CoreWeave'e AI Cloud: From Infrastructure to the AI Loop

I'm at CoreWeave's Fully Connected 2026 in San Francisco this week, the company's first-ever global user conference, and I sat down with Susanne Seitinger, VP of Product Marketing at CoreWeave, and my colleague Holger Mueller to talk about the company's big announcement: Forge.

Forge is CoreWeave's opinionated statement on what Susanne calls the "AI loop," the full set of steps it takes to make AI genuinely useful inside an organization. It's built on what CoreWeave calls metal-to-model thinking: everything connects back to the hardware layer so you can optimize across the entire stack, not just the parts of it you can see from an application dashboard.

What struck me most is that this isn't a one-time delivery. As Susanne put it, the model delivery lifecycle is "never a one-time shot, it's continuous, a virtuous, infinite loop." You get value earlier in that loop because you aren't stuck waiting for a signal on whether something is actually working. And the variable most teams assume is driving change, the data, is only part of the story. The infrastructure underneath it, the GPUs, changes constantly too, with real cost implications for what you can afford to run and how.

The only way to know if any of it is working, Susanne told us, is to talk to your customers directly and bring that signal back into the loop. That's exactly what we saw happening on stage with customers like Caterpillar and Capital One. Caterpillar is thinking about its business over the next hundred years, and that changes what kind of cloud partner it needs. This isn't about shaving costs. It's about scaling up and building a long-term partnership around much bigger objectives.

That takes me to the five themes I keep coming back to in nearly every enterprise AI strategy conversation right now:

1. Margin compression. Most of the Fortune 500 are only growing in single digits. There's no room left to find that growth through cost-cutting alone.

2. Exponential scale. Nothing's interesting anymore unless you're doing it 10x, 100x, or 1000x better. Incremental improvement doesn't move the needle.

3. A mindset shift. This isn't about cost savings anymore. It's about expansion as a core strategy.

4. Infinite possibilities through cycle expansion. This is the one people underestimate. If you can get 2x, 5x, or 10x more "at-bats" than your competitor, more reps, more iterations, more chances to learn, that's not a cost story. That's a transformation story.

5. Strategy and execution becoming one thing. With infinite possibilities on the table, you don't have time to mess up. It's a game of velocity, not speed.

   ---

Holger's favorite part of the announcement was the backward compatibility CoreWeave has built in: the ability to run everything from the newest Rubin architecture back to Volta from 2017, six full GPU generations, working in concert through a standard operating model CoreWeave calls Mission Control. That consistency is what actually lets you scale. As Holger pointed out, some things have to stay consistent, or scaling simply isn't possible.

That consistency is also what makes a concept like infinite computing practical rather than theoretical. Nothing is truly infinite, Holger quickly noted, but for the practical purpose of running your business, you don't need to know exactly how much capacity exists. You use what helps your business, what you can afford, what accelerates you, and it becomes practically infinite. That reframes how you design and automate processes from the ground up: you stop asking for ten GPUs or fifty, and start asking for as many as it takes to transform the business.

The through-line across the entire conversation was partnership and composability. CoreWeave is deliberately building openness into Forge, including a partner network and search capabilities through Query Forge, because agents need to talk to the rest of the world to be genuinely useful. And the roadmap itself comes directly from customers. As Susanne said, "we built Arena because we heard that's what they need." That kind of direct signal from the field is exactly what the next phase of enterprise AI strategy has to be built on.

My takeaway: the future of cloud isn't a single switch from "build" to "buy." It's a spectrum from turnkey to build-to-suit, and for the first time, one platform shows the whole spectrum. That's the opportunity in front of enterprises right now, and it's why conversations like this one matter more than ever.