Asus officially launched its end-to-end AI factory platform at Asus AI Tech 2026 in Seoul.
Asus mounts 227 kW artificial intelligence push
Turnkey factory setup brings pre-build digital testing but leaves deliveries unconfirmed
In a nutshell
Asus has expanded from manufacturing server components to delivering complete computing factories, bundling liquid-cooled processors with pre-installation software simulations. The platform pairs high-density computing racks with digital modeling tools from Schneider Electric, AVEVA, and IBM to stress-test power grids and cooling systems virtually. Asus asserts the platform cuts operational risks, but with prices, delivery timetables, and paying customers still unannounced, the rollout remains an unproven blueprint until physical shipments arrive.
Highlights
- Asus flagship AI POD combines 36 Vera processors and 72 Rubin graphics chips into a liquid-cooled rack.
- The equipment cabinet operates at high electrical densities reaching up to 227 kilowatts.
- Asus claims ten times the inference performance-per-watt compared to previous hardware generations.
- Simulation software from Schneider Electric, AVEVA, and IBM models facility heat and power digitally.
- Enterprise prices, regional delivery schedules, and initial customer contracts remain unannounced.
From the Editor’s Diary
Hardware performance gains now generate so much heat and consume so much electricity that selling processors requires selling the power and cooling infrastructure to survive them.
Who's involved
Asus
Taiwanese multinational hardware maker and server infrastructure provider
goal → Move from standalone computer boards into full-service corporate computing plant rollouts
Nvidia
American designer of accelerated graphics chips and processors
goal → Supply advanced Vera Rubin computer racks and the Omniverse simulation blueprints that run them
Paul Ju
Asus Senior Vice President and Head of Infrastructure Solution Business Group
goal → Establish Asus as a single-source contractor for processing power, cooling systems, and operational oversight
Schneider Electric
French multinational specialist in digital power management and plant automation
goal → Provide digital simulations that map out heat and electrical currents for ultra-dense server racks
In short
TL;DR: Asus has rolled out a unified computing package to build complete artificial intelligence factories, combining liquid-cooled hardware with software simulations before physical assembly. The systems promise heavy efficiency gains, but lack confirmed shipping dates, pricing, or buyers.
Q: What did Asus release, and what makes its rollout uncertain?
- Unveiled a complete system combining supercomputing racks, facility simulation tools, and runtime management.
How it unfolded
Asus previews liquid-cooled Vera Rubin system
Asus signaled its ambitions early in 2026 by testing customer appetites for extreme rack densities, aiming to address the mounting heat generated by advanced processing hardware. The company demonstrated a liquid-cooled architecture built around Nvidia Vera Rubin NVL72 setups, designed to absorb up to 227 kilowatts of electrical heat per equipment cabinet. That preview established the physical boundary limits of the machinery, though it stopped short of showing how warehouse operators could manage such loads at scale.
Full platform rollout addresses power risk
Six months after unveiling the bare hardware, Asus expanded the concept into an operational system designed to remove installation friction for facility operators. At its Seoul conference, the manufacturer linked its physical racks to digital twin blueprints—computer simulations that mimic real-world physics inside Nvidia Omniverse graphics software. Partnering with Schneider Electric, AVEVA, and IBM, Asus introduced pre-build testing for electrical, heat, and network stresses so operators can spot structural bottlenecks before buying equipment. The integrated offering turns hardware into a turnkey product, but its commercial credibility faces immediate friction: Asus withheld delivery dates, prices, and launch customers, leaving production timelines tied entirely to upcoming component shipments from Nvidia.
Where things stand
Official specifications confirm that the platform runs on 36 Vera central processors and 72 Rubin graphics processors per rack, backed by digital twin modeling software. Commercial viability remains unverified because Asus has not disclosed retail costs, shipping windows, or purchase orders. Delivery will depend on component supply from Nvidia, which determines when production data centers can plug in the hardware.