top of page

📢 STT 訂閱專區已上線

免費文章會照常更新,一篇都不會少。訂閱是「加強版」——每週深度週評、財報法說的完整判讀、所有長篇深度報告全包。

免費讓你跟上,訂閱讓你看懂、能做判斷。

月訂 NT$199|年訂 NT$2,000(約 NT$167/月)
👉 立即訂閱: vocus.cc/salon/simpletechtrend

OCP Global Summit 2025: Dell on Scaling Data Center Infrastructure

3 days ago
3 min read

Introduction

At OCP Global Summit 2025, Dell shared how it is using OCP standards and modular design to evolve data center infrastructure for the AI and HPC era. From DC-MHS to Open Rack v3 to liquid cooling, Dell stressed that standardization + modularity + liquid-cooling innovation is its core strategy for coping with rapid GPU generational change and soaring power draw.


Key Points

Long-Term OCP Involvement and DC-MHS

Dell has worked with OCP for more than a decade, initially focusing on NIC, networking and storage specifications, and later investing heavily in DC-MHS (Data Center Modular Hardware System). Beyond helping write the spec, Dell has moved its entire server portfolio to DC-MHS capability, creating what it calls the industry's most modular compute platform. This lets Dell respond quickly to market demand in its product roadmap while gaining significant advantages in speed and scale.


Exascale Design and Liquid Cooling

For its exascale designs, Dell is focused on liquid cooling, cabling and open rack design, and participates in the OCP cold-plate workstream to bring open AI systems into production. With GPU power growing faster than the CPU curve, Dell expects real-world GPU power to often run 20–40% above launch specs, so facilities must prepare for much tougher thermal requirements in advance.


The Rapid Jump in GPU Scale

In just two years, AI training clusters have grown from 4,000 GPUs (about 10 MW) to 18,000 GPUs, and now to 100,000+ GPU data halls. Today, Dell customers often use 4 MW as the basic deployment unit, scaling in multiples up to 40 MW. This explosive demand is accelerating innovation in liquid cooling and high-performance infrastructure.


Liquid-Cooling Solutions and Enclosed Racks

Dell continues to innovate on OCP liquid-cooling systems, covering flow design, cold plates, thermal interface materials and CDUs (Cooling Distribution Units), and has developed its own durable quick-disconnect couplings. More recently it has pushed an enclosed Door Heat Exchange Enclosure that cuts cooling power by 60% and works with non-chilled water, greatly improving flexibility and deployment efficiency.


Putting Open Rack v3 to Work

Building on Open Rack v3, Dell optimized CDU connections, busbar design, quick disconnects and cable trays to deliver a 480 kW rack that can be upgraded to 1 MW. This allowed Dell to deploy 50,000 GPUs in six weeks, supporting customers at unprecedented density and speed.


Modular Supercomputers for HPC and AI

Working with Berkeley Lab and the University of Texas at Austin, Dell built a modular HPC supercomputer that fits 27,000 CPU cores and 144 GB200 GPUs in a single rack, with future support for accelerators such as NVIDIA B200/B300 and AMD MI355/MI450. It signals that HPC and AI are entering a new generation that is modular, upgradable and scalable.


Conclusion

Dell's core message at OCP 2025: OCP standardization and modular design, combined with liquid cooling and high-density racks, let data centers support the explosive growth of AI and HPC faster and at lower cost. This makes Dell a major contributor to the OCP community and underpins its leadership position in next-generation AI data center infrastructure.


Further Perspectives

  1. Technology impact

    • Dell's 480 kW → 1 MW rack design implies that future data centers will broadly move to "megawatt-per-rack" designs, directly driving demand for the liquid-cooling supply chain, power electronics and silicon photonics interconnects.

  2. Supply chain observations

    • Dell is not just a server vendor; it is developing its own liquid-cooling modules, CDUs and quick disconnects, which may put it in coopetition with traditional cooling and power vendors.

    • The launch of modular HPC and AI systems will tighten collaboration among NVIDIA, AMD and Dell, and also means more competitive pressure on ODM/white-box players.

  3. Market trends

    • From Oracle's gigawatt AI campus and Microsoft's Fairwater to Dell's modular supercomputers, the overall trend is clear: AI data centers are shifting from "supercomputers" to "super campuses".

    • Dell's "modularity + liquid cooling + high density" strategy targets mid-to-large cloud and research markets, giving it a differentiated position beyond the hyperscalers.

Recent Posts

See All

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page