Is the data center slowly becoming the AI data center?

19.06.2024 352 0

Over the past year or so a lot has been changing in the IT industry. The main “culprit” is artificial intelligence (AI). AI has slowly engulfed the IT sector and has become the driving force behind many of the key decisions for the expansion and building of new IT infrastructure, services, and investments.

And all of this is just the beginning. AI will continue to reshape not only the IT industry, but the entire world for years to come! Naturally, data centers have received most of the attention during the initial stages of AI transformation. After all it’s exactly because of data centers we can have the AI revolution in the first place.

Those facilities provide all the computing and storage resources needed for AI development, training, processing and usage, and over the past year or so AI has become the de facto leader of the data center. A lot of data centers have been reworked to better fit the requirements of AI – so much so, that there’s a bit of a debate forming: are data centers slowly becoming simply “AI centers”? If we look at some of the latest developments in the sector, it might seem plausible. Let’s explore.

AI is conquering the data center energy

One popular topic lately has been that AI demands so many resources that data center operators can’t keep up with – including even building the new facilities. Also, another problem is the lack of energy supply. At least that’s the worry. In fact, many forecasts say there will be an energy crunch for data centers soon. Some even said, there won’t be enough power for everyone within two years.

A recent article by Wired states that there is indeed a huge chase happening to secure green energy sources right now globally, and data center operators are among the most active in this pursuit. For example, AWS recently announced it has reserved more than half of the 800MW output of the new wind farm off the coast of Scotland.

And in Europe there’s a big data center competition forming. Microsoft will invest $3.2 billion in data centers in Sweden. It will also double its data center footprint in Germany. And it will invest a further $4.3 billion for, you guessed it, AI data centers in France. All while Amazon is also working on a $8.5 billion worth of projects in Germany and $17.1 billion in Spain. Google will invest billions in Europe, again mostly for data centers, too. And most of this new computing power will go towards AI.

The more AI we have, the more resources we will need. “There is a recognition that as power demand increases, the industry will have to find alternative energy sources,” says Colm Shorten, senior director of data center strategy at real estate services company JLL to CleanTechnica.

All of this will bring energy issues, grid conflicts and other challenges that need to be solved. “In the future, technologies like advanced nuclear reactors, renewable energy sources, and energy storage solutions will be crucial in making this possible,” says Kilian Wagner, an expert in sustainable digital infrastructures at German digital association Bitkom. And yes, all of this is mainly to fuel AI algorithms.

It’s all about the hardware

All of the major chip and other data center equipment makers are already fully embracing the AI reality. As siliconAngle says, “the transformation of data centers into artificial intelligence data centers is reshaping the industry, highlighting the importance of innovation, efficiency and sustainability.”

For example, Dell has developed a whole strategy to integrate AI into data centers. “The amount of focus on innovation around AI has just been mind-blowing and staggering. The call to action has been clear, and it’s been exciting,” says Brian Payne, vice president of cloud product management at Dell Technologies Inc.

He also notes that the surge in power needs for data centers is going to be massive. “We’ve got to think about how do we cope with 8X growth in power over the next six years, all driven by gen AI infrastructure. Well, the good news is that we have a legacy of designing systems for pushing the envelope. If you think back to the advent of the hyperscale public cloud, we were very much involved in building out those data centers, which were pushing the envelope at that point in time … how do we make sure that the design within the system is as efficient as possible. That’s absolutely necessary because, at the end of the day, what’s good for sustainability is also good for business.”

Dell wants to offer end-to-end solutions for customers to develop sustainable data center plans and integrate generative AI into their operations. Payne adds, that the approach includes the usage of more recyclable materials and focusing on the circular economy. Those are key to the modern AI data center, the company says.

His colleague Varun Chhabra, Senior Vice President of Product Marketing, ISG and telecom at Dell adds: “Whether you’re retrofitting your existing data center, you’ve got to think through the entire stack. What’s the position that your hardware vendor is taking on? How these are built, how these are designed. What’s the software layer in terms of being able to manage these? What are the services and professional services that you’re getting to be able to actually think about this from the ground up?”

