Starting with Vera Rubin OEM model disadvantage.
( not financial advice )
In this post
OEM means : DELL, HPE, SMCI, Cisco, Lenovo etc.
ODM means : Foxconn/Ingrasys, Wistron, QCT/Quanta, Inventec, Mitac, Pegatron etc.
========x========
With the Vera Rubin reference design,
- Contract ODM vendors are likely to catch up to OEM cooling and system design advantages.
>Traditional data centers heavily rely on air cooling, which consumes significant energy to move and condition air. Vera Rubin NVL72 systems instead use warm-water, single-phase direct liquid cooling (DLC) with a 45-degree Celsius supply temperature. Liquid cooling captures heat far more efficiently than air, enabling higher operating temperatures, reducing fan and chiller energy, and supporting dry-cooler operation with minimal water usage.
- Contract ODMs such as Wistron in Dallas–Fort Worth and Foxconn in Houston have also expanded U.S. manufacturing, reducing OEM's unique US based scale advantage.
- With the Vera Rubin, NVIDIA also planning to sell NVIDIA branded servers directly made by contract ODM.
- In my view, OEMs may already be near its peak achievable margin of around 10%, and breaking above that could be difficult as contract ODMs offer comparable systems at thinner margins closer to 4%.
- contrary to popular belief. OEMs do not have any software advantage over contract ODMs. For example : Every vendor has cloud management product based on opensource products like Ansible, OpenStack, cloudstack etc. For cooling NVIDIA offers NVIDIA Data Center GPU Manager that includes cooling management features.
The only remaining OEM advantage is that Contract ODMs typically require customers to fund inventory upfront, while OEM often pays NVIDIA and suppliers upfront, builds and delivers systems, and may wait up to 90 days for customer payment.