AWS powers up AI portfolio with custom 3nm chip, boosts global infra by 3.8 GW

Las Vegas, Dec 3 (PTI): To lower the exorbitant costs of developing artificial intelligence (AI) and meet the insatiable global demand for computing power, Amazon Web Services (AWS) has unveiled its most powerful custom AI chip yet, and announced a massive expansion of its global physical infrastructur.

AWS powers up AI portfolio with custom 3nm chip, boosts global infra by 3.8 GW

Las Vegas, Dec 3 (PTI): To lower the exorbitant costs of developing artificial intelligence (AI) and meet the insatiable global demand for computing power, Amazon Web Services (AWS) has unveiled its most powerful custom AI chip yet, and announced a massive expansion of its global physical infrastructure.

During his keynote address at AWS re:Invent 2025 here on Tuesday, CEO Matt Garman introduced Trainium 3, the company's next-generation chip designed specifically for training massive AI models.

Garman stated that the new Trainium 3 chip offers four times the performance of its predecessor, and is built using a "3-nanometre" process.

To put it simply, a smaller nanometre number means manufacturers can pack more processing power into a smaller space, making the chip significantly faster and more energy-efficient.

"Trainium already represents a multi-billion dollar business today and continues to grow really rapidly," Garman said, adding that AWS has already deployed over a million Trainium chips to date.

AWS also introduced the Trainium 3 UltraServer, which connects 144 of these chips together to function as a single integrated system, allowing developers to train huge AI models in a fraction of the time it usually takes.

Garman also teased Trainium 4, which is in the works, and said the model will deliver six times the FPU performance, four times more memory bandwidth, and twice the memory bandwidth capacity than Trainium 3 to support large models.

Switching from silicon to the buildings that house it, Garman highlighted that AWS added 3.8 gigawatts of power capacity to its network in 2024. To put that in perspective, that is roughly equivalent to the energy output of multiple large nuclear power plants.

"AWS has by far the largest and most broadly deployed AI cloud infrastructure anywhere in the world. Our global network of data centres spans 38 regions, 120 availability zones, and we have already announced plans for three more regions. In fact, in the last year alone, we have added 3.8 gigawatts of data centre capacity, more than anyone in the world.

"We have the world's largest private network, which has increased 50 per cent over the last 12 months to now being more than 9 billion kilometres of centres of terrestrial and subsea cable. That's enough optical cabling to reach from the earth to the moon, and back for 11 times," Garman noted.

Amazon Web Services (AWS) is the cloud computing arm of Amazon. AWS's Indian customer base includes Razorpay, Zerodha, HDFC Life, Sony Liv, Airtel, Zomato, MakeMyTrip and Mahindra, among others.

The company has cloud regions in Hyderabad and Mumbai. PTI ANK DIV DIV

(This story is published as part of the auto-generated syndicate wire feed. No editing has been done in the headline or the body by ABP Live.)