NVIDIA Inference Context Memory Storage Platform (now NVIDIA CMX Context Memory Storage Platform)
https://www.nvidia.com/en-us/data-center/ai-storage/cmx/
Renamed NVIDIA CMX Context Memory Storage Platform (2026). A BlueField-4 (STX) powered "AI-native context tier" that extends GPU memory with a shared pod-level KV-cache tier for long-context, multi-turn agentic inference, with Dynamo routing and DOCA Memos SDK managing cache across compute and storage nodes. Announced CES Jan 2026; offered via storage partners (Dell, HPE, IBM, NetApp, VAST, WEKA, etc.).
Renamed · Verified 2026-10-23 · fresh as of 2026-10-23 · 43 cited facts, 1 with a third-party source · 2 recorded changes · Suggest a correction
At a glance
- Tier
- T2
- Section
- Dedicated memory layers
- Latest release
- not applicable — hardware platform; not OSS fix
- License
- proprietary (hardware/DOCA) + Dynamo Apache-2.0 repo fix
- GitHub
- not applicable — proprietary platform fix
- Pricing
- searched not found fix
- Funding
- Public (NVDA); ~$3T+ market cap (2026) vendor fix
- Freshness
- 2026-10-23 fix
- Layer
- agent-infra repo fix
- Licensing
- commercial repo fix
Memory taxonomy
- storage
- kv
- retrieval
- similarity
- persistence
- session
- update
- replacement
- unit
- token
- governance
- opaque
- conflict
- n/a
When to use
Optimised for: petabyte-scale RDMA-accelerated KV-cache placement (5x tokens/sec; 5x power efficiency); multi-turn TTFT reduction
Anti-fit: not for non-NVIDIA-Rubin/BlueField stacks; not for small-deployment / edge use; not for non-KV-cache memory needs (semantic conflict resolution etc.)
Pros & cons
Pros
Bridges GPU HBM and storage with hardware-accelerated context offload; broad partner ecosystem; aligned with Rubin cluster architecture; positions KV-cache as a first-class storage tier.
Cons
GA H2 2026 — not yet shipping; locked to NVIDIA Rubin/BlueField stack; multipliers reported without baseline absolute numbers; opaque tenant-isolation details.
Claims & capabilities
Up to 5x tokens/sec; up to 5x greater power efficiency vs traditional storage; reduced time-to-first-token in multi-turn
Technical surface
- API surface
- DOCA APIs; NIXL transfer library; Dynamo runtime
- Backend storage
- custom (BlueField-4 + partner SSD/NVMe arrays)
- Deployment
- On-prem rack-scale (NVIDIA Rubin cluster + partner storage); managed via DOCA
- Embedding model
- not applicable — KV cache (raw tokens; no embedding model)
- Multi-tenancy
- hard-isolation (BlueField DPU enforces)
- MCP
- searched not found
- A2A
- searched not found
- OpenTelemetry
- first-class (DOCA telemetry / OTel via partner integrations)
Alternatives to NVIDIA Inference Context Memory Storage Platform (now NVIDIA CMX Context Memory Storage Platform) → live options ranked by shared design, licence, hosting and MCP support
Similar systems
Other dedicated memory layers in the catalog, ranked by inbound references.
- Mem0 T1
Universal memory layer for AI agents. Three concurrent stores (vector + graph + KV); LLM-extracted facts; concurrent retrieval via ThreadPoolExecutor.
- Zep & Graphiti T1
Bi-temporal knowledge graph (event time + ingestion time). Strong on chronological reasoning and contradiction tracking. Graphiti is the open-source core.
- Cognee T1
"Extract–Cognify–Load" pipeline that turns raw input into a typed, queryable knowledge graph for agent recall.
- Hindsight (Vectorize) T1
Standalone memory service from Vectorize. Open source. Biomimetic four-network design (World, Bank, Observation, Opinion). Ships an MCP memory server.
- Letta / MemGPT T1
Originated as MemGPT (UC Berkeley), where the model pages its own memory between context, recall and archival tiers. The product now centres on Letta Code, a stateful agent harness whose agents rewrite their own memory, skills and prompts over time.
- Memori T2
Structured, persistent memory layer for AI agents. Open source.
Sources (42 cited facts, 1 third-party)
Each fact links to the page it came from; the badge says who published it (how to read sources).
