To The Moon

GPT-6 Astra is another reminder that we’re still very early in the AI cycle.

OpenAI says Astra was built using more than 100,000 GPUs at its Texas facility, making it one of the largest training runs the company has ever done.

And the bigger story isn’t just how capable the model is.

It’s what happens when models keep getting this much smarter.

Astra is already pushing into work that would have sounded unrealistic just a few years ago — complex coding, computer use, scientific research, cybersecurity and multi-step professional workflows. OpenAI says it reaches state-of-the-art results across several of these areas.

But intelligence at this level comes with a massive requirement:

Compute.

Every generation raises the amount of infrastructure needed to train and run these systems. GPUs, data centers, power, cooling and networking all have to scale with it.

And that infrastructure cannot be built overnight.

That’s where I think the market may still be underestimating the opportunity.

The gap between what most people think AI can do and what frontier models are actually capable of is getting wider — not smaller.

If model capability keeps accelerating while physical infrastructure takes years to catch up, compute could remain constrained for a long time.

That makes the AI infrastructure trade much bigger than just the next GPU cycle.

We may still be very early. 👀