How does the closed-loop cooling system work?

The system circulates a water-and-glycol coolant through servers and heat exchangers before reusing it. Meta expects the coolant to remain usable for up to a decade.

What results did Meta see from using reinforcement learning in cooling?

In a data-center pilot, the approach reduced energy consumed by air-cooling supply fans by an average of 20% and lowered water consumption by 4% across varying weather conditions.

What is Meta's planned capital expenditure for 2026?

Meta expects full-year 2026 capital expenditures of $130-$145 billion, primarily for servers, data centers, and network infrastructure.

AI Business & IndustryYahoo Finance AI2h agoAug 29, 2026, 01:01 JST2 min read

Meta uses closed-loop liquid cooling across most new AI data centers

Send on LINE

Key takeaway

Meta is deploying closed-loop liquid cooling in most of its newest AI data centers. This reduces water use and allows higher GPU density.
The company is also using reinforcement learning to cut energy and water consumption.
These efforts support its AI infrastructure buildout.

3 Key Points

What happened
Meta announced it is using closed-loop liquid cooling across the majority of its newest AI-optimized data centers. The system circulates a water-and-glycol coolant through servers and heat exchangers, and Meta expects the coolant to remain usable for up to a decade.
Why it matters
The cooling method reduces ongoing water consumption and enables higher GPU density within server racks. In a pilot, reinforcement learning cut energy used by air-cooling supply fans by an average of 20% and lowered water consumption by 4%, supporting Meta's AI infrastructure expansion.
What to watch
Meta spent $31.1 billion on capital expenditures in Q2 2026, mainly for servers, data centers, and network infrastructure. It expects full-year 2026 capital expenditures of $130-$145 billion as it maximizes computing capacity during 2026 and 2027.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Get news like this every morning — free

Context & Analysis

Meta's cooling efficiency push comes as its compute demand rises, making data-center efficiency critical. The closed-loop system directly cuts water use, while direct-to-chip cooling allows more GPUs per rack without expanding data-center space. These gains are expected to support Meta's broader AI infrastructure expansion as it scales compute-intensive models and services.

Meta is deploying AI across recommendations and advertising, and developing personal agents, business agents, and enterprise offerings, creating multiple ways to monetize its growing compute capacity. The company believes near-term AI capacity remains highly valuable, with numerous ROI-positive opportunities in its core business.

Meanwhile, Meta faces competition from Reddit and Snap in AI. Reddit recently launched AI moderation tools that block 23 million spam views daily and cut response time for hateful content to under five seconds. Snap is leveraging AI through products like SPECS and Lens+ to drive subscription revenue and engagement. These moves highlight the broader AI race as Meta scales its infrastructure.

FAQ

How does the closed-loop cooling system work?: The system circulates a water-and-glycol coolant through servers and heat exchangers before reusing it. Meta expects the coolant to remain usable for up to a decade.
What results did Meta see from using reinforcement learning in cooling?: In a data-center pilot, the approach reduced energy consumed by air-cooling supply fans by an average of 20% and lowered water consumption by 4% across varying weather conditions.
What is Meta's planned capital expenditure for 2026?: Meta expects full-year 2026 capital expenditures of $130-$145 billion, primarily for servers, data centers, and network infrastructure.

Yahoo Finance AIRead Original Article

Send on LINE