Oxide's Second-Gen Cloud Computer Ships With Micron DDR5 and SSDs
Micron DDR5 RDIMMs and 7000-series SSDs now ship in Oxide's second-gen on-premises cloud rack, with 1.5 TiB per sled and a Micron Ventures equity stake attached.
By Daniel Okafor
4 min read

What's News
- Oxide announced on November 16, 2026 that Micron DDR5 memory and 7000-series SSDs ship in its second-generation Cloud Computer, with up to 1.5 TiB of memory per compute sled.
- Micron invested in Oxide through Micron Ventures.
- The rack supports up to 24 sleds, 36 TiB of DRAM, 6.5 PiB of storage, 12.8 Tbit/s switching and up to 30 kW power draw.
- Oxide sells the rack outright with no metered billing, per-core fees or egress fees.
- Each sled runs one AMD EPYC 9005 Series processor with up to 192 cores and 384 threads.
By Daniel Okafor
Micron DDR5 memory and 7000-series SSDs are now shipping inside Oxide Computer's second-generation on-premises Cloud Computer, with the largest memory configuration reaching 1.5 TiB per compute sled. Oxide announced the qualification on November 16, 2026. Micron has also taken an equity stake in Oxide through its venture arm, Micron Ventures.
The deal deepens a hardware partnership that spans both memory and storage for on-premises cloud racks. It targets a specific buyer: organizations that keep workloads in their own data centers because of security, sovereignty, latency or operational-control requirements, yet still want cloud-style operations.
Jeremy Werner, senior vice president and general manager of Micron's Core Data Center Business Unit, said: "AI and data-intensive workloads are changing how enterprises design compute infrastructure in their own data centers. Micron DDR5 RDIMMs and Micron 7000-series SSDs help Oxide Cloud Computer customers achieve cloud-scale capabilities within infrastructure they own and control."
What sits inside the rack?
Each sled in the second-generation system carries one AMD EPYC 9005 Series processor, with up to 192 cores and 384 threads. Memory runs at 6400 MT/s, and each sled has two 100GbE connections.
The memory options come in three tiers. The largest uses twelve 128GB Micron DDR5 RDIMMs, which is what produces the 1.5 TiB per-sled maximum. Storage splits across two drive families:
- Up to ten Micron 7600 SSDs provide NVMe block storage, configurable as distributed or local.
- Two Micron 7450 SSDs serve as boot drives.
At rack scale the numbers grow considerably. The chassis supports up to 24 second-generation sleds, 36 TiB of DRAM and 6.5 PiB of raw storage. Oxide separates those hardware totals from what guests actually see. Per-rack maxima exposed to tenants are 7,875 vCPUs, 30.6 TiB of DRAM and 1.7 PiB of NVMe block storage.
The networking and power envelope matters as much as the compute. The rack carries 12.8 Tbit/s of switching capacity and a maximum listed power draw of 21.6 kW in redundant mode, or 30 kW in non-redundant mode. Actual specifications vary by configuration.
How does the commercial model work?
Oxide sells the rack outright. There is no per-minute billing, no per-seat licensing, no per-core fees and no egress fees. Customers buy the hardware once and operate it themselves.
Co-founder and CTO Bryan Cantrill framed the Micron relationship and investment as an extension of Oxide's co-designed hardware-and-software approach. The goal, in his telling, is to keep the economic and security advantages of ownership while stripping away its traditional operational burden.
The trade-off for customers is direct. They take ownership of the rack and its power requirements, and in exchange they get isolated, self-service capacity under a purchase-based model rather than metered cloud billing. For workloads with heavy data movement, the absence of egress fees alone can shift the cost equation.
Why HPC first?
The Cloud Computer is currently available for high-performance computing data-center deployments. That market fits the product's design constraints well.
Inside an HPC center, the system runs services researchers need outside batch queues. It provides isolated tenant spaces and resource quotas for individual teams or projects. Researchers provision virtual machines through an API, a command-line interface or a web console using self-service provisioning.
Standard VMs and containers support existing tools. Terraform and Ansible deployments can instantiate the Slurm and Flux schedulers, the two most common HPC workload managers. Oxide's HPC product page lists:
- REST APIs, a Terraform provider and CLI access
- Compatibility with Kubernetes, Jupyter, GitLab, Grafana and Prometheus
That compatibility allows interactive notebooks, data pipelines, monitoring and orchestration to run alongside an HPC cluster rather than in a separate infrastructure silo.
What does this mean for the market?
The Micron investment signals supplier confidence in Oxide's ownership-based model, and it locks in a qualified memory and storage supply chain for a company that builds its own rack-scale hardware. Werner's comment points to the demand driver: AI and data-intensive workloads are pulling cloud design patterns back into enterprise-owned data centers. Oxide's bet is that a growing set of buyers will pay for the rack up front to avoid the metered bill — and Micron now holds a stake in that bet paying off.
Source: unite.ai
More from Daniel Okafor
Show full bio
Correspondent covering money, funding and market shifts around entrepreneurship, startups and venture funding at Business Bearings.
2,189 articles