{"resourceId":"nvidia-dsx-maxlps-technical-20260824","versions":[{"version":"external-bd78677776292829a4c3e07868808cd2c9cf4fedb6d68e081ad1e50319a6a8fa","resource":{"id":"nvidia-dsx-maxlps-technical-20260824","title":"DSX MaxLPS technical deep dive: DPS Developer Preview and up-to-40% Rubin projection","organization":"NVIDIA","sector":"AI factory software / facilities engineering","geography":"Global (vendor technical blog)","publishedAt":"August 24, 2026","publicationDate":"2026-08-24","eventDate":"2026-08-24","sourceName":"NVIDIA Technical Blog","sourceLabel":"Vendor technical guidance","sourceUrl":"https://developer.nvidia.com/blog/maximizing-ai-factory-performance-per-watt-with-nvidia-dsx-maxlps/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["infrastructure","operating-model","governance-procurement"],"finding":"Aug 24, 2026 NVIDIA technical blog defines MaxLPS as Maximum Land Power Shell—dynamic power allocation via Dynamic Power Software (Developer Preview), workload-profile power solutions/APPM, and 45°C liquid-cooling inlet design. An illustrative 100 MW waterfall leaves ~60 MW for AI load after facility/rack/ops deductions. Vendor examples reclaim 170 kW stranded in a 540 kW static plan and show GB200/GB300 rack power cuts enabling roughly 35–39% more racks at preserved throughput; the post projects up to 40% more Rubin GPUs in a 100 MW factory. DPS does not change the site power envelope.","sledRelevance":"Newly relevant alongside today’s Lambda PoC: explains control-loop prerequisites and that Rubin “+40%” is a planning projection requiring early facilities design for 45°C and rack-position optionality—not a drop-in software toggle for SLED halls.","evidence":"DPS Developer Preview continuous control loop; DSX Exchange optional IT/OT bus also Developer Preview; Figure 4 GB200 125→90 kW and GB300-class 136→101 kW examples; Table 1 100 MW factory projections; guidance to validate before infrastructure is fixed; scoped NVL72 inference power pilot recommended.","architectureImplications":"Site must be designed for MaxLPS rack-position optionality, east-west network sized to MaxLPS GPU count, and warm-water cooling; Day-1 population can be lower than the lifecycle target.","governanceImplications":"Treat 40% Rubin capacity and zettaflop tables as planning estimates pending measured site pilots; separate software validation from irreversible facilities decisions.","securityPrivacyImplications":"IT/OT event bus (DSX Exchange) expands attack surface to building systems if adopted; still in Developer Preview.","caveats":"Vendor-authored; DPS/Exchange in Developer Preview; Rubin capacity gains are projections; illustrative waterfalls are not a specific customer site audit.","streamIds":["nvidia"],"roles":{"sales":"Position MaxLPS as a facilities-plus-software operating model, not a SKU discount. Lead with the buyer’s fixed land/power/shell constraints, then explain DPS reclaiming stranded rack headroom without raising the utility interconnect. Explicitly label the up-to-40% Rubin figure as a projection requiring 45°C design readiness—and pair it with any measured PoC rather than selling the projection table alone.","engineering":"Fit for teams planning Vera Rubin or densifying Blackwell under a hard megawatt cap. Integrate DPS topology models (utility→rack→GPU), workload profiles, and optional DSX Exchange. Prerequisites: telemetry, policy engine, and thermal design for 45°C inlet. Constraints: Developer Preview software; proof of value is a static-vs-managed pilot tracking throughput, latency, errors, power, and policy compliance.","delivery":"Delivery work starts before concrete is poured: validate power topology, cooling, telemetry, and rack-position optionality. Ownership: facilities engineering with platform software. Skills: power-policy ops and CDU/TCS loop literacy. Risks: Day-1 training-heavy mix blocks densification; chillers still needed in heat waves; preview software maturity."},"roleDetails":{"sales":{"customerProblem":"Utility interconnect and shell limits block AI expansion more than chip supply.","stakeholders":["facilities","CIO/CTO","procurement","utility"],"discoveryQuestions":["Is the site designed for 45°C DLC inlet?","Can you leave rack positions dark for later fill?","Is DPS on the software roadmap?"],"valueHypothesis":"Early MaxLPS-ready design can raise lifecycle GPU count inside a fixed MW without promising immediate +40%.","potentialEngagement":"Site MaxLPS readiness review plus scoped software pilot.","unsupportedClaims":["40% more Rubin GPUs guaranteed","DPS GA everywhere","Cooling power always reclaimed"]},"engineering":{"fit":"Fits greenfield AI factories and major retrofits; poor fit for unmanaged racks without telemetry.","architectureIntegration":"DPS groups plus WPPS/APPM profiles; size network for MaxLPS GPU count; optional DSX Exchange to BMS.","prerequisites":["power telemetry","policy engine","thermal design validation","space for optional racks"],"constraints":["Developer Preview DPS/Exchange","workload-mix dependent Day-1 fill","climate-specific free cooling"],"security":"Harden any IT/OT bus; least-privilege power policy changes.","proofOfValue":["static vs managed pilot metrics","policy compliance logs","thermal inlet targets met"]},"delivery":{"work":"Parallel facilities design gate and software pilot; do not wait for full Day-1 population.","dependencies":["mechanical/electrical engineer","NVIDIA DSX SE","platform SRE"],"ownership":"Facilities program lead with platform co-owner.","skillsAdoption":"Train ops on power policies and TCS loop behavior.","governanceCheckpoints":["Design freeze only after MaxLPS optionality review","No +40% in budget without pilot","Preview software risk accepted"],"acceptanceCriteria":["Site design checklist signed","Pilot metrics filed","Rollback power policy documented"],"risks":["Retrofit cost exceeds reclaim","Preview software delays","Training mix prevents densification"]}},"retrievedAt":"2026-09-19T03:12:30Z","enrichedAt":"2026-09-19T03:17:22Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"None stated; denser factories may intensify facilities staffing needs.","procurementImplications":"Separate facilities-readiness BOM from GPU purchase; require Developer Preview risk language.","operatingModelImplications":"Capacity becomes a lifecycle fill strategy tied to workload-mix evolution.","sourceVerification":{"openedUrl":"https://developer.nvidia.com/blog/maximizing-ai-factory-performance-per-watt-with-nvidia-dsx-maxlps/","referenceExcerpt":"NVIDIA projects that MaxLPS combined with data center power planning can enable up to 40% more Rubin GPU capacity","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}