For all the excitement artificial intelligence has generated, success is still eluding many companies. A recent PricewaterhouseCoopers LLP study found that just 20% of enterprises are achieving at least three-quarters of the revenue and efficiency gains AI promises. Gartner estimates that at least half of generative AI projects were abandoned last year, and other estimates have put the figure closer to 80%.

The industry has coined a term — “AI pilot trap” — to describe AI experiments that demonstrate isolated value but fail to evolve into drivers of enterprise-scale transformation. Organizations may successfully build a chatbot or automate a narrow workflow in one department but struggle to expand those successes across the broader business. One key reason is that organizations often underestimate the time and resource commitments needed to scale AI initiatives safely and reliably. Cockroach Labs Inc.’s “State of AI Infrastructure 2026” report found that 83% of leaders believe their data infrastructure will fail without major upgrades in the next 24 months.

AI infrastructure differs from conventional data processing platforms in significant ways. Whereas traditional enterprise systems are optimized for repetitively processing large volumes of business transactions, database queries and application workloads using CPUs, AI infrastructure requires massive parallel processing of unstructured data powered by graphics processing units.

AI workloads continuously move enormous data volumes between storage, memory and processors, making high-speed networking and memory bandwidth as important as compute power. AI clusters of hundreds or thousands of interconnected GPUs create special power, cooling and networking requirements that can overwhelm traditional data processing infrastructure.

This feature is part of SiliconANGLE Media’s exploration of how AI factories, private cloud AI and vetted partner ecosystems are helping enterprises move from isolated pilots to production-scale AI outcomes. (* Disclosure below.)

Introducing the AI factory

The concept of an “AI factory” has emerged to describe the infrastructure needed to meet the full range of AI project demands, including model training, inference, retrieval-augmented generation, agentic AI and production operations.

TheCUBE Research has called AI factories “data centers of the future,” describing them as an architectural shift that’s as significant as the internet and cloud computing. The entire infrastructure stack is shifting, researchers observed, from CPU-centric to GPU-centric computing, from application hosting to intelligence generation and from static workflows to agentic orchestration systems.

Hewlett Packard Enterprise Co.’s solution for these unique requirements is Unleash AI, a program that delivers production-ready enterprise AI on infrastructure that provides the necessary power, governance, security and scale. Unleash AI centers on a curated ecosystem of vetted ISV partners who integrate domain- and industry-specific solutions with HPE Private Cloud AI (PCAI) to turn AI potential into results.

Unleash AI solutions are much more than applications. They include a full package of pre-configured infrastructure, systems software and controls that together serve as an “easy button” for AI success.

Co-engineered AI

PCAI is a turnkey stack based on HPE ProLiant Compute Gen12 servers equipped with Nvidia Corp.’s accelerated computing stack to apply parallel processing to the unique needs of AI workloads. PCAI’s pre-integrated hardware and software support hybrid cloud deployment, a cloud-like operational model managed by the HPE GreenLake enterprise hybrid cloud platform, support for on-premises, edge and colocation deployments and multi-tenant enterprise security and governance controls.

Nvidia’s accelerated compute infrastructure delivers a full-stack solution combining GPUs for model training and inference, high-speed networking technologies, a suite of AI enterprise software, containerized microservices that accelerate the deployment of AI models and a catalog of pre-trained, customizable AI workflows.

Unleash AI delivers pre-vetted ISV applications on top of that foundation. Control and auditability features comply with regulations such as the European Union’s AI Act and the General Data Protection Regulation. Enterprise-class security features such as data isolation, identity and access controls, sovereign AI deployment models, compliance tooling, AI model governance controls, runtime protection and auditability and remediation capabilities guard against data exfiltration and unintended disclosure of sensitive information.

Unleash AI partners leverage this underlying infrastructure to add value and simplify deployment. For example, Kamiwaza Corp. worked with SHI Corp., HPE and Nvidia to create an AI agent that automates compliance with Section 508, a law that requires federal agencies to ensure their information technology assets are accessible to people with disabilities.

Many government agencies have websites spanning thousands of pages, dating back years. Scanning and fixing such a massive corpus of information manually is an overwhelming task. Kamiwaza uses a large language model to understand a website’s content and a visual model to assess visual accessibility. Agents can find and fix noncompliant content automatically.

The company estimates it can condense four to six person-years of manual labor into a single week of automated processing.

Kamiwaza uses HPE’s Sovereign AI features to process data where it resides, avoiding expensive and risky data migrations to the cloud. It uses a sophisticated inference mesh that leverages PCAI’s multi-location capabilities. When an inference request is made, the orchestration engine dynamically routes it to the HPE infrastructure stack closest to the required data. HPE’s integrated environment allows the software to direct an AI agent exactly where to move the inference task, so that information is processed locally using the enterprise’s existing security protocols.

PCAI’s turnkey architecture simplifies lifecycle management of the AI stack. This consistent foundation allows Kamiwaza’s solution to scale to hundreds or thousands of agents on the same underlying hardware, according to Luke Norris, the firm’s co-founder and chief executive officer.

“It’s a very tight solution that has a very tight [return on investment] that’s bundled nearly like an appliance that can be installed and turned on and get immediate ROI,” he said in an interview on theCUBE, SiliconANGLE Media’s livestreaming studio.

The integrated platform is also extensible. Once established, organizations can add new agents for use cases such as housing management or fire detection without needing to replace or purchase entirely new architecture.

AI at the edge

Another Unleash AI partner, ProHawk Technology Group Inc., uses HPE PCAI as a high-performance, secure foundation for its software that transforms degraded or obscured video feeds into clear imagery for use in image recognition, security, automated inspection and vehicle navigation. The technology performs roughly 33 million pixel optimizations every second, a massive computational task.

ProHawk uses Nvidia’s RTX 6000 PRO GPUs within the HPE PCAI stack to process sensor data as soon as it’s captured, without the latency penalties of moving it to the cloud. The software leverages PCAI’s ability to co-locate with existing sensors to deliver crystal-clear images to vehicle and maritime operators in near real time.

HPE Private Cloud AI’s Sovereign capabilities keep data and models behind an organization’s firewall for security and compliance while providing an audit trail that meets strict standards for forensic and legal admissibility. By fully integrating with HPE infrastructure, ProHawk can deliver a complete pixel-by-pixel restoration solution without the complex manual setup typically required for high-end computer vision.

ROI mandate

The window for AI experimentation is rapidly closing. Business leaders are now demanding results. Recent research by SAP SE’s Concur division found that half of finance leaders say evaluating ROI is their biggest AI challenge. AI champions need to focus on delivering measurable and sustainable results.

Unleash AI demonstrates how innovative technology vendors are helping them succeed by leveraging a vetted partner ecosystem and AI-ready infrastructure to deliver results out of the box and directly to the bottom line.

(* Disclosure: TheCUBE is a paid media partner for HPE’s “Unleash AI Momentum” interview series. Neither HPE, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Image: theCUBE Research/Gemini