CoreWeave’s Vera Rubin announcement is impressive, but it does not settle the bigger question hanging over the company: is this the next great AI infrastructure winner, or simply the most aggressive borrower in the room?

Being first to deploy NVIDIA’s Vera Rubin NVL72 system gives CoreWeave real credibility. This is not a generic cloud provider adding another product page. CoreWeave has built its identity around moving faster than traditional hyperscalers when new NVIDIA hardware arrives. For AI companies that care more about immediate access to cutting-edge compute than about buying everything from one giant cloud platform, that speed matters.

The market’s excitement is therefore understandable. CoreWeave is positioning itself as the specialist cloud for the AI era, and every successful launch deepens the perception that it is becoming an essential part of NVIDIA’s ecosystem.

But calling CoreWeave a classic picks-and-shovels business is too generous. NVIDIA sells the shovels. CoreWeave buys them, often with borrowed money, builds expensive infrastructure around them and then hopes customers rent enough capacity at high enough prices to justify the investment.

That model can be extraordinarily profitable when AI compute is scarce. It can also become painful very quickly if supply catches up, rental prices fall or customers shift workloads elsewhere.

The main risk is not whether demand for AI will disappear. It is whether CoreWeave can earn attractive returns after paying for chips, data centres, power, cooling, networking and interest. Revenue growth and a huge backlog look impressive, but they do not automatically translate into durable free cash flow. The company remains highly capital intensive, carries substantial debt and depends heavily on a relatively small number of very large customers.

Its close relationship with NVIDIA is also both an advantage and a vulnerability. Early access to new systems can help CoreWeave win business, but NVIDIA has every incentive to support multiple cloud partners. CoreWeave does not control the most valuable component in its own offering. It must keep spending to remain at the technological frontier, while much of the industry’s economic value may continue flowing to NVIDIA.

My view is that CoreWeave is a genuine technology and execution leader, but its stock is not a low-risk way to invest in AI infrastructure. It is a leveraged bet that demand for premium GPU capacity will remain strong enough, for long enough, to outrun depreciation, competition and financing costs.

Vera Rubin strengthens the bull case because it proves CoreWeave can keep delivering advanced systems quickly. It does not remove the financial risk. The company may become one of the defining infrastructure platforms of the AI boom, but it could also discover that being first to deploy every generation of hardware is an expensive race with no finish line.