Over half of enterprise AI projects shelved as infrastructure complexity stalls adoption

Over half of enterprise AI stalls on infrastructure mess

The answer seems to be educating the enterprise workforce, and creating smarter use cases More than half of AI projects have been delayed or canceled within the last two years citing complexities with AI infrastructure, according to a research report commissioned by DDN, a data optimization company in partnership with Google Cloud and Cognizant. About two-thirds of the 600 IT and business decision-makers surveyed at US enterprises with 1,000 or more employees said their AI environments are too complex to manage. "If you look at the enterprise, there's just enormous enthusiasm to deploy AI, but the problem is that the infrastructure, the power, and the operational foundation that is required to run it just aren't there," Alex Bouzari, CEO of DDN, told The Register. "And so as a result, it pops up in the financial elements with IT projects getting delayed, the GPUs being underutilized, power costs going up. And so the economics, I think, for lots of organizations don't pencil out because of these challenges." This isn't the first study that has found AI projects coming up short in the enterprise. MIT's widely cited Project NANDA found 95 percent of organizations are seeing zero measurable return from their generative AI investments. Gartner predicted that more than 40 percent of agentic AI projects will be canceled by the end of 2027. Forrester found that 25 percent of planned AI spend would be delayed into 2027, as only 15 percent of AI decision-makers reported an EBITDA lift for their organization. While some 97 percent of the decision-makers surveyed in the DDN study believe that scaling AI for their organization will need to happen in the cloud, Bouzari isn't so sure that is much of a panacea for the infrastructure dilemma. "The same challenges that you would have on prem will follow you into the cloud," he said. "I mean, cloud needs unified data, and the cloud needs orchestration at scale. So, it's all of these considerations. There's an education process which needs to take place within the IT organization." Founded in 1998, DDN works with some of the biggest names in the AI race including Nvidia, xAI, and Google, to optimize the flow of data into and out of AI infrastructure, a capability that has taken on heightened relevance as the cost and power used by those systems has grown. Bouzari said there is a widening gulf between the early movers in AI that made big bets and have turned pilot projects into salable products that generate ROI, and many enterprises who are just starting in AI today. Complicated infrastructure appears to be one of the significant roadblocks that are stopping adoption. "I think that the education process is something that the facilitators can enable. I mean, if you look at organizations like Accenture and Deloitte, resellers who know how to deploy complex, turnkey business solutions for organizations, I think there's a ramp in that curve, which is starting to take place, and then we will have an accelerated adoption." Rather than defaulting to customer service chatbots when the topic of use cases comes up, vendors and advisors need to help find capabilities that bridge an organization's data with AI. "As opposed to, I'm going to lower my customer service cost from 3.7% of revenue to 3.1% of revenue," Bouzari said. "That is really short changing what AI can do." ®

Businesses are struggling to achieve AI success - this report reveals why

Enterprises are also under pressure from a sustainability angle Artificial intelligence isn't proving to be the golden key for many businesses, with two in three (65%) admitting their AI environments are too complex to manage and more than half (54%) have cancelled AI projects over the past two years due to infrastructure issues. And infrastructure, according to DDN's latest State of AI Infrastructure Report, is exactly what's holding businesses back, quickly followed by energy. Looking ahead, 97% agree cloud is essential to scaling AI initiatives, with hybrid AI workloads expected to grow 162% over the next 12 months. DDN's report reveals the considerable role that third parties play, with 72% relying on external expertise and only 12% depending solely on in-house teams. This is notable, because 83% agree that teams are struggling today and 98% admit to AI skills gaps, further underscoring the need for outside help. The study also found that most failures can be traced back to silos, either in storage, compute or data pipelines. "Enterprises are discovering that scaling AI isn't a compute problem - it's an integration problem," DDN CTO Sven Oehme wrote. "If your infrastructure isn't unified, your AI can't learn efficiently." Other common reasons for failure include legacy technologies, poor cloud strategies, and the complexity of stacking tools instead of simplifying them. "Without modern, unified infrastructure, AI can't scale," DDN CEO Alex Bouzari said, slating companies for chasing models and GPUs instead of focusing on "the data layer underneath." All of this against a backdrop of increased pressure from stakeholders and regulators. Most (93%) are now actively trying to reduce AI energy impact, with around half (47%) citing power and cooling as the top inefficiencies. "Tokens per watt" is therefore emerging as a new performance metric for AI efficiency.