Lenovo unveiled a suite of new enterprise servers specifically designed to handle AI inferencing workloads.

Showcased at CES 2026 in Las Vegas, the ThinkSystem and ThinkEdge servers cover an array of sizes, each catering to specific enterprise inference needs. The ThinkSystem SR675i V3, for example, is the powerhouse unit, designed to power some of the largest workloads with what the Chinese firm described as “massive scalability.”

“Enterprises today need AI that can turn massive amounts of data into insight the moment it’s created,” Ashley Gorakhpurwalla, EVP and president of Lenovo’s Infrastructure Solutions Group, explained. “With Lenovo’s new inferencing-optimized infrastructure, we are giving customers that real-time advantage, transforming [a] massive amount of data into instant, actionable intelligence that fuels stronger decisions, greater security, and faster innovation.”

Unlike AI training, wherein a model is steadily built using vast datasets across an often inordinate amount of time, inference takes an already complete model and uses it to output new information. Inferencing runs continuously, meaning enterprises can and are increasingly looking to shift away from costly training cycles and reiterations of their AI models and systems.

Lenovo’s CES showcase sought to capture that enterprise demand for inference-capable hardware with servers such as the ThinkSystem SR650i, an easy-to-deploy unit capable of offering high-density GPU compute.

Also unveiled at Las Vegas was the ThinkEdge SE455i, a more compact offering touted for use in environments like retail, telecom, and the industrial edge. Lenovo claims the SE455i “brings AI inferencing capabilities anywhere data is located” while ensuring rugged reliability.

The units are powered by AMD's EPYC processors and Nvidia accelerated computing hardware also under the hood, while Lenovo’s Neptune liquid cooling tech removes heat from the source.

In an attempt to sway enterprises looking to upgrade but conscious of costs, the new server range is available via Lenovo’s TruScale pay-as-you-go pricing model, which provides access to the on-premises hardware with no upfront costs but instead a variable monthly charges based on actual usage.

The firm said offering the servers via its TruScale model will enable businesses to “achieve peak performance and efficiency without compromising agility, security, or budget.”

Lenovo’s CES launch and its move to capitalize on the rise of inferencing follow its storage server update last December that sought to take advantage of the VMware uncertainty.

The vendor unveiled new all-flash storage and hyperconverged infrastructure offerings to help enterprises running AI applications at the other end of the stack so they can “extract maximum value” from their vast troves of data.