IBM is pushing artificial intelligence processing away from distant cloud data centers and toward the edge, where companies can run AI models directly on their own servers. The company announced the Power S1112, a compact single-socket server designed specifically for on-premises AI inference, alongside new automation tools that let enterprises manage increasingly complex AI workloads without constant human oversight.

The timing reflects a broader industry shift. As businesses deploy more autonomous AI agents, the costs and complexity of relying entirely on cloud-based models are becoming harder to justify. IBM's Institute for Business Value projects that enterprises will deploy an average of 1,661 AI agents by 2027, a 38% increase from current levels, leaving IT teams managing hundreds of thousands of autonomous decisions daily. That scale demands infrastructure that can handle routine tasks locally, without constantly reaching out to cloud providers.

What Makes Local AI Inference Attractive to Enterprises?

Running AI models on-premises solves several practical problems that cloud-dependent systems create. When autonomous agents make repeated API calls to cloud services, the costs can spiral unpredictably. A business using cloud-based AI agents for heavy workloads might spend $6,000 to $8,000 per year on API bills alone, compared to roughly $384 annually in electricity costs for a local edge system handling the same work. Beyond cost, there's the security angle: every API call to an external cloud service carries the risk of exposing sensitive corporate data, customer information, or proprietary code.

The financial case for edge inference becomes clearer when you look at the long-term payback. Cloud AI remains an ongoing operational expense, month after month. Local hardware, by contrast, becomes a permanent asset that pays for itself over time. Once the initial investment is recovered, the business benefits from substantially lower operating costs while retaining full ownership of its AI infrastructure.

How to Deploy Agentic AI Across Your Organization?

Enterprises typically choose from three deployment models, each suited to different operational needs and risk tolerances:

• Hosted Model: The AI agent runs on an edge device but depends entirely on cloud-based language models via APIs for the actual workload. This approach offers access to the most advanced frontier models but features less predictable costs and ongoing cloud dependencies.
• Hybrid Model: Expected to dominate enterprise deployments, this model splits work between local on-device models and cloud APIs. Routine tasks are handled locally at no additional cost, while highly complex reasoning tasks are routed to the cloud, balancing cost predictability with cutting-edge capabilities.
• Fully Local Model: Both the AI agent and the language model reside entirely on-premises. This approach requires optimized edge compute hardware with high memory bandwidth and offers the tightest controls over operational costs and data privacy.

Real-world results show the potential. Wyndham Hotels and Resorts, which operates over 9,300 franchises globally, deployed agentic AI to automate routine workflows. The company reduced the time to update global brand standards from 30 days to just 1.5 days, a 94% time reduction. AI agents now handle 28% of incoming customer calls completely autonomously, and average call handle times have dropped by 30% to 50%, freeing human staff to focus on high-value guest interactions.

What Technical Advantages Does the Power S1112 Offer?

IBM's new server is built around the Power11 processor, which includes specialized hardware called Matrix Math Acceleration designed specifically for AI inference. The company claims the S1112 delivers twice the per-core performance of its older Power S914 model and three times that of the Power S814, based on published benchmarks for 4-core configurations. In terms of energy efficiency, IBM reports up to 69% better efficiency than the S914 in a smaller physical footprint. A 10-core S1112 configuration delivers 539 compute performance watts per watt of power consumed, compared to 319 for an 8-core S914.

The server supports up to 512GB of DDR5 memory and can run IBM AIX or IBM i operating systems. It fits in a standard 2U rack form factor, making it suitable for deployment in existing data centers or edge locations outside the main data center. IBM is also introducing Power Expert Care Premium Essentials, a support tier exclusive to the S1112 that includes priority access to IBM specialists and faster response times.

Alongside the hardware, IBM announced Power Autonomous Operations, a control plane for automating day-to-day infrastructure management. In internal testing across an eleven-system Power environment, the platform resolved a capacity-related issue in 3.33 minutes, compared to 52.59 minutes using traditional manual workflows, a roughly 15-fold reduction in resolution time. The platform includes an embedded AI agent that supports natural-language interaction, reducing the need for deep domain expertise to manage and tune Power environments.

"The goal is to let enterprises pursue rapid AI deployment without trading off system stability, positioning increased automation in Power as a way to handle routine availability, optimization, and security tasks while preserving control and resilience," said Hillery Hunter, General Manager for IBM Power and CTO at IBM Infrastructure.

Hillery Hunter, General Manager for IBM Power and CTO at IBM Infrastructure

The Power S1112 is expected to be generally available on July 24, 2026. Power Autonomous Operations is scheduled for general availability on September 23, 2026. For organizations already managing complex IBM i applications, IBM also released the IBM Bob Premium Package for i, an AI-driven development assistant that helps engineers understand existing codebases and onboard new developers more quickly. Early adopters report that new developers understand complex IBM i applications 60% faster with the tool.

The shift toward edge inference reflects a maturing understanding of where AI workloads belong. Not every task requires the power of a frontier cloud model, and not every business can afford the ongoing costs of cloud-dependent AI. By bringing inference capabilities on-premises, IBM is betting that enterprises will choose local control, predictable costs, and data security over the convenience of cloud APIs.