A new Google study puts up some fascinating numbers about the holistic environmental impacts of AI, and how to measure them: Not just water, not just power, and not just based on conditions during active queries. On one hand, the Google researchers found that other measurement approaches weren’t factoring in some major sources of energy use; on the other, they also found that some public estimates of AI inference impacts were considerably higher than what their numbers showed. And what exactly did they show? A median text prompt for Gemini AI consumes just 0.24 watt-hours—less than watching nine seconds of TV—and about five drops of water. Still, with over 400 million monthly users making billions of queries, including heavier image and video requests, the cumulative impact is considerable. Click through to get the whole picture.
Kelly Hill
Executive Editor
RCRTech
AI Infrastructure Top 3
Pinpointing AI impacts: A single text prompt for Gemini AI uses the energy equivalent about 9 seconds of TV time, plus 5 drops of water. Now multiply that by the queries of 400 million monthly active users.
New China chip: Nvidia is reportedly building a new Blackwell-based AI chip for China, better than the H20 but not quite at the level of its Ultra GPUs, as Chinese regulators push big tech firms to pause H20 purchases amid security reviews.
Vodafone Idea, IBM & AI: Vodafone Idea and IBM have launched an AI innovation hub in India to co-develop automation tools and embed AI in DevOps, aiming to modernize IT systems and strengthen telecom service resilience.
AI-Powered Telecom Infrastructure
Supermicro, in collaboration with NVIDIA, delivers AI-powered infrastructure tailored for telcos, enhancing operational efficiency, network management, and customer experiences. Explore now
AI Today: What You Need to Know
Meta hiring pause: After making headlines with eye-popping salary offers for AI talent, WSJ reports that Meta has frozen hiring in its AI unit amid an ongoing reorganization and investor concerns about costs.
OpenAI eyes IaaS: OpenAI is expecting to spend trillions on AI infrastructure, and the company is looking to monetize those investments by developing an infrastructure-as-a-service offering, per its CFO.
Meta’s KC DC: Meta’s new Kansas City data center is officially open and serving traffic, the company said. The site, which is LEED Gold certified, has been in the works since 2022 and represents a $1 billion investment.
Lower AI expectations?: Sam Altman has warned investors about AI bubble hype, while Eric Schmidt urged Silicon Valley to shift focus from superhuman AI. Is there a recalibration happening in high-profile AI discourse?
Questioning AI valuations: U.S. tech stocks slid amid growing investor doubts over AI-driven valuations. Big names like Nvidia and Palantir plunged, sparking concerns about an overheated AI rally ahead of public remarks by the Fed chair.
DeepSeek’s latest splash: DeepSeek’s new V3.1 model, boasting 685 billion parameters and ultra-low costs, challenges U.S. giants like OpenAI and Anthropic and reflects the latest salvo in the global competition for AI dominance.
AI infra jobs: Data centers face soaring demand but struggle to find skilled tradespeople like electricians and HVAC techs. There are roughly 400,000 skilled trade jobs unfulfilled in America now, and the number is growing.
DC MOUs: Samsung and LG signed MoUs with Vietnamese partners to develop major hyperscale data centers, signaling deepening South Korea–Vietnam tech ties and expanding Southeast Asia’s role in global AI and cloud infrastructure.
Powering the West: A newly announced $15 billion power company merger would create a Western utility with the scale to meet power needs that are increasingly driven by data centers, according to company execs.
Caterpillar, energized: Caterpillar and Hunt Energy have teamed up to provide independent energy options for data centers. With an initial project expected to launch in Texas, the multi-year effort aims to deliver up to 1 GW of power for data centers across North America.