WEKA breaks the exabyte-per-rack barrier with compression and Micron SSDs

What happened?

WEKA unveiled the WEKApod 3, a line of storage appliances designed to run its new NeuralMesh 6 software. The top configuration, Prime Max, promises 1.1 exabytes (1,100 PB) of effective capacity in a single 56U rack, breaking the exabyte-per-rack barrier for the first time. However, this figure heavily relies on data reduction technologies: the raw hardware capacity is only 441.5 PB, using approximately 1,800 units of the Micron 6600 ION 245.76 TB SSD. Jeremy Werner, senior vice president of Micron's data center business unit, highlighted that this architecture achieves 15.8 PB in a single 2U chassis.

The system delivers 10.2 TB/s and 210 million IOPS, outperforming competitors by 114% in performance density, according to WEKA. The NeuralMesh 6 software includes compression, deduplication, fingerprinting, and similarity hashing, with a write overhead of less than 5%. WEKA contractually guarantees a reduction ratio of up to 6x on AI training data. Hardware can be ordered now, with deliveries starting in fall 2026, while NeuralMesh 6 will be available in the second half of 2026, with free upgrades for existing customers.

Why is it important?

This launch marks a milestone in storage density, but also reveals a key trend: the reliance on software to overcome physical hardware limitations. Amid data center space and power shortages (a 49 GW deficit is forecast in the U.S. through 2028, according to Morgan Stanley), optimizing storage is critical. WEKA argues that inefficient storage directly competes with GPUs for space and power budgets. Historically, advances in storage density have depended on NAND improvements (e.g., moving from SLC to QLC) or faster interfaces (NVMe over PCIe). However, WEKApod 3 represents a qualitative leap by combining dense hardware with aggressive, contractually guaranteed data reduction software.

Additionally, vertical integration (WEKA manages its own supply chain) enables more predictable pricing and delivery timelines, unusual in an OEM-dominated market. The AlloyFlash tiering automatically between TLC and QLC (30-40% cheaper per TB) optimizes costs based on workload. This approach contrasts with solutions like Pure Storage or NetApp, which rely on third-party disk supply and often face delays. WEKA's ability to control its supply chain gives it a competitive edge in a market where component availability is a bottleneck.

Consequences and context

The launch comes as U.S. data center construction slowed in 2025, and grid connection queues extend from 4 to 7 years. This pressures vendors to maximize density. However, the promise of 1.1 exabytes should be taken with caution: without compression, the system offers less than half. WEKA believes AI data patterns enable high reduction rates, but not all workloads benefit equally. For example, raw video data or pre-compressed files may achieve much lower ratios, while language model training data (with high redundancy) could reach or exceed 6x.

Analyst Steve McDowell of NAND Research notes that production-scale inference workloads require different metrics, such as tokens per rack and cost per inference, which buyers should consider. Additionally, the 10.2 TB/s and 210M IOPS are leading figures but verifiable only under optimal conditions; in real-world mixed read/write environments, performance may be lower. WEKA's contractual guarantee mitigates some risk but also means the company must deliver or face penalties.

Compared to previous events, the WEKApod 3 surpasses the previous density record, the Dell PowerScale F900 (which offered ~60 PB per rack with hard drives), but does so with flash and reduction software. Other competitors like VAST Data or Pure Storage have achieved densities of hundreds of petabytes per rack, but none have exceeded the exabyte. This milestone could accelerate all-flash storage adoption in data centers where hard drives were previously used for cost, especially if the effective price per TB proves competitive.

What should readers know?

• Effective vs. raw capacity: The 1.1 exabyte figure is after compression; raw capacity is 441.5 PB. Actual results depend on data type. WEKA guarantees up to 6x on AI data, but other workloads may achieve less.
• Performance: 10.2 TB/s and 210M IOPS are leading figures but verifiable only under optimal conditions. The contractual guarantee includes performance, offering some assurance.
• Availability: Hardware can be ordered now, with deliveries from fall 2026. NeuralMesh 6 arrives in H2 2026; current customers get a free upgrade.
• Market context: Data center power and space shortages drive high-density solutions. WEKA bets on vertical integration to avoid supply bottlenecks, potentially offering better delivery timelines than competitors.
• Contractual guarantee: WEKA offers reduction ratio and performance guarantees, adding confidence but also risks if unmet. This is unusual in the industry and may pressure other vendors to offer similar guarantees.
• Market impact: The WEKApod 3 could shift density expectations in enterprise storage, forcing competitors like Dell, NetApp, or Pure Storage to accelerate their own high-density solutions. Moreover, the reliance on software to achieve these figures underscores the importance of system-level optimization, not just hardware.