{"resourceId":"itdaily-netapp-novus-100-tbs-figures-from-netapp-and-omdia-projections-20260929","versions":[{"version":"external-5848fc2347b2e5620b24981e866cf879dd8375bf403605827b7945d13e483f1c","resource":{"id":"itdaily-netapp-novus-100-tbs-figures-from-netapp-and-omdia-projections-20260929","title":"ITdaily: Novus 100 TB/s figures come largely from NetApp and Omdia projections; practice has yet to prove them","organization":"ITdaily","sector":"Technology trade press","geography":"Europe (Belgium)","publishedAt":"29 September 2026","publicationDate":"2026-09-29","eventDate":null,"sourceName":"ITdaily","sourceLabel":"Jonkers, \"NetApp Novus keeps hundreds of thousands of GPUs well-fed,\" ITdaily, 29 Sep 2026","sourceUrl":"https://itdaily.com/news/cloud/netapp-novus/","evidenceClass":"independent-reporting","outcomeClass":"mixed","topics":["infrastructure","governance-procurement"],"finding":"ITdaily (29 September, newly relevant now that the PEAK:AIO deal has closed and 8 November coverage repeats NetApp's superlatives) reports NetApp's claim that Novus exceeds 100 TB/s and serves hundreds of thousands of GPUs. It says Omdia observed near-linear scaling as ONTAP clusters were added to a single namespace and, on that basis, expects up to 100 TB/s sequential read and tens of exabytes. The reporter concludes the throughput and scale figures come largely from NetApp and Omdia's projections, so practice has yet to prove them. NetApp says the first release uses Supermicro infrastructure with AFF A90 data nodes and is orderable, with software-defined deployment later.","sledRelevance":"Limited direct SLED relevance: Novus targets neoclouds and GPU-as-a-service providers. It is a caution for university research centers and public agencies that may see 100 TB/s claims in vendor pitches. Not deployment evidence.","evidence":"Opened the full article. Separates NetApp claims, Omdia projection and the reporter's caution; no independent measured benchmark, customer or SLED example is provided. Omdia's test scale is not stated.","architectureImplications":"Metadata is separated from the data path using standard NFS; a software-defined option is described as future, so current deployments depend on the specific hardware configuration.","governanceImplications":"Require audited test conditions and sustained results in any bid that cites 100 TB/s.","securityPrivacyImplications":"The article says little about security; multi-tenant and security features are asserted by NetApp without testing detail.","caveats":"Single trade-press report; the test scale and workload behind Omdia's projection are not given; the figure is a projection of sequential reads, not a deployed or mixed-workload result; predates the acquisition close.","streamIds":["netapp"],"roles":{"sales":"The point is to keep the claim honest. NetApp says Novus exceeds 100 TB/s, but ITdaily notes the figures come largely from NetApp and Omdia projections that practice has yet to prove. For a university or agency asking about AI storage, position NetApp credibly: validated ONTAP data services, plus a new architecture still being proven. Ask what throughput and metadata behavior the customer needs. Do not quote 100 TB/s as an achieved result or imply any SLED site runs Novus.","engineering":"Use this as a reminder to test, not to size. The 100 TB/s figure is an Omdia projection of sequential reads after a scaling test of unstated size. SLED research workloads often stress small-file metadata, checkpoints and mixed tenants, which sequential bandwidth does not cover. Ask NetApp for the audited node and client counts, file-size mix and sustained results, plus the client kernel and NIC support matrix. Treat Supermicro-plus-AFF A90 as the only described configuration until others are published.","delivery":"Delivery should plan proof before commitment. Work: define a representative workload test, specify acceptance on sustained throughput and metadata operations, and agree the client image. Owners are research computing and storage operations. Acceptance criteria below are proposals, not observed results. The main risks are buying on projected figures, undefined failure behavior across the metadata and data tiers, and a software-defined option that is not yet available."},"roleDetails":{"sales":{"customerProblem":"Buyers meet headline throughput claims without independent evidence at their scale.","stakeholders":["Research computing director","CIO or CTO","Procurement lead","Storage manager"],"discoveryQuestions":["What throughput and metadata rates do your AI jobs actually need?","What evidence will you require before accepting a vendor throughput claim?"],"valueHypothesis":"Evidence-based testing may reduce procurement risk; this source does not measure any customer benefit.","potentialEngagement":"A bounded proof-of-value test plan for an AI or research storage purchase.","unsupportedClaims":["Novus achieves 100 TB/s in production","Any SLED customer benefit","Superiority over other vendors","Guaranteed GPU utilization gains"]},"engineering":{"fit":"Neocloud and very large AI file workloads; limited fit for typical SLED sizes.","architectureIntegration":"Novus Data Director on qualified Supermicro servers with AFF A90 data nodes; standard NFS client access per NetApp.","prerequisites":["Client kernel, NIC and NFS version matrix","Representative dataset and access pattern","Network design for the storage fabric"],"constraints":["Only one configuration described","Software-defined deployment is future","Projected figure covers sequential reads only"],"security":"Multi-tenant and security features are vendor-asserted; verify identity, export policy and isolation on both tiers.","proofOfValue":["Sustained metadata-heavy small-file test","Failure test of metadata and data tiers separately","Tenant isolation check"]},"delivery":{"work":"Define workloads, run the proof test, and decide on evidence.","dependencies":["NetApp test environment access","Client images and network design"],"ownership":"Research computing with storage operations.","skillsAdoption":"Staff need pNFS client tuning and a runbook covering two failure domains.","governanceCheckpoints":["Test plan approval","Go/no-go review on measured results"],"acceptanceCriteria":["Sustained throughput and metadata rates meet agreed thresholds on the customer workload (proposal)","Recovery from single-tier failure without client remount (proposal)"],"risks":["Buying on projected figures","Unclear failure behavior across tiers","Software-defined option not yet available"]}},"retrievedAt":"2026-11-08T22:07:00-06:00","enrichedAt":"2026-11-08T22:25:00-06:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"No evidence on skills or staffing.","procurementImplications":"Treat 100 TB/s as a design target when evaluating bids; ask for audited node count, client count, file-size mix and sustained throughput.","operatingModelImplications":"Separate metadata and data tiers would add operational components; the article does not discuss operations.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://itdaily.com/news/cloud/netapp-novus/","referenceExcerpt":"so practice has yet to prove whether Novus delivers on these promises.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}