From the NVIDIA edition of October 14, 2026
DGX Cloud Create docs: Run:ai managed limits — no SOC 2/HIPAA, H100-only, no GPU splitting
NVIDIA · Technology vendor · Global (AWS/GCP managed service)
- Publisher
- NVIDIA DGX Cloud Create Documentation
- Original publication
- undated documentation page (retrieved 2026-10-14); why now: day after Run:ai open-source release
- Source retrieved
- 2026-10-15
What happened
NVIDIA's DGX Cloud Create limitations page states the managed Run:ai offering is restricted: customer workloads cannot use privileged containers, host networking, HostPath mounts and related Kubernetes features; customer admins cannot create/edit clusters, nodes, node pools or Run:ai control-plane settings. Only H100 GPUs on AWS and GCP are supported (not GB200 NVL72). NVCF, NVIDIA Blueprints, multi-node inference, GPU splitting/partial GPUs and mixed GPU types in one cluster are unsupported. The cluster is not SOC 2 Type 2 nor HIPAA compliant. Why now: day after the Run:ai open-source release, buyers comparing community self-host vs managed DGX Cloud Create need these constraints.
Why it matters
High for universities and agencies evaluating managed GPU clouds for research or pilots with regulated data.
Evidence and measured results
Vendor documentation of product limits; page undated in retrieved HTML; applies to managed service, not all Run:ai deployments.
Limitations and uncertainty
Docs page undated; limits may change; distinct from self-hosted open-source Run:ai.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-15; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Buyers need clear limits of managed GPU cloud before placing research or pilot workloads with regulated data.
- Stakeholders
- Research computing director
- CISO / compliance
- Cloud procurement
- PI / data steward
- Discovery questions
- Does any workload involve PHI, CUI or other regulated data?
- Do you need GPU fractions or MIG?
- Is GB200 or multi-node inference required?
- Value hypothesis
- Interpretation: the managed service can speed non-regulated H100 research if limits are acceptable; it cannot satisfy HIPAA/SOC 2 needs per this page.
- Potential engagement
- Fit assessment comparing DGX Cloud Create limits to self-hosted open Run:ai for one research workload class.
- Unsupported claims
- HIPAA or SOC 2 compliance
- GB200 support
- GPU splitting on this SKU
- That limits match self-hosted Run:ai
Pre-sales engineering
Role takeaway
- Fit
- Moderate for non-regulated H100 research on AWS/GCP; poor for HIPAA/PHI, Blueprints, NVCF, partial GPUs or GB200.
- Architecture and integration
- Managed Run:ai on DGX Cloud Create with restricted Kubernetes capabilities and customer admin roles.
- Prerequisites
- AWS or GCP tenancy via NVIDIA
- H100-compatible workloads
- TAM channel for cluster changes
- Constraints
- Not SOC 2 Type 2 or HIPAA
- H100 only (no GB200 NVL72)
- No GPU splitting/partial GPUs
- No Blueprints/NVCF/multi-node inference
- Security
- Platform blocks high-risk Kubernetes features; compliance certifications for the cluster are explicitly absent for SOC 2 Type 2 and HIPAA.
- Proof of value
- Classify data before onboarding
- Test a non-regulated workload end-to-end
- Confirm Nsight/profiling needs against documented limits
Delivery
Role takeaway
- Work
- Run a compliance and feature-fit gate, then onboard only eligible workloads with documented exceptions.
- Dependencies
- NVIDIA TAM
- Institutional compliance office
- Cloud networking/identity
- Ownership
- Research computing with CISO sign-off for data classes.
- Skills and adoption
- Users adapt to reduced admin powers; ops rely on TAM for cluster changes.
- Governance checkpoints
- Data-class approval
- Feature-fit sign-off
- Exit criteria if limits block science
- Acceptance criteria
- Proposed: written HIPAA/SOC 2 disqualifier acknowledgment
- Proposed: pilot workload completes without needing blocked features
- Risks
- Regulated data placed on non-HIPAA cluster
- Science blocked by missing fractions/Blueprints
- Confusion with open-source Run:ai capabilities
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Managed multi-tenant restrictions; no MIG/partial GPU; H100-only today.
Governance
Who approves, reviews and stays accountable for outcomes?
Customer System Administrator capabilities removed; TAM-mediated changes.
Security and privacy
What data, permissions and controls need testing?
Hardened Kubernetes policies; explicitly not SOC 2 Type 2 or HIPAA—block PHI/regulated workloads.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Reduced admin surface may ease ops but blocks advanced profiling (Nsight limits).
Procurement
What should contracts, pricing and exit terms secure?
Compliance buyers must not assume HIPAA/SOC 2 from this SKU.
Operating model
Which teams own the service once it runs?
Managed service tradeoff: less control, NVIDIA-operated plane.
Publication history
- 2026-10-14NVIDIA · Issue 397 resources
Stable resource ID: nvidia-dgx-cloud-create-runai-limitations