NVIDIA and Microsoft widen Vera Rubin NVL72 Azure access as NVIDIA's post-earnings slide deepens

Vera Rubin is now running in Microsoft's cloud at meaningful scale. Announced at Ignite, the widened Azure preview gives customers VMs on NVL72 rack-scale systems aimed at long reasoning chains and agentic inference. GPU Cloud PCs extend the hardware to virtual desktops.
The NVL72 design connects 72 GPUs into a single NVLink domain, so very large models and long-context inference can run as if on one giant accelerator. That matters for reasoning models that generate long internal token streams. Earlier in the week at SC26, NVIDIA also introduced a Rubin HPC GPU for scientific simulation and expanded DGX Spark developer kits.
The financial picture is mixed. Q3 FY2027 revenue reached a record $115.84B, up 103%, with data center at $107.61B. Shares slid anyway and the drop deepened in the following days. Community threads cite a reported SEC probe into NVIDIA's accounting and a $2.7 trillion rout, and the slide coincided with a broader AI sell-off that erased the Nasdaq's 2026 gains around Anthropic's IPO.
The skeptical case is less about demand and more about sustainability. Investors are asking whether hyperscaler capex can keep growing, and Meta's net-zero reversal shows the power constraints. Watch for any confirmation of the SEC matter and Rubin's general-availability timeline on Azure.