Lenovo, Nvidia launch AI cloud gigafactory

Lenovo and Nvidia have unveiled a new program aimed at building gigawatt-scale AI cloud infrastructure, announced at CES 2026 in Las Vegas.

The initiative, called Lenovo AI Cloud Gigafactory with Nvidia, targets enterprise and cloud providers seeking to deploy large-scale AI workloads more efficiently.

Lenovo said the program will use Nvidia’s advanced computing platforms and its own liquid-cooling technology to help providers set up AI factories capable of supporting complex applications.

The platform includes access to Nvidia’s Blackwell Ultra architecture, and the newly announced Vera Rubin NVL72 system.

The program is designed to cut deployment time for AI infrastructure, and features integrated solutions spanning hardware, networking, and support services.

Lenovo claims its AI cloud solutions serve eight of the world’s top 10 public cloud providers, and that it can design, manufacture, and integrate custom AI cloud systems in-house.

🔗 Source: Lenovo

🧠 Food for thought

Implications, context, and why it matters.

Lenovo AI factory plan lacks firm buyers, clear dates

Lenovo announced the AI Cloud Gigafactory with no firm customer commitments, pilots, or delivery schedules for the NVIDIA-based systems though it powers eight of the top 10 public cloud providers 1.
NVIDIA’s Vera Rubin NVL72 has entered production 2, but broader availability is not expected until the second half of 2026 3. Blackwell Ultra systems in the program have no stated ship window, so near-term revenue looks uncertain 1.
The announcement omits prices, facility specs or power needs so total cost of ownership cannot be assessed 1. Morgan Stanley pegs liquid-cooling parts for a GB300 (Blackwell Ultra) NVL72 rack at about $49,860 4, which is one slice of spend for gigawatt-scale rollout.

Build liquid-cooling capacity for Rubin systems

Data center operators plus colocation providers (data centers that lease space, power, cooling to customers) should ready sites for direct liquid cooling (coolant delivered via cold plates directly to chips) 2 5. The Vera Rubin NVL72 needs warm-water single-phase cooling with a tubeless, cableless design.
Utilities plus power infrastructure firms can stand out with rack-level power smoothing and site-level battery energy storage solutions, since Rubin’s variable AI workloads push power swings that strain utilization 5.
Cooling vendors should design gear for the 3,600W thermal design power (the heat a chip is designed to dissipate) of future Rubin Ultra GPUs 4. Standard liquid cooling may fall short, so immersion cooling (submerging hardware in non-conductive fluid) or embedded cooling could handle these loads.