Most people overestimate what AI can do today. But they completely underestimate what it will be able to do in a year or two.

When you move at this incredible pace, every small improvement in technology compounds.

Just as a thought experiment, if you get better by 1% every day, you would be about 37x better by the end of the year.

Now imagine if you are not improving by just 1% but by much more.

So far, AI has mostly been limited to text. For most people, AI means chatbots like ChatGPT. You type something, it answers. But soon, we will move beyond natural language processing and LLMs. AI will become agentic and multimodal.

What comes next is much bigger.

AI will move beyond just understanding words. It will see images, hear sounds, watch videos, and understand the real world. This is what multimodal means. One system that can read, listen, watch, and respond all at once.

At the same time, AI will become agentic.

That means it will not just answer questions. It will take actions. It will plan steps, make decisions, use tools, and work toward goals on its own. Instead of asking AI to write an email, you will ask it to handle a task. It will decide what to do, when to do it, and how to do it.

AI will be able to call other AI agents to help. It will work with images and videos. And most importantly, it will handle real world tasks.

2026 is likely the year we see a real breakthrough moment. Think ChatGPT, but for physical AI. Robotaxis will likely roll out at scale, with humanoid robots following a year or two later.

Nvidia just released a full new robotics stack. That includes robot foundation models, simulation tools, and specialized hardware. It wants to be a full stack robotics provider or the default platform for robotics, just like Windows became the default for PCs or Android for smartphones.

These are multi-trillion markets emerging from 0.

All of these require exponentially more compute: Text → Reasoning → Agents → Images → Videos → Physics → Real-word Applications

There is basically no limit for how much processing power we can use. Every jump in compute unlocks new things we simply could not do before.

To support this shift, the world needs AI factories. These are data centers designed not to store information, but to produce intelligence at massive scale. They require new breakthroughs in compute, memory, efficiency, and system design.

There is practically no limit to how much intelligence we can use.

New large-scale data center projects are being announced almost every week. And outside of the hyperscalers, only a handful of companies can actually deliver compute at this scale.

Right now, 5 neocloud and HPC players clearly stand out:

HUT8 $HUT

Nebius $NBIS

Coreweave $CRWV

Applied Digital $APLD

IREN $IREN

AI is one of the rare platform technologies. It does not just create new products. It makes almost every other technology better at the same time.

Most people still underestimate what AI can do, but it’s improving so quickly that what seems impressive today will feel normal in a few months. And markets often underprice these second order effects at first, then reprice them quickly once people understand.

This is the new industrial revolution.

If you have not yet explored make sure the checkout the AI Infrastructure portfolio, it is worth taking a look.