He feels the days of traditional data centers are gone. We are now moving to AI data centers which are also “AI factories designed for efficiency and sustainability”.

Cisco and Nvidia think alike

The competition is heating up on all fronts. And it forms more alliances. Cisco and Nvidia announced they have co-developed an all-in-one AI data center solution. It’s called “Cisco Nexus HyperFabric AI clusters,” because of course companies love giving long, complex names to their products. But while the name might be complex, the goal of the product is to simplify and help enterprises implement and manage AI data center infrastructure in data centers, colocation facilities and edge locations. Cisco also adds that the Nexus HyperFabric AI clusters will be the first data center product line that can be cloud-managed. The trials with selected customers will start in the fourth quarter of 2024 and general availability will follow soon after.

Speaking to DataCenterKnowledge, Cisco’s Vice President of Product Management for Data Center Networking, Murali Gandluru, says the product will allow enterprises to “partake in the AI evolution and easily design, deploy, operate, and manage full-stack AI infrastructure in a plug-and-play model.”

Nvidia’s involvement is no surprise either. The company has masterfully positioned itself as one of the leading players in AI hardware. Its AI processors have become the benchmark for the industry, and they are bringing in billions in revenue every quarter. As a result, Nvidia is at historic heights, propelling the company well on the way to a $3 trillion market cap. And the company has secured partnerships with Dell, Hewlett Packard Enterprise, Lenovo and other key players all of which are also developing AI hardware solutions for data centers.

Intel is joining the party

A few years ago Intel was this unstoppable force in the CPU world. Its products were powering the vast majority of consumer PCs and servers globally. Then the company started to lose its grip on the PC market due to the strong competition from AMD. And then AMD and Nvidia started chipping away at the server market, too.

Now the booming AI data center market gives Intel a great opportunity to gain some much needed returns. The company kicked off June 2024 with its new generation Xeon 6 data center chips which are specially made for AI, too.  The new processors come in two variants: P-core (performance) and E-core (efficiency). The P-core series will be for the maximum power and will have additional integrated chiplets and will support 136 PCIe lands and up to 128 cores.

The E-core models aren’t to be neglected, either. Intel says they are made with size in mind and will require 67% fewer server racks to achieve the same computing power. This will make them more efficient, but suitable for hosting websites, running media, handling database calculations and some AI models. The result will be 4.2x performance per unit of rack space and 2.6x performance per watt. All Xeon 6 chips will support Compute Express Link (CXL) I, II and III, along with 12 channels of memory and more.

The P-cores will be available later in the third quarter, while the E-cores are already on the market. Intel hopes these chips will attract the attention of data center operators and will bring in revenue and make the competition with Nvidia a little more heated. How the Xeon 6 chips compare to Nvidia and AMD’s offerings on the pure hardware level, remains to be seen.

Amazon has something to say, too

AWS has been the most popular cloud service for a while, but the company isn’t relying on that for the future. It fully realizes the changes AI will bring and it has started to adopt them, all while making its own hardware, too.

The new AWS CEO, Adam Selipsky, has his hands full. The company is undergoing major transformations to adopt Gen AI, deploy Nvidia’s GH200 chips in data centers, and roll out its new Trainium AI processors as well. AWS is also investing in energy diversification for its data centers, including eyeing the options for more nuclear-powered facilities. And yes, all of this is to better accommodate the needs of AI in the data center.

It’s safe to say that the data center is indeed slowly becoming the AI data center. Everything will be “AI first, and the rest of the services second.” This is just the natural development of the segment, and it will bring some changes for other services. But apart from possible energy constraints and lack of enough colocation space, there probably won’t be much of a change in the overall picture. At least initially, as the layers will slowly shift and adapt to the new developments and opportunities.

Leave a Reply

Your email address will not be published.