NVIDIA made DGX Cloud broadly available on 8 August 2023, turning access to a dedicated slice of its AI supercomputing stack into a monthly subscription. The offer bundled high-end GPU capacity with NVIDIA's enterprise AI software, starting on Oracle Cloud infrastructure rather than requiring customers to build the whole machine room themselves.
NVIDIA Watch analysis
What happened
The company described each DGX Cloud instance as eight 80GB Tensor Core GPUs with 640GB of GPU memory, priced from $36,999 per instance per month. It presented the service as dedicated training capacity for generative AI and other demanding workloads, supported by NVIDIA's own specialists and software stack.
Oracle Cloud was the first infrastructure provider named for general availability. NVIDIA said Microsoft Azure support was expected in the following quarter and Google Cloud support later. Those dates were forward-looking availability plans in NVIDIA's release, not proof that every region or configuration was ready on announcement day.
Why it matters
The strategic shift was bigger than another GPU launch. DGX Cloud let NVIDIA sell ongoing access to an integrated computing environment, moving part of the buying decision from capital equipment towards a recurring service. For teams blocked by procurement lead times, that promised a faster route to serious training capacity.
The trade-off was control. A bundled service can simplify setup and support, but it also concentrates hardware, software and operating choices with one supplier. The launch price was clear enough to start an honest budget conversation, although the cited announcement supplied no independent comparison of utilisation, performance per pound or long-term ownership cost.
Our read
This was NVIDIA putting a meter on the supercomputer and offering to mind the boiler. That may be exactly what a team needs when model work is waiting on infrastructure, but convenience is not the same as cheapness. The interesting test is whether customers gained useful weeks of progress rather than merely acquiring a splendidly expensive monthly direct debit.
What to watch
- Whether the promised Azure and Google Cloud availability arrived with comparable regions, capacity and service terms.
- How consistently customers could secure capacity during periods of heavy demand for high-end accelerators.
- Whether portability between DGX Cloud and customers' existing cloud or on-premises environments proved practical.
- Which teams could justify the monthly price through faster delivery, higher utilisation or reduced infrastructure work.
Discussion spark: Would you pay a premium for managed access if it removed months of infrastructure delay, or would the tighter supplier dependency stop you?
Sources and evidence
- NVIDIA DGX Cloud Now Broadly Available, Unlocking Instant Access to AI Supercomputing (8 August 2023, 13:00 UTC)
- NVIDIA Newsroom RSS record for the DGX Cloud announcement (8 August 2023, 13:00 UTC)
Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.