min read · · Solid Team

According to research DDN commissioned with Google Cloud and Cognizant, two-thirds of 600 IT and business decision-makers at U.S. enterprises with more than 1,000 employees say their AI environments have become too complex to manage. The result: more than half of the AI projects launched in the past two years have been postponed or canceled outright.

This picture isn't unique to one study. MIT's widely cited Project NANDA finds that 95 percent of organizations see no measurable return from their generative AI investments. Gartner forecasts that more than 40 percent of agentic AI projects will be canceled by the end of 2027. Forrester reports that 25 percent of planned AI spending has been pushed to 2027, and that only 15 percent of decision-makers report an EBITDA lift at their organization.

The problem isn't picking the technology. Most AI tools on the market today genuinely work. The problem is that infrastructure, data readiness and the right use case aren't keeping up with the pace.

Why Do Enterprises Get Stuck?

DDN CEO Alex Bouzari says there's a strong appetite inside enterprises to roll out AI, but the infrastructure, power and operational foundation to run it often aren't ready. The result: projects slip, GPUs run underutilized, energy costs climb, and for a lot of organizations the math just doesn't work.

This isn't just a technical detail — it becomes a straightforwardly financial problem: IT projects blow through budget, and return on investment becomes impossible to calculate.

Moving to the Cloud Isn't a Fix by Itself

97 percent of respondents believe scaling AI needs to happen in the cloud. But Bouzari is cautious here: the challenges enterprises face internally just move to the cloud with them — the cloud needs unified data and orchestration at scale too. In his view, what's really missing is an education and maturation process that has to happen inside the IT organization.

Where the Breakdown Happens

  • Complex, fragmented infrastructure. When data, compute and orchestration layers operate disconnected from each other, AI projects get bogged down in infrastructure problems before they ever reach production.
  • Low-quality or hard-to-reach data. No matter how good the model is, scattered and inconsistent data delays a project before it even starts.
  • Unclear business value. Most projects that start from a "let's use AI" motivation move forward without clearly defining which concrete problem they're solving.
  • Weak risk controls and rising cost. Risks and costs surface as the project progresses; without planning for them upfront, budget and timeline balloon unexpectedly.
  • Shallow use cases. Squeezing AI into a narrow goal like "cut customer service cost from 3.7 percent to 3.1 percent" leaves most of its potential on the table.

What Successful Organizations Have in Common

According to Bouzari, the gap that keeps widening is between organizations that invest in AI early and decisively — turning pilots into products that actually generate revenue — and those just getting started. Two things close that gap:

An internal learning process. Working with partners who have real experience deploying complex, enterprise-scale solutions can speed up that maturation process.

Depth in use-case selection. Instead of shallow goals focused only on cost reduction, finding use cases that genuinely combine AI with the organization's own data to produce measurable business value.

Evaluate Your Own Project

  • Can our infrastructure — data, compute, orchestration — actually carry AI at this scale?
  • Is the data we're using accessible, consistent and clean enough?
  • Is the use case we picked producing real business value, or is it just the "easy and shallow" choice?
  • Who's tracking risk and cost growth, and how often?
  • Is there clear ownership that will keep this project running 12 months from now?

In Short

  • More than half of the AI projects launched in the past two years were postponed or canceled due to infrastructure complexity
  • Independent research from MIT, Gartner and Forrester all point to the same picture
  • The cloud alone isn't the fix — unified data and orchestration are needed in every environment
  • Shallow use cases (cost-cutting only) limit AI's potential
  • The organizations pulling ahead invest in both infrastructure maturity and deep use-case selection together

Source: The data used in this article draws on The Register's "Over half of enterprise AI stalls on infrastructure mess" (based on the DDN / Google Cloud / Cognizant 2026 AI Infrastructure Report), and on the MIT Project NANDA, Gartner and Forrester research it cites.

If you'd like to talk about this in the context of your own project, Let's Talk About Your Project /