NVIDIA
2026-06-26
NVIDIA Rubin Mandates 100% Liquid Cooling with 45°C Warm Water, Reshaping Data Center Thermal Design
Summary
Key Takeaways
On June 21, 2026, NVIDIA detailed the Rubin platform's liquid cooling design, the world's first 100% liquid-cooled AI compute platform, eliminating all fans. Unlike the hybrid cooling of Blackwell/GB200, Rubin uses 45°C warm water inlet (55°C return), eliminating chillers and enabling dry cooler or free cooling, drastically reducing energy and water consumption. A single inlet/outlet delivers coolant to multiple high-power chips on a board, enabling a simpler tray-level cooling architecture. Full liquid cooling allows higher rack density: a system previously requiring six rack units now fits in two. Mass production begins H2 2026, with NVIDIA mandating that all data centers must transition to liquid cooling, making it a requirement rather than an option. GMI estimates the global AIDC server liquid cooling market will reach $60B in 2026 and $271B by 2035 (CAGR ~18%).
Why It Matters
NVIDIA's move is a lock-in strategy masquerading as innovation. By mandating liquid cooling for Rubin, NVIDIA creates an adoption barrier: customers must invest in liquid infrastructure before deploying Rubin, raising switching costs to AMD or Intel. The 45°C warm water design hides physical limitations: retrofitting data centers with CDUs, piping, and leak detection incurs high CapEx; warm water may cause tail latency in sensitive optical modules on RoCEv2 or NVLink networks; maintenance complexity rises. The tray-level cooling architecture ties coolant loops to NVIDIA's NVSwitch and Grace CPU, making third-party liquid cooling solutions difficult to integrate. This shifts control from data center operators to NVIDIA, stripping architectural flexibility.
PRO Decision
[Vendors] AMD and Intel should partner with liquid cooling vendors (CoolIT, Asetek) to create open-standard liquid cooling platforms that decouple from NVIDIA's tray design, support 45°C warm water but also lower temps, and reduce customer switching costs. Push OCP to define liquid cooling interface standards to break NVIDIA's lock-in.
[Enterprises] CIOs and architects must conduct zero-trust technical audits: demand full lifecycle TCO models including CapEx for retrofitting, maintenance training, and leak insurance. Insist on cross-platform compatibility so liquid infrastructure supports AMD, Intel, and future custom chips. Require NVIDIA to open coolant loop interface specs to avoid single-vendor lock-in.
[Investors] See through the PR: NVIDIA's liquid cooling mandate increases supplier concentration risk, boosting short-term revenue but potentially slowing adoption due to customer pushback. Look for independent growth in liquid cooling infrastructure vendors (Vertiv, nVent) and white-box server makers (Supermicro) that can offer NVIDIA-agnostic solutions. Beware of NVIDIA stock price already pricing in a liquid cooling monopoly premium.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)