
New hardware, same AI gold rush
CoreWeave spent its Fully Connected AI conference flexing what looks like the next chapter of the AI infrastructure boom: Nvidia’s Vera Rubin NVL72 platform. The headline grabber? Cognition reportedly became the first customer to run production workloads on it, with token throughput up to 4.8× higher than Nvidia’s GB200 NVL72. That’s the kind of stat that makes AI builders sit up straight.
Cantor says the story still has legs
Cantor Fitzgerald’s Brett Knoblauch kept an Overweight rating on CRWV and a $176 price target, basically telling investors the gap between where CoreWeave is today and where it could be later is still pretty chunky. The analyst’s pitch: the company is adding software, expanding customer types, and building recurring revenue instead of just renting out shiny GPUs like a one-trick pony.
More than a GPU rental shop
CoreWeave also rolled out Forge, an integrated software platform with Free, Pro, and Enterprise tiers. Translation: it wants to be the place where you train, evaluate, and deploy models without duct-taping ten tools together like some late-night startup hackathon.
It also:
- added a partner network featuring CrowdStrike, VAST Data, and ClickHouse
- said it now has enterprise inference customers like Ennoble Care, Canva, and Capital One
- claimed it has 10x demand for every megawatt of capacity
- stayed confident it can hit its 8 GW target by 2030
Price hikes, bigger dreams
The company also said it raised prices by 25% in July and another 10% since then. That won’t really show up in 2026, but it could start to matter in 2027 and beyond — which is Wall Street code for “the margin story might get juicier later.”
Big picture: CoreWeave is trying to evolve from expensive AI plumbing into a more durable platform business. If that works, today’s infrastructure spend could look a lot less like a fad and a lot more like a moat.