NVIDIA CEO Jen-Hsun Huang had this to say at Tuesday's launch. Yesterday, NVIDIA unveiled its next-generation GH200 Grace Hopper superchip platform, built for the era of accelerated computing and generative AI.

Jen-Hsun Huang noted that in order to meet the growing demand for generative AI, data centers need accelerated computing platforms that are tailored to specific needs. The new GH200 chip platform offers superior memory technology and bandwidth, the ability to boost lossless connected GPU aggregation performance, and a server design that can be easily deployed throughout the data center.

It is worth mentioning that the incoming wave of big models is spawning a variety of AI-native applications, driving a surge in arithmetic demand, and the data center market specifically designed to cope with data-intensive AI applications is rapidly emerging.

Data centers usher in new changes

As established cloud computing providers race to retrofit data centers with advanced chips and make other upgrades to meet the demands of artificial intelligence software, some upstart builders see an opportunity to develop new facilities from the ground up, analysts note, according to the Wall Street Journal.

A data center resembles a large warehouse equipped with multiple racks of servers, networking, and storage for storing and processing data. AI data centers have more servers using high-performance chips than traditional data centers, so AI data center servers can consume an average of 50 kilowatts or more per rack, compared to about 7 kilowatts per rack in traditional data centers.

This means that AI data centers will need to build additional infrastructure capable of delivering higher power, and because the additional power usage generates more heat, AI data centers will also need other cooling methods, such as liquid cooling systems, to prevent equipment from overheating.

Manju Naglapur, senior vice president at services and consulting firm Unisys, noted:

Purpose-built AI data centers house servers that utilize AI chips, such as NVIDIA's GPUs, and can run multiple computations simultaneously as AI applications sift through huge data stores. These data centers are also equipped with fiber optic networks and more efficient storage to support large-scale AI models. AI data centers are highly specialized buildings that require a significant investment of money and time. Spending in the global AI infrastructure market is expected to reach $422.55 billion by 2029, with a compound annual growth rate of 44 percent over the next six years, according to research firm Data Bridge Market Research.

DataBank CEO Raul Martynek said the pace of AI deployment is likely to lead to a shortage of data center capacity in the next 12 to 24 months.

AI arithmetic upstart secures $2.3 billion in funding

Currently, various giants are betting on AI data centers, and "real estate benchmark" Blackstone has sold its house to invest in AI data centers.Meta has also said that it will build a new AI data center.

As mentioned in the previous article, CoreWeave, an AI computing power newcomer, took NVIDIA H100 collateralized loan and obtained debt financing of $2.3 billion (about 16.5 billion yuan).

CoreWeave said the funds will be used to accelerate the construction of AI data centers, which is another funding round for the company after it received $221 million in April and $200 million in May this year.Founded six years ago, CoreWeave has seven AI data centers online and is expected to double by the end of this year.

CoreWeave is working with NVIDIA and Inflection AI to build a mega AI server cluster with the goal of running 22,000 NVIDIA H100s. if built, it would be the largest AI server cluster in the world.

It's worth noting that according to CoreWeave's official website, their services are 80% cheaper than traditional cloud computing vendors. NVIDIA's latest HGX H100 servers, which contain eight H100s with 80G of video memory and 1T of RAM kind of thing, start at as little as $2.23 (16 yuan) per hour.

And compared to its predecessor, the new GH200 Grace Hopper platform's dual-chip configuration increases memory capacity by 3.5x and bandwidth by three times, with 144 Arm Neoverse high-performance cores, 8 petaflops of AI performance, and 282GB of the latest HBM3e memory technology in a single server.

No wonder in this era of LLM explosion, Jen-Hsun Huang still boldly said "the more you buy, the more you save"!

NVIDIA CEO Jen-Hsun Huang had this to say at Tuesday's launch. Yesterday, NVIDIA unveiled its next-generation GH200 Grace Hopper superchip platform, built for the era of accelerated computing and generative AI.

Jen-Hsun Huang noted that in order to meet the growing demand for generative AI, data centers need accelerated computing platforms that are tailored to specific needs. The new GH200 chip platform offers superior memory technology and bandwidth, the ability to boost lossless connected GPU aggregation performance, and a server design that can be easily deployed throughout the data center.

It is worth mentioning that the incoming wave of big models is spawning a variety of AI-native applications, driving a surge in arithmetic demand, and the data center market specifically designed to cope with data-intensive AI applications is rapidly emerging.

Data centers usher in new changes

As established cloud computing providers race to retrofit data centers with advanced chips and make other upgrades to meet the demands of artificial intelligence software, some upstart builders see an opportunity to develop new facilities from the ground up, analysts note, according to the Wall Street Journal.

A data center resembles a large warehouse equipped with multiple racks of servers, networking, and storage for storing and processing data. AI data centers have more servers using high-performance chips than traditional data centers, so AI data center servers can consume an average of 50 kilowatts or more per rack, compared to about 7 kilowatts per rack in traditional data centers.

This means that AI data centers will need to build additional infrastructure capable of delivering higher power, and because the additional power usage generates more heat, AI data centers will also need other cooling methods, such as liquid cooling systems, to prevent equipment from overheating.

Manju Naglapur, senior vice president at services and consulting firm Unisys, noted:

Purpose-built AI data centers house servers that utilize AI chips, such as NVIDIA's GPUs, and can run multiple computations simultaneously as AI applications sift through huge data stores. These data centers are also equipped with fiber optic networks and more efficient storage to support large-scale AI models. AI data centers are highly specialized buildings that require a significant investment of money and time. Spending in the global AI infrastructure market is expected to reach $422.55 billion by 2029, with a compound annual growth rate of 44 percent over the next six years, according to research firm Data Bridge Market Research.

DataBank CEO Raul Martynek said the pace of AI deployment is likely to lead to a shortage of data center capacity in the next 12 to 24 months.

AI arithmetic upstart secures $2.3 billion in funding

Currently, various giants are betting on AI data centers, and "real estate benchmark" Blackstone has sold its house to invest in AI data centers.Meta has also said that it will build a new AI data center.

As mentioned in the previous article, CoreWeave, an AI computing power newcomer, took NVIDIA H100 collateralized loan and obtained debt financing of $2.3 billion (about 16.5 billion yuan).

CoreWeave said the funds will be used to accelerate the construction of AI data centers, which is another funding round for the company after it received $221 million in April and $200 million in May this year.Founded six years ago, CoreWeave has seven AI data centers online and is expected to double by the end of this year.

CoreWeave is working with NVIDIA and Inflection AI to build a mega AI server cluster with the goal of running 22,000 NVIDIA H100s. if built, it would be the largest AI server cluster in the world.

It's worth noting that according to CoreWeave's official website, their services are 80% cheaper than traditional cloud computing vendors. NVIDIA's latest HGX H100 servers, which contain eight H100s with 80G of video memory and 1T of RAM kind of thing, start at as little as $2.23 (16 yuan) per hour.

And compared to its predecessor, the new GH200 Grace Hopper platform's dual-chip configuration increases memory capacity by 3.5x and bandwidth by three times, with 144 Arm Neoverse high-performance cores, 8 petaflops of AI performance, and 282GB of the latest HBM3e memory technology in a single server.

No wonder in this era of LLM explosion, Jen-Hsun Huang still boldly said "the more you buy, the more you save"!