OpenAI’s Stargate is getting bigger, and that matters more than you think

OpenAI just announced they’re scaling up Stargate again. If you’ve been following the compute arms race in AI, this isn’t exactly a surprise, but the scale of what they’re doing is worth paying attention to.

Stargate is their big bet on compute infrastructure for AGI. Not the kind of infrastructure where you spin up a few GPU instances in the cloud and call it a day. I’m talking about purpose-built data centers designed from the ground up to run the kind of models that make today’s largest training runs look like homework assignments.

The announcement is short on specifics, but the direction is clear: more capacity, more quickly. They’re adding data center capacity to meet what they describe as “growing AI demand.” That phrase is doing a lot of heavy lifting, because what they really mean is that training and inference at AGI scale requires hardware that most organizations can’t even imagine.

This is higher than I expected. A few years ago, the conventional wisdom was that we’d hit diminishing returns on compute scaling. Instead, we’re seeing the opposite: the ceiling keeps moving. Stargate isn’t just about building bigger clusters, it’s about building them in a way that actually works at that scale. Power delivery, cooling, networking, reliability at tens of thousands of accelerators, these are hard engineering problems that most people never think about.

I’ve seen this pattern before in other industries. When hyperscalers build their own infrastructure, they stop being constrained by what’s available and start defining what’s possible. OpenAI is doing the same thing here. They’re not waiting for the cloud providers to catch up, they’re building their own path.

There’s a practical angle too. If you’re building on top of OpenAI’s API, this means more capacity for inference, lower latency, and fewer bottlenecks. The days of getting rate-limited because everyone else is also trying to run the same model might start to fade. But don’t hold your breath, demand is growing faster than supply, even with these expansions.

The flip side is that this kind of infrastructure spending is eye-wateringly expensive. We’re talking tens of billions of dollars over the next few years. OpenAI is betting that AGI will generate enough value to justify the cost. If they’re right, it’s the best investment in history. If they’re wrong, it’s a very expensive lesson.

I’d like to see more transparency around the actual architecture and power consumption of these new data centers. The environmental impact alone is worth discussing, but that’s a topic for another post.

For now, the takeaway is simple: the compute infrastructure race is real, it’s accelerating, and Stargate is one of the biggest bets in the game. Whether you think AGI is five years away or fifty, the hardware being built today will shape what’s possible tomorrow.