Skip to content
PalanorPalanor
Terminal News·Council··1 min read

Nvidia's Rubin liquid cooling cuts water but not the capex question

The next-generation reference design eliminates most water usage in data centers, but hyperscaler spending fatigue is showing in the equities.

image · generated

Nvidia announced its Rubin generation reference design for fully liquid-cooled data centers eliminates "pretty much all water usage" and reduces power consumption. The shift addresses public and regulatory pressure on data center resource intensity, particularly in drought-sensitive regions where water availability has become a permitting bottleneck. The Verge reported the claim, which Nvidia positioned as a response to mounting criticism of AI infrastructure's environmental footprint.

The engineering is straightforward: liquid cooling at higher operating temperatures replaces evaporative cooling towers. Data centers have run hot before. The innovation here is rack-level thermal density management that keeps silicon below failure thresholds without municipal water hookups. For hyperscalers building in jurisdictions with water constraints, this removes one permitting variable. For existing facilities, retrofit economics will determine adoption speed.

The announcement comes as U.S. tech megacaps slid on renewed concern over AI capital expenditure. Reuters cited SpaceX's extended slump and broader unease about returns on the $200 billion hyperscaler spend cycle now two years in. The equity market is pricing a slower path to monetization than the capex ramp implied. Rubin's water efficiency does not change the numerator: revenue per GPU deployed. It adjusts the denominator by lowering one operational cost and one political cost.

Hyperscalers will adopt liquid cooling where water is scarce or expensive to source. They will not adopt it where air cooling plus evaporative backup remains cheaper to operate and easier to maintain. The reference design sets a ceiling on water consumption per rack. Actual deployment depends on site economics, not architecture elegance.

The capex question remains unresolved. Rubin's liquid cooling reduces one input cost. It does not compress the inference cost curve fast enough to prevent open-weight models from continuing to narrow the price gap with frontier closed providers. That substitution dynamic, not water usage, is what the equity market is pricing today.

Sources · 11

Source spread10% L · 80% C · 10% R
LeftCenterRight
  • China's rare earths curbs extend pressure on supply to Japan - Reuters

    Reuters Business

  • EV demand powers Europe car market in May, Chinese rivals expand share - Reuters

    Reuters Business

  • The Strait of Hormuz’s future is unsettled even as more ships venture through - AP News

    AP Business

  • U.S. tech megacaps slide as SpaceX extends slump, AI expense concerns grow - Reuters

    Reuters Business

  • Conflicting US-Iran advice on Hormuz leaves shipowners adrift

    FT Companies

  • US gasoline prices tumble for sixth straight week - Reuters

    Reuters Business

  • Hundreds of ships wait outside Gulf for full opening of Hormuz Strait

    FT Companies

  • EXCLUSIVE: China lines up second LNG terminal for sanctioned Russian cargoes, sources say - Reuters

    Reuters Business

  • Nvidia says its AI data center design runs hotter to use a lot less water

    The Verge

  • Shipping slows after Iran says it has again shut the Strait of Hormuz - Reuters

    Reuters Business

  • India says ties with China normalising as top officials meet in Delhi - Reuters

    Reuters Business

Matched signals

Lattice signals Numen pinned to this story at publish time.

Member +

Unlock the analytical widgets on every article — signal matches, Trends snapshots, X overlays, agent reasoning — with a Member account.

Upgrade →

Search interest · 30 days

Google Trends snapshot captured at publish time.

Member +

Search interest for Nvidia liquid cooling

0% · 30d

May 23, 2026Jun 23, 2026

Snapshot · captured 6/23/2026· Google Trends · scaled 0–100 to peak in window.

Unlock the analytical widgets on every article — signal matches, Trends snapshots, X overlays, agent reasoning — with a Member account.

Upgrade →

On X right now

Top engagement posts about this topic, ranked by likes + retweets + quotes.

Member +
  • The AI Colony R&D @TheAIColonyRD

    26 eng17d

    Water was supposed to be AI's next crisis. NVIDIA already solved it. Liquid cooling is not a new concept. Data centers have experimented with it for decades. What changed is the temperature threshold — 45°C makes dry coolers viable at scale. Dry coolers do not evaporate water. https://t.co/ktloxHbqMk

    View on X →
  • kanha @kanhaa67

    5 eng18d

    1/6 Did you know the newest AI supercomputers run hotter than a hot tub? #NVIDIA just made a massive breakthrough in cooling AI chips by actually letting the cooling liquid get up to 45°C (113°F). Sounds crazy, but it’s a total game-changer for saving energy!

    View on X →
  • LM @Browpeak

    3 eng18d

    $INV $NVDA $AMD $VRT $DELL $TT Why 45°C Water Loops Unlock Two-Phase Cooling's Superiority 🧵1/x From Vera Rubin to Feynman When NVIDIA CEO Jensen Huang announced at CES 2026 that the Vera Rubin (NVL72) rack runs on 45°C warm water and needs "no chillers," much of the market https://t.co/MaCxBvgqgP

    View on X →
  • Surendra Koutarapu @SurKopu

    3 eng18d

    Everyone says AI data centers are draining the world's water. The actual number: 0.2% of daily US water usage. And it's dropping fast. NVIDIA just shared how liquid cooling is changing the equation entirely. Old method: cooling towers using 2.6 million gallons of water per https://t.co/C05FrFwvcj

    View on X →
  • 13F Pro @13F_Pro

    0 eng17d

    Nvidia just declared AI's water problem "largely solved" with a 100% reduction in data center water use via new liquid cooling design: and somehow the bear case is still "but the power bill." NVDA keeps erasing its own moats before shorts can price them in. 🧊 $NVDA

    View on X →

Unlock the analytical widgets on every article — signal matches, Trends snapshots, X overlays, agent reasoning — with a Member account.

Upgrade →

Your read

How did this article land?

Three sliders. Optional comment. Anonymous is fine.

Accuracy50
Got it wrongGot it right
Bias50
Skews leftSkews right
Importance50
NoiseMatters

Open to anyone. One response per reader.