- What it is Hardware-accelerated cluster KV-cache storage tier vendor
- What it does & what's distinct Renamed NVIDIA CMX Context Memory Storage Platform (2026). A BlueField-4 (STX) powered "… vendor
- Claims / benefits Up to 5x tokens/sec; up to 5x greater power efficiency vs traditional storage; reduced t… vendor
- Modalities text (KV-cache; modality-agnostic at storage layer) vendor
- Created Announced 2026-04 (NVIDIA GTC season); GA H2 2026 (BlueField-4 GA dependency) vendor
- License proprietary (hardware/DOCA) + Dynamo Apache-2.0 repo
- Mindshare NVIDIA newsroom + The Register + HPCwire + StorageReview coverage 2026-04 vendor
- Funding Public (NVDA); ~$3T+ market cap (2026) vendor
- Customers / scale Design-win partners: AIC, Cloudian, DDN, Dell Technologies, HPE, Hitachi Vantara, IBM, N… vendor
- Compliance SOC 2 / ISO via partner storage vendor (NVIDIA stack itself enterprise-grade) vendor
- Data handling RDMA-only data path; tenant isolation handled by partner storage controllers vendor
- Deployment On-prem rack-scale (NVIDIA Rubin cluster + partner storage); managed via DOCA vendor
- HQ US (Santa Clara CA) vendor
- Founders / pedigree NVIDIA; co-founders Jensen Huang, Chris Malachowsky, Curtis Priem (1993) third-party
- Memory volume / scale Petabyte-scale KV-cache capacity (cluster-level) vendor
- Performance 5x tokens/sec and 5x power efficiency multipliers (no absolute baseline disclosed) vendor
- Code/weights release Code public (Dynamo/NIXL OSS) + hardware proprietary repo
- API surface DOCA APIs; NIXL transfer library; Dynamo runtime vendor
- Latency p50/p99 sub-microsecond RDMA fabric (Spectrum-X) vendor
- Throughput line-rate 800G Ethernet (Spectrum-X) vendor
- Backend storage custom (BlueField-4 + partner SSD/NVMe arrays) vendor
- Multi-tenancy hard-isolation (BlueField DPU enforces) vendor
- Encryption at-rest + in-transit (DPU-accelerated crypto) vendor
- SSO / RBAC partner-dependent (storage vendor handles SSO/RBAC) vendor
- Consistency strong (RDMA in-cluster) vendor
- Tombstoning replacement (LRU/LFU at KV-cache layer) vendor
- Namespace primitives per-cluster / per-tenant (DPU-enforced) vendor
- Forgetting policy LRU/replacement (cache eviction) vendor
- OTel first-class (DOCA telemetry / OTel via partner integrations) vendor
- Import / export partner-dependent (storage vendor APIs for export) vendor
- Validated verticals gigascale agentic inference operators; cloud GPU/HPC providers; LLM inference vendors (d… vendor
- Time to running months (rack-scale BlueField-4 deployment; depends on Rubin GA H2 2026) vendor
- Anti-fit not for non-NVIDIA-Rubin/BlueField stacks; not for small-deployment / edge use; not for … vendor
- Optimised for petabyte-scale RDMA-accelerated KV-cache placement (5x tokens/sec; 5x power efficiency);… vendor
- Adjacent infrastructure requires NVIDIA Rubin cluster + BlueField-4 DPU + Spectrum-X Ethernet + DOCA + NIXL + Dy… vendor
- Pros Bridges GPU HBM and storage with hardware-accelerated context offload; broad partner eco… vendor
- Cons GA H2 2026 — not yet shipping; locked to NVIDIA Rubin/BlueField stack; multipliers repor… vendor
- Project & sources announcement; Dynamo OSS; DOCA platform; BlueField page vendor
- architecture-deployment-locus local vendor
- licensing-bucket commercial repo
- positioning-platform-vs-product platform vendor
- taxonomy-three-layer-rollup agent-infra repo
What changed
Changes to this record's lifecycle, licence, release, pricing and hosting facts, from the catalog's git history. Subscribe (Atom)
- 2026-10-23
- Name “NVIDIA Inference Context Memory Storage Platform” → “NVIDIA Inference Context Memory Storage Platform (now NVIDIA CMX Context Memory Storage Platform)”
- Homepage “https://nvidianews.nvidia.com/news/nvidia-bluefield-4-powers-new-class-of-ai-native-storage-infrastructure-for-the-next-frontier-of-ai” → “https://www.nvidia.com/en-us/data-center/ai-storage/cmx/”
- 2026-10-22
- License “proprietary (BlueField-4 firmware) + NIXL/Dynamo software stack (mixed)” → “proprietary (hardware/DOCA) + Dynamo Apache-2.0” source