From the NVIDIA edition of September 20, 2026
NVIDIA megawatts-to-tokens: Eos/SVP production Flex signals and Lambda MaxLPS densification
NVIDIA · AI factory infrastructure / power management · United States (Santa Clara SVP program; Lambda PoC; Manassas target)
- Publisher
- NVIDIA Blog
- Original publication
- September 15, 2026
- Source retrieved
- 2026-09-21
- Event date
- 2026-09-15
What happened
Sep 15, 2026 NVIDIA blog ties DSX factory co-design to two measured threads: (1) Emerald Conductor on NVIDIA’s Eos AI factory in Silicon Valley Power’s Flexible Load Interconnect Program—automated 4 MW→3 MW reduction with high-priority jobs continuing, claiming >200 successful demand signals; (2) Lambda’s first DSX MaxLPS validation on HGX B200—19 nodes in a 16-node full-power budget, ~4M→5M tokens/s (+24%) and +23% perf/W, with up-to-40% more Vera Rubin NVL72 capacity projected in suitable environments. Clarifies Santa Clara is Conductor/production flex proof rather than a dedicated DSX Flex install, with Manassas 96 MW as the first dedicated DSX Flex commercial deployment target. Also notes ~120 kW heat per GB200 NVL72 liquid-cooled rack as a cooling constraint.
Why it matters
Fresh Sep 15 primary consolidating production Flex signal counts with MaxLPS densification—useful for SLED facilities comparing interruptible-load programs vs densification software. Still vendor-authored; SVP/Eos and Lambda contexts differ from typical campus halls.
Evidence and measured results
SVP >200 signals; 4→3 MW automated drop; Lambda +24% tokens/s and +23% perf/W; 40% Rubin projection; Manassas 96 MW named; GB200 ~120 kW heat callout.
Limitations and uncertainty
Vendor blog; MaxLPS and Flex overlap prior edition themes but this page adds Eos/SVP production signal volume and explicit non-DSX-Flex labeling for Santa Clara; Rubin +40% is projection.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-21; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Fixed megawatts limit tokens; utilities resist firm AI loads.
- Stakeholders
- facilities
- AI platform
- utility liaison
- finance
- Discovery questions
- Is the bottleneck interconnect or stranded rack power?
- Can you auto-shed noncritical jobs?
- Do you meter rack draw?
- Value hypothesis
- Layered DSX-style flex + densification can raise useful work per MW when instrumented.
- Potential engagement
- Diagnostic: classify bottleneck, then one flex or MaxLPS pilot.
- Unsupported claims
- 40% more Rubin GPUs guaranteed
- Santa Clara equals DSX Flex GA
- Any site gets 24% tokens
Pre-sales engineering
Role takeaway
- Fit
- AI factories with mixed workloads and utility programs; weaker for tiny always-on clusters.
- Architecture and integration
- Conductor/DSX Flex path for grid signals; MaxLPS for intra-rack allocation; cooling design for ~120 kW/rack class heat.
- Prerequisites
- power telemetry
- priority classes
- liquid cooling readiness where applicable
- Constraints
- Vendor authorship
- Lambda PoC specifics
- Eos/SVP context
- Security
- Harden automation that sheds load without human click.
- Proof of value
- Signal success rate
- MW target hit
- tokens/s densification with SLO hold
Delivery
Role takeaway
- Work
- Implement and drill both control planes; document naming/scope for ops.
- Dependencies
- utility program
- NVIDIA DSX/MaxLPS access
- cooling vendors
- Ownership
- Facilities lead with platform SRE.
- Skills and adoption
- Cross-train power and inference ops.
- Governance checkpoints
- Label Conductor vs DSX Flex correctly
- Projection vs measured gates
- Fail-safe firm mode
- Acceptance criteria
- ≥agreed signal success
- No priority SLO breach
- Densification budget never exceeded
- Risks
- Naming confusion in contracts
- Cooling cliff
- Overfit to Lambda peaks
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?
Combine workload-priority flex orchestration with MaxLPS stranded-power reclaim; plan liquid-cooling heat rejection alongside power software.
Governance
Who approves, reviews and stays accountable for outcomes?
Separate “production Conductor flex” from “DSX Flex productized deploy” in buyer language to avoid scope confusion in RFPs.
Security and privacy
What data, permissions and controls need testing?
Automated demand response without operator touch increases need for secure automation and fail-safe firm mode.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Automated shedding may delay batch research; publish flex calendars to users.
Procurement
What should contracts, pricing and exit terms secure?
May inform utility tariffs and DSX-related BOM; separate software vs facility contracts.
Operating model
Which teams own the service once it runs?
Production AI factories become participants in utility flex programs with automated runbooks.
Publication history
- 2026-09-20NVIDIA · Issue 158 resources
Stable resource ID: nvidia-megawatts-to-tokens-dsx-20260915