AEWIN's OCP 2026 filing lists a 1U two-phase direct liquid-cooled server, BG18-A10710, with one 600 W CPU and two single-width 600 W GPUs. The same release names a 2U model, BG28-A10710, with four PCIe 6.0 x16 double-width slots for 600 W GPUs, and a 4U model, BG48-A20813, with up to eight double-width GPU slots. The packet contains no figure, so the cooling loop, manifold, CDU, and rack layout are not shown.

The useful constraint is not the GPU count but the power budget per chassis. A 1U box that must carry a 600 W CPU plus two 600 W GPUs leaves little room for air cooling, so the 2P DLC claim is the load-bearing part of the story. The release says rack-scale systems can exceed 200 kW with reduced PUE, but it does not tie that number to a measured rack, a specific CDU, or a validated PUE. For a buyer, the document proves a product lineup and a cooling partnership; it does not prove that the 1U, 2U, and 4U units will hit those rack-level targets in a live data center.

What is still unmeasured is the actual PUE, the per-chip cooling capacity, and the rack power draw under mixed CPU and GPU load. The filing also omits GPU model names, CPU part numbers, and any test data for the 200 kW rack claim. Those gaps matter because a 600 W component is a thermal specification, not a proof of sustained operation.

The constraint is the 600 W component budget inside a 1U chassis, not the GPU count. The release proves a lineup and a cooling partnership, but it does not prove measured rack power or PUE.

The document leaves out measured PUE, per-chip cooling capacity, GPU model names, and rack power draw under load.

A 1U box with a 600 W CPU and two 600 W GPUs needs liquid cooling to stay within chassis limits. The 200 kW rack claim is useful, but it is not tied to a measured CDU, PUE, or sustained load.