From the NVIDIA edition of September 16, 2026
DSX Facilities Infrastructure Reference Design v2.0 for 250 MW-class AI factories
NVIDIA · AI factory facilities / NCP reference design · Global reference (NVIDIA Cloud Partner design guide)
- Publisher
- NVIDIA DSX Documentation
- Original publication
- August 19, 2026 (Design Guide v2.0 cited on page)
- Source retrieved
- 2026-09-17
What happened
NVIDIA's DSX Facilities Infrastructure Reference Design (Design Guide v2.0, August 19, 2026) documents a 250 MW-class example site—157 acres, 1,536 GPU racks, 110,592 GPUs—with 34.5 kV campus MV backbone, 45°C liquid-cooling design point with dry coolers (no evaporative water), N+1 CDUs, cabinet TDP from 198 kW (MGX Gen 1.1) to 330 kW (Vera Rubin NVL72), and cooling plant not UPS-backed on the GPU side.
Why it matters
Sets the true facility bar for NVIDIA AI-factory scale. Directly cautionary for SLED: most campuses and agencies cannot host 198–330 kW cabinets or 250 MW campuses. Use as a sizing/transferability boundary when partners pitch AI-factory language for education or government rooms.
Evidence and measured results
Docs page cites NVIDIA DSX Facilities Infrastructure Design Guide v2.0 (August 19, 2026); example table lists 250 MW-class IT load, 96 Scalable Units, 157-acre footprint, 240 MW compute IT+mechanical and 22.8 MW core loads; text states 45°C liquid-cooling design point, dry coolers with no evaporative water use, chillers on 4.16 kV, and cabinet TDP 198–330 kW.
Limitations and uncertainty
Reference design for NVIDIA Cloud Partners at hyperscale; not a measured SLED deployment. Supplemental figures via NVOnline. Actual sites vary.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-17; 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 hear AI-factory language without grasping megawatt, acreage, and liquid-cooling prerequisites.
- Stakeholders
- facilities VP
- CIO
- capital planning
- utility liaison
- Discovery questions
- What is peak rack kW and loop temperature available?
- Is evaporative cooling restricted by water policy?
- Who owns MV substation and BESS decisions?
- Value hypothesis
- A reference-aligned facility plan reduces stranded GPU purchases and schedule risk.
- Potential engagement
- Facility gap assessment against DSX Scalable Unit assumptions for a single hall or lab.
- Unsupported claims
- Campus can host 250 MW DSX without new capital
- Dry coolers always eliminate chillers
- UPS covers GPU cooling plant
Pre-sales engineering
Role takeaway
- Fit
- Fit as planning baseline for greenfield AI halls; poor fit as-is for legacy SLED machine rooms.
- Architecture and integration
- Co-design 34.5 kV backbone, CUB/CDU loops, CIN spine, and compute HACs as Scalable Units.
- Prerequisites
- Utility capacity study
- Liquid-cooling plant design
- Fiber distance plan ≤500 m
- Constraints
- 198–330 kW cabinet TDP
- GPU cooling not UPS-backed
- NCP-oriented scale
- Security
- Plan SMN/TAN separation; map to agency zoning separately.
- Proof of value
- Signed facility readiness checklist
- Pilot Scalable Unit heat-rejection test
- Documented UPS/backup scope
Delivery
Role takeaway
- Work
- Multi-year site/facility build with mechanical and electrical partners.
- Dependencies
- utility
- mechanical OEMs/CDU partners
- network plant
- Ownership
- Owner's facilities engineer with AI platform input.
- Skills and adoption
- IT ops must learn liquid-loop and MV coordination.
- Governance checkpoints
- Capital board approval
- Permitting (stormwater/NPDES)
- Factory acceptance of CDU N+1
- Acceptance criteria
- Measured loop temps at design point
- Documented rack power budget per SU
- Written deviation log from DSX guide
- Risks
- Schedule slip on long-lead MV gear
- Underestimating mechanical load
- Applying hyperscale design to undersized rooms
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?
Treat power (34.5 kV), dual-temperature facility water, N+1 CDUs, and CIN fiber ≤500 m as first-class design objects—not bolt-ons after GPU selection.
Governance
Who approves, reviews and stays accountable for outcomes?
Utility interconnect, BESS/generation permitting, and EPA/state stormwater notes imply multi-year capital governance beyond IT procurement.
Security and privacy
What data, permissions and controls need testing?
Network Core separates Tenant Access and Secure Management Network; page does not specify sovereign or FedRAMP controls.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
24/7 NOC/SOC/FOC staffing model assumes industrial ops workforce, not typical campus IT shifts.
Procurement
What should contracts, pricing and exit terms secure?
NCP reference and Marketplace CDU lists imply partner-constrained BOM; SLED RFPs should separate facility package from GPU SKU.
Operating model
Which teams own the service once it runs?
AI factory ops with dual-temperature plants and non-UPS GPU cooling require runbooks distinct from traditional HPC rooms.
Publication history
- 2026-09-16NVIDIA · Issue 119 resources
Stable resource ID: nvidia-dsx-facilities-infra-ref-design-v2-20260819