Partner POV | Futuristic AI needs futuristic data centers, and it's just starting

This article was written by Steven Carlini, Vice President of Innovation and Data Center for Schneider Electric. 

One of the most intriguing movie scenes of all time, in my opinion, was when Harrison Ford interviews Sean Young in the original 1982 Blade Runner. Ford's character attempts to determine if Young's character is a replicant—an artificial being engineered to look exactly like a human. Young was very convincing as a human in her performance, but Ford caught on to some implanted memories of childhood.

Although I'm not a futuristic detective trying to identify robots posing as humans, many of my friends and family think of me as a "computer guy," but now I'm being portrayed as an "AI guy" who can somehow explain to them convincingly why they shouldn't be terrified by AI. Many are expressing heartfelt concerns about what AI means for them and their place in the world. The most common question I'm asked is: How do we stop AI from taking over? Well, everything? They're referring to "technological singularity," which is a hypothetical future point in time when AI becomes uncontrollable, resulting in unforeseeable consequences for human civilization.

My position is that we're at the beginning of a very long journey. AI isn't being developed to displace humanity. Instead, it's being developed to make the things we do in our personal lives and work lives more efficient and effective. The value and functionality of AI aren't going to mature overnight. It took the human brain over 3 million years to evolve to where it is now.

Explaining AI and LLMs

What I see happening broadly is that AI is being explained to the public by the technical programmers who developed it. For example, here's TechTarget's definition of LLMs: "An LLM is the evolution of the language model concept in AI that dramatically expands the data used for training and inference. In turn, it provides a massive increase in the capabilities of the AI model."

Although this is true and does sound advanced, it doesn't describe what an LLM can do today. Many people vastly overestimate the capabilities of AI. Right now, the capabilities of LLMs include the following:

• Searching and summarizing; in other words, helping people find needed content and presenting it concisely.
• Answering general questions, such as for customer support lines and general inquiries.
• Market research and competitor analysis.
• Predictive analytics.
• Copiloting and assisting in both personal and work functions.
• Workflow automation.
• Digital twinning.

Overall, LLMs are used to help people and businesses learn valuable information, predict the future and ultimately make good decisions—not exactly replacing people's role in society.

Crossing the chasm to more advanced AI

Where AI can take things is the more interesting topic, and it includes areas such as:

• Automated Healthcare: Personalized treatment plans, AI-assisted surgeries and even predictive healthcare models.
• Autonomous Manufacturing: Predicting what needs to be built, sourcing the material and managing the operations.
• Enhanced Virtual and Augmented Reality: Immersive experiences blurring the line between real and virtual worlds.
• Autonomous Transportation: Automobiles, planes and ships guided and operated by machines.

Advanced AI needs new IT

All of this AI automation won't magically happen. In the case of AI, IT infrastructure will need to be built to develop and train the LLM models and even more IT infrastructure will be needed to scale globally to support working versions of these LLMs closer to the user.

It's not efficient or fast to train AI LLMs in legacy data centers. The IT servers needed for advanced AI are larger and heavier. They have new components like GPU (graphics processing unit) accelerators, DPUs (data processor units) and CPUs (central processing unit) arranged in clusters, a grouping that could be hundreds of servers and thousands of GPUs. These clusters are highly dense, use quite a bit of energy and generate a large amount of heat. Overall, demand for AI computing to train LLMs is far outpacing the AI data center capacity that the industry is racing to build out. Building out as much AI capacity as possible is a trend that's expected to keep going strong for at least a few more years.

Moving LLMs closer to the user and putting them to work after they've been trained is called edge AI or inference, which is the process of running live data through a trained AI model to make a prediction, solve a task or generate content. For AI to transform industries and applications through process efficiency and task automation, sufficient data center capacity must be installed at the edge. Although these edge AI data centers will run compressed LLMs and there will be a variety of sizes and capabilities, they'll need to operate at the edge to deliver tangible benefits of lower data transfer volume and cost, higher speed and lower latency. This buildout of working models at the edge will follow the development of training models that are starting now and will run for many years.

The future of AI is exciting and is expected to provide many benefits. It won't happen overnight and not without high-powered data center and IT capacity. The AI journey is well underway, but realistically, we're at the beginning. I believe AI will advance to better serve society in our personal and business lives. Keep in mind that Blade Runner was made in 1982 and was supposed to be based in 2019. As far as I know, no robots are running around that are hard to distinguish from actual humans. Although AI progress seems like a sprint, in reality, it's a marathon.