Nvidia's Next-Gen AI Architecture: How Chips Running Hotter Could Decouple Data Centres from Water Woes

Nvidia's Next-Gen AI Architecture: How Chips Running Hotter Could Decouple Data Centres from Water Woes​

The Revolution of Heat: Introducing 100% Liquid Cooling for AI Factories​

Nvidia has offered a significant answer to the rising environmental scrutiny surrounding Artificial Intelligence infrastructure: making data centre machines run deliberately hotter. The company's newest Rubin-generation AI infrastructure is designed to operate effectively with liquid cooling running at temperatures up to 45 degrees Celsius (113 degrees Fahrenheit). This temperature is warm enough that it directly challenges traditional, water-intensive cooling systems prevalent in the current AI boom.

The significance of this development lies in its design philosophy. The Rubin generation is stated as the first to achieve 100 percent liquid cooling, meaning every chip and networking component is cooled by a closed-loop liquid system without internal fans. This technology forms part of Nvidia’s DSX AI factory reference design, providing a critical blueprint for future large-scale AI infrastructure.

Solving the Water Footprint: From Evaporative Cooling to Zero Consumption​

Conventional data centres heavily rely on cooling towers that use evaporation to remove waste heat, a process which consumes vast amounts of water. This is a major flashpoint as cloud companies and AI developers rapidly expand their "AI factories." Nvidia’s 45°C liquid-cooling architecture provides a viable alternative for facilities located in climates suitable for dry coolers.

By utilizing this advanced system, cooling-related water use can be reduced from approximately 2.6 million gallons per megawatt per year (in conventional cooling-tower systems) down to nearly zero. Ali Heydari, Nvidia’s director of data centre cooling and infrastructure, confirmed that the DSX reference design achieves "zero water consumption" when dry coolers are used, potentially requiring chillers only for a small part of the year in certain climates.

Financial Impact and Efficiency Gains in Hyperscale Facilities​

Cooling is not merely an environmental concern; it accounts for substantial costs. McKinsey has estimated that cooling can consume around 40 percent of a data centre’s energy consumption, making efficiency gains crucial. Nvidia projects that by transitioning to liquid-cooled infrastructure, a 50-megawatt hyperscale facility could save over $4 million annually in cooling-related water and energy expenses.

This technological leap comes at a critical time for the company's growth trajectory. In the first quarter of fiscal 2027, Nvidia reported record revenue of $81.6 billion, an increase of 85 percent year-over-year, with data centre revenue alone reaching $75.2 billion, up 92 percent.

The Broader Context: AI's Resource Strain and the Global Shift​

The rapid expansion in model size and enterprise AI adoption is generating intense pressure on computing resources. More compute requires denser racks and generates higher heat loads, amplifying the strain on power and cooling systems globally. Nvidia’s solution, therefore, represents more than a simple hardware upgrade; it is a direct response to the debate surrounding AI's resource footprint.

The shift toward liquid cooling could allow data centre operators to build denser AI facilities without escalating cooling-water demand at the same pace. However, experts caution that while Nvidia addresses on-site water usage sharply, it does not erase the broader water concerns tied to the entire lifecycle of AI infrastructure, including manufacturing and power generation.

India’s Accelerating Data Centre Growth and Policy Relevance​

The development is particularly pertinent for India. The country's data centre capacity has grown from around 520 MW in 2020 to nearly 1.5 GW by mid-2025, with projections showing it could reach between 4.5–6.5 GW by 2030, according to the Council on Energy, Environment and Water (CEEW).

As of January 2026, India had approximately 271 data centres, led by markets in Mumbai, Chennai, Hyderabad, and Bengaluru. CEEW estimates that Indian data centres used around 150 billion litres of water in 2024, a figure projected to more than double by 2030. This rapid buildout makes cooling technology a critical policy issue for sustainable growth.

Caveats and Adoption Timeline for Next-Gen Cooling​

The claim requires careful assessment. Nvidia’s focus is mainly on on-site water used for cooling, not the broader indirect water footprint associated with power generation or manufacturing supply chains. Furthermore, the benefits are climate-dependent; Nvidia refers to "favourable climates," and some regions may still need chillers for a portion of the year.

Adoption, however, is expected to be staggered. Existing data centres will continue running older systems, meaning not all facilities can be immediately redesigned for fully liquid-cooled racks. The shift is primarily anticipated in new hyperscale facilities and AI factories built specifically for next-generation chips.