Beyond the hype: A strategic AI roadmap to tangible returns
For organizations investing in AI projects, realizing ROI can be rife with obstacles. These three tips can help leaders define and find success in their enterprise AI initiatives.
Intelligence and advice powered by decades of global expertise and comprehensive coverage of the tech markets.
Published: 18 Sep 2026
Is the shine starting to fade from AI?
According to research from Omdia, a division of Informa TechTarget, 60% of organizations said they are starting to see the initial hype around AI wane as teams realize how difficult it is to operationalize. Businesses are spending far too much money for leadership to not expect a significant payback, but navigating a path to achieving tangible financial benefits from AI is an effort fraught with constantly evolving complexities. It is time for overinflated expectations and aimless experimentation with AI to come to an end.
Common hurdles that organizations face on the path to AI success include the following:
Multiple influencers and business units promoting their own use cases.
Trapped or poor-quality data.
Ever-changing landscapes of models, infrastructure options and application platforms.
With four out of five organizations allocating significant levels of investment to AI initiatives, there is increased visibility into AI projects and inherent opportunities and risks for the decision-makers. As pressures on AI infrastructure teams rise, a convoluted landscape of AI tools and infrastructure options adds to the difficulty of achieving positive financial returns -- even if the organization gets multiple projects to production.
3 tips to improve your success in AI
Put a hyperfocus on the high-value use case. Instead of racing to get a portfolio of AI projects out the door, prioritize one well-defined, high-value use case. Define success for that project upfront and track quantifiable metrics. The implementation approach that follows -- including the data, the infrastructure, the platform and the tools -- should then be optimized around that use case.
The alternative to this approach would be either to evaluate the implementation options divorced from the use case or to begin the planning phase with multiple potential use cases in mind. While those approaches can lead to capable production AI systems, they decrease the likelihood of achieving an optimized design and finding tangible financial success compared with concentrating on one high-priority AI use case.
Accept the inevitable distributed AI ecosystem. The potential downside of the use case-centric approach to AI implementation is that once you move to another use case, the "right" implementation for your second or third use cases -- or sets of use cases -- might require a completely different approach. That approach might involve using public cloud, colocation, edge vs. a data center deployment, different accelerators, additional vendors, etc.
Given the diversity of AI use cases, a distributed approach to building an AI portfolio was always the likely outcome. In addition, the constant and rapid evolution of offerings in the AI space necessitates that you should avoid locking your organization into any one specific implementation design. For example, in the past few weeks, Equinix announced the Equinix Inference Exchange, a partnership with Together AI and Nvidia to provide users with a turnkey deployment of Together AI's inference platform. It supports more than 200 open source models on validated Nvidia infrastructure within Equinix facilities, mitigating concerns over power, cooling and space.
The same week, at VMware Explore, Broadcom announced the VMware AI Factory, which includes VMware Cloud Foundation (VCF) with VCF Private AI Services. It helps automate provisioning and lifecycle management on validated hardware options from players such as Dell, Cisco, Lenovo and Supermicro. The VMware AI Factory should simplify integration into existing VMware environments, making it easier to evaluate and deploy.
These announcements, however, join an already crowded landscape for AI that includes the major public clouds -- AWS, Google and Microsoft -- the neoclouds and a wealth of technology vendors, such as Cisco, Dell, HPE, Lenovo, Nutanix, Nvidia and Red Hat.
62% of organizations agreed that managing the infrastructure for AI is currently more difficult than building the AI models themselves.
Pressure test vendors and products. As the list of AI tools and vendors continues to grow, it is imperative to stay educated on the landscape and recognize that no single vendor offers everything you need for your AI project to be successful. To this end, push the design responsibility back on your vendors. Many tout extensive expertise in building successful AI implementations. Have them prove it. Do not be satisfied with hearing from potential vendors about why their widget is superior to the competition's. Prioritize vendors that can bring in the right technology partners and have an end-to-end conversation on how to optimize your environment for your AI use case.
The diverse and complex nuances of optimizing AI infrastructure design create complexity, add cost and slow down projects. According to the previously cited Omdia research, 62% of organizations agreed that managing the infrastructure for AI is currently more difficult than building the AI models themselves. Unfortunately, there is no perfect answer for every use case, but there is likely a perfect answer for your use case. Success in AI depends on finding that answer.
Scott Sinclair is Practice Director with Omdia covering the storage industry.
Omdia is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.
