From “card hoarding” to “efficiency is king”: Dell (DELL.US) “AI factory” expands, and the computing power business of new cloud vendors is about to change

The Zhitong Finance App noticed that Dell Technology (DELL.US) shares surged nearly 6% before the market on Wednesday. Earlier, the company announced the launch of a complete AI Factory (AI Factory) program with NVDA.US (NVDA.US), covering AI solutions and ecological innovation, new desktop intelligence solutions, and dedicated AI infrastructure.

Dell said that more than 6,500 companies have already adopted Dell AI factories in collaboration with Nvidia, up from 5,000 in the previous quarter.

New products include support for Reflection's Beam model, AI observability in collaboration with DynaTrace, and Mistral, FogSphere, and SUSE Linux.

Strategically, Dell is betting on “smart AI”: the agency predicts that intelligent AI will become the largest workload in 2028, with inference costs reduced by 90% and token consumption increasing 24 times by 2030. Dell uses “own infrastructure+end-to-end hosting” to seize this wave of computing power dividends and compete head-on with pure public cloud leasing routes.

The computing power business of new cloud vendors is about to change

Dell Technology and Nvidia are teaming up to expand the “AI factory” combo and are rewriting the competitive landscape of enterprise-level AI computing power — the ones that most directly feel the pressure are new cloud vendors that have started by “renting GPUs” in the past two years.

Dell's core narrative is “owning is more cost-effective than leasing”: the cost efficiency of own AI infrastructure is 90% higher than that of a public cloud within two years. The superposition of the DSX POD containerized AI factory reduces the deployment cycle from 6 months to 14 days, and the accounting logic of the enterprise's “self-built computing power” has been completely reshaped.

Previously, the core selling points of new cloud vendors were eliminating heavy assets, being flexible on demand, and increasing volume quickly; now, Dell uses factory prefabrication+end-to-end hosting to reduce the threshold and billing period for “self-construction”, which is tantamount to directly removing part of the rental demand that would otherwise flow to new cloud vendors.

But this is not a one-sided downside. The DSX POD is essentially NVIDIA's reference design. New cloud vendors such as CoreWeave (CRWV.US), Nebius (NBIS.US), Lambda, Crusoe, and IREN (IREN.US) are its early users — they can also use this modular solution to reduce their own expansion costs and shorten delivery cycles. In other words, the separation of “manufacturing” and “operation” of computing power is accelerating, and the core competitiveness of new cloud vendors is shifting from “how many GPUs are in their hands” to “operational efficiency, software stack depth, and customer acquisition capabilities.”

The real dividing point is software and ecology. Dell weaves an end-to-end web with open models, observability, knowledge assistants, and security governance. Small to medium cloud vendors that only rent bare metal computing power and lack upper software capabilities will face being squeezed in the middle — losing to hyperscale clouds at the top and losing to Dell's “full stack+own” at the bottom.

The industry logic has moved from “scarce computing power” to “efficiency is king”. When inference costs are expected to drop 90% by 2030 and token consumption will increase 24 times, what determines victory or loss is no longer about hoarding cards, but who can use a lower unit token cost to deliver ROI. Dell is betting on “enterprise-own+intelligent AI,” while new cloud vendors need to prove they are irreplaceable in terms of dedicated chips, elastic scale, and data gravity. A reshuffle around “who controls the last mile of enterprise AI implementation” has only just begun.