In our lifetime, few technologies have significantly transformed the way we live and work than Artificial Intelligence (AI). And even fewer technologies have evolved as fast as AI. Indeed, in the last decade alone, AI has emerged from impressive but niche applications to transformative tools that have impacted multiple industries.

Let’s take a quick journey over the last 10 years and I’ll share some thoughts on where AI will go from here.

While the term AI has been in our vocabulary longer than a decade, I’ve chosen a 10-year span because 2015 is when OpenAI was founded. But it wasn’t until 2017 that Google released the transformer architecture that laid the groundwork for modern natural language processors like GPT and BERT (Bidirectional Encoder Representations from Transformers), which Google released in 2018. BERT was a key development in AI’s journey because it revolutionized Natural Language Processing (NLP) by enabling contextual word understanding.

Perceptive AI
By 2019, OpenAI unveiled GPT-2 but did not release it due to ethical concerns. Up to this point, the context of AI was perceptive. Meaning, the Large Language Models (LLM) could ingest different types of content (text, images, sounds) and extract its meaning. We started to see more intuitive chatbots, autonomous cars, robots that interacted in a much more natural way, and it began to be used to help with medical diagnosis and treatment.

Generative AI
GPT-3 followed in 2020, its 175 billion parameters capable of generating human-like text. Writers could enter a few prompts and complete articles could be written – to the trained eye, it was clear that it was artificially generated, but we could tell it would get better over time.

In 2021, DALL-E was released and we all played around with creating AI-generated images from text prompts. Through 2022 and 2024, new competitors to OpenAI emerged, notably Google’s Gemini and Anthropic’s Claude. Everyday web searches started to produce AI-generated results.

Agentic AI
By 2024, we saw the emergence of Agentic AI. Agentic AI is the term used when AI does more than just ingest and spit out results. Now AI has agency: it can understand, reason, plan, and take action. It uses tools and it learns. AI Assistants can book travel arrangements, set up meetings, and conduct research.

Physical AI
With Physical AI, we start to see the combination of AI with robotics and real-world interactions. AI will enable robots to physically manipulate and respond to their own environment. In 2025, physical AI is already in use in simulated warehouses and in manufacturing and logistics applications – it’s easy to see how human-like robots can be used in the home.

It is important to note that each of these transformations does not mean the former stage is now obsolete. Rather, AI developers are continuing to expand the capabilities for every type of AI application. For example, in Agentic AI, one direction could be that instead of waiting for prompts or commands, AI will be capable of setting long term objectives, much like human strategists do today, and then setting into motion the tasks to accomplish them.

AI’s Impact on Supercomputing Infrastructure

Two things come to mind as I look at the trajectory of where AI can take us.

One is that the data center infrastructure industry desperately needs to keep up with exponential growth in data creation demand in an efficient and sustainable way. The amount of energy needed to process the data that these AI types use far outweighs the forecast of energy supply. Innovation in this area and in cooling technologies will be key.

Secondly, as we rely more and more on autonomous modes – whether it be self-driving cars making navigation and breaking decisions or humanoid robots in our home helping with chores and childcare, the data processing needs to be close to where the output is being used. Here, I’m talking about Edge and how modular data centers are ideal for this.

In fact, it is not inconceivable that each household in the future contains its own single rack data center that drives the data processing needed in the home, but is also connected to other data centers for longer term data processing and analysis.

What’s Next for AI?

In the area of Perceptive AI, the models could expand to include touch and smell and then combine with deep learning for logical reasoning.

One criticism of AI today is that the results generated can be considered to be biased. Much work needs to be done to ensure that the AI’s training sets are bias free, that its results are not misused (deepfakes, misinformation), and that it is transparent so that its decision-making is explainable.

Decentralized AI agents could collaborate across various domains for collective intelligence.

Humanoid robots could use AI to provide home care, emergency response, and of course, household chores.

We are also moving toward Artificial General Intelligence (AGI) where AI can teach itself new skills without supervision and continually evolves itself based on past experiences. Lastly, medical science has long been exploring brain-computer interfaces where our brain’s neural network interacts directly with AI.

If the last decade is any indicator, the next decade may see these advances and many more.