Lenovo expands hybrid AI portfolio to reduce enterprise inference costs

Lenovo has expanded its Hybrid AI Advantage portfolio with new AI inference platforms and agentic AI tools designed to help enterprises run AI workloads across devices, workstations, data centres and cloud environments.

The update is aimed at organisations moving AI from trials into production, where the ongoing cost of running AI models and agents can become harder to manage. Lenovo said the new portfolio is intended to help enterprises deploy AI agents faster, reduce token costs and run workloads closer to where company data is created and used.

The company is positioning the expanded portfolio around hybrid AI infrastructure, where AI workloads can be placed across AI PCs, devices, workstations, data centres or cloud environments depending on performance, cost, security and governance needs.

New platforms focus on private AI deployment

The expanded portfolio includes new inference-optimised platforms developed with NVIDIA, Intel, Red Hat and Canonical. Lenovo is targeting enterprises that want to run private AI workloads while maintaining control over data location, governance and operating costs.

One addition is a CPU-only Lenovo Hybrid AI Platform with Red Hat. The platform is built on Red Hat AI Enterprise and powered by Intel Xeon 6 processors with integrated AI acceleration. Lenovo said it is designed to process about twice as many AI requests at the same time, while improving throughput, latency and time-to-first-token.

The platform is intended for enterprise AI workloads such as retrieval-augmented generation, HR support and customer service assistance. By combining Lenovo servers, storage and Red Hat AI Enterprise in a validated architecture, the company said organisations can scale private AI across on-premise and hybrid cloud environments while maintaining security, governance and operational control.

Lenovo is also introducing Lenovo Hybrid AI Platform 221 in two configurations. The Canonical version uses Canonical Ubuntu and Canonical Kubernetes architectures for building, testing and deploying private AI applications, copilots and personalised enterprise tools. The Red Hat AI Enterprise version is designed for organisations moving AI into governed production environments, with lifecycle management and safeguards across deployment environments.

Lenovo said these systems can be ready in as little as a few weeks.

Cost control becomes a key part of the pitch

Lenovo is using AI running costs as a central part of the update. The company said its inference-optimised architecture can deliver up to 8X lower cost per token than cloud-based infrastructure services for workloads requiring sustained CPU and GPU use. It also claimed up to 18X lower cost per million tokens compared with model-as-a-service APIs.

The focus on cost comes as enterprises assess how much AI usage will cost once deployments move into regular operation. Lenovo cited industry research stating that 92% of organisations deploying agentic AI report costs exceeding expectations.

For enterprise buyers, the practical issue is whether hybrid infrastructure can make AI costs more predictable without adding management complexity across devices, private infrastructure and cloud environments.

Agentic AI tools target enterprise workflows

Lenovo is also adding agentic AI capabilities for autonomous and long-running agents. These tools are designed to support AI deployments that run from desktops and workstations to data centre infrastructure.

The company highlighted Lenovo AI Library solutions, including Knowledge Super Agent use cases, which allow employees to search and synthesise information from multiple enterprise systems through a single AI-powered interface. Lenovo said independent analysis has shown these deployments can save thousands of employee hours across an organisation.

Lenovo is also developing NVIDIA NemoClaw skills for AI operations. These capabilities are intended to help IT teams detect issues earlier, automate troubleshooting and respond to technology problems more quickly.

The company said it is co-developing autonomous AI agents, skills and solutions for NVIDIA NemoClaw with customers globally. It is also introducing personal AI Factory environments on Lenovo ThinkStation PGX, which will give developers a local environment for agentic AI workloads, including NVIDIA NemoClaw blueprints.

Lenovo described ThinkStation PGX as both an entry point and endpoint for local AI execution, with a path from proof of concept to larger deployments through ThinkStation PGX and ThinkStation PX systems. Initial customer engagement will focus on limited-access co-development programmes.

The company also outlined a planned retail use case in which an AI-powered kiosk would act as a digital associate. The kiosk is intended to help customers find products, check inventory, discover promotions and receive personalised assistance, while reducing employee workload in stores.

Governance remains part of the infrastructure update

Lenovo is pairing the AI deployment update with infrastructure and management capabilities aimed at data privacy, governance and security.

The additions include Nutanix Compute only Cluster on Lenovo ThinkSystem Servers, a CPU-only virtualisation platform designed to reduce cost and complexity while preserving performance and reliability. Lenovo also highlighted XClarity One, which provides zero-trust management, visibility, control and automation across hybrid infrastructure.

The company said supply chain and hardware root-of-trust protections are part of its approach to helping organisations maintain compliance and governance requirements as AI moves from experimentation into more autonomous enterprise use.