Nvidia entered the cloud computing arena in 2023 with DGX Cloud, aiming to rival giants like AWS, Google Cloud, and Microsoft Azure by offering premium access to its high-performance AI GPUs bundled with specialized software. CEO Jensen Huang envisioned this as a full-stack AI platform that could generate $150 billion in revenue, surpassing AWS's annual sales, while hedging against hyperscalers developing custom chips to reduce reliance on Nvidia hardware. The strategy sought recurring high-margin revenue and greater ecosystem control through tools.
Technical and operational hurdles quickly emerged as major setbacks. Running infrastructure (across diverse data centers from AWS, Google, and Microsoft proved nearly impossible to troubleshoot, with fixes incompatible across environments, leading to persistent performance issues. Pricing exacerbated the problem: DGX Cloud charged $36,999 monthly for H100 instances at launch, far exceeding rivals* who slashed their own Nvidia GPU prices by up to 45%, deterring potential customers despite early promotions to firms like ServiceNow and SAP.
Strategic realities forced Nvidia's retreat, as expanding the service risked alienating its top customers; AWS, Google, and Microsoft—which account for about half of its GPU revenue. Huang grew reluctant to push forward amid fears of backlash, prompting a team restructure that shifted DGX Cloud primarily to internal use for chip design, AI model development, and research. The newer DGX Cloud Lepton marketplace also faltered and integrated into engineering, signaling a full pivot away from direct cloud competition.
This episode underscores a classic misadventure: Nvidia discovered the harsh realities of cloud operations through painful experience, choosing not to let it erode focus from its dominant core of AI chip innovation. Far from derailing hardware supremacy, the pullback reinforces Nvidia's strengths amid booming data center demand.
NVIDIA'S CLOUD FLIGHT: SMART REALIGNMENT OR LESSON IN STAYING IN YOUR LANE?