WAIC 2026: DapuStor Highlights Enterprise SSD Innovation for AI Infrastructure

WorkAI.TV Editorial Desk
3 Min Read

Share with your CTO

Chinese SSD maker DapuStor is positioning itself as a full-stack storage partner for AI infrastructure, not just a component supplier. At WAIC 2026 in Shanghai, the company showed an enterprise SSD portfolio spanning QLC drives up to 245TB, Gen5 TLC drives with up to 3.5x transparent compression, and a low-latency SCM drive rated at 120 DWPD for inference workloads. Joint demonstrations with Weka and Gigabyte framed the pitch around GPU utilization and Time to First Token, the two metrics AI infrastructure teams currently live or die by.

What this means for your business

The storage layer is where AI infrastructure budgets quietly bleed out. GPU utilization numbers look disappointing not because the GPUs are slow but because they’re waiting on data, and most teams don’t audit that gap until they’ve already overspent on compute. DapuStor’s portfolio is built around exactly that problem. If your AI infrastructure team is already benchmarking NVMe storage against training throughput or TTFT targets, this product set belongs in that evaluation, particularly for shops considering alternatives to Western suppliers amid ongoing supply-chain pressure.

The transparent compression claim deserves scrutiny. A 3.5x effective capacity multiplier on the R6C TLC drive sounds attractive, but compression ratios are workload-dependent, and AI training datasets, which are often already compressed or in binary formats, tend to compress poorly. The number is real under the right conditions; the conditions matter more than the number. Any procurement conversation should lead with your actual dataset composition before letting that 3.5x figure anchor the capacity math.

DapuStor’s partnership framing with Weka is the more durable signal here. Weka has genuine traction in AI training environments at scale, and a validated joint architecture means DapuStor isn’t asking enterprise buyers to absorb integration risk alone. The question your storage and AI platform teams should weigh at the next vendor review isn’t whether Chinese SSD suppliers are competitive on specs, they clearly are, but whether your organization’s procurement policy and geopolitical risk tolerance have kept pace with what the hardware market now looks like.

Concept deep-dive: DWPD

DWPD stands for Drive Writes Per Day, a measure of how many times you can overwrite an SSD’s full capacity daily over its rated lifespan. Think of it as an endurance budget. Consumer SSDs typically run below 1 DWPD. At 120 DWPD, DapuStor’s X5 SCM drive is built for workloads that rewrite data constantly, like KV Cache in inference serving, where the cost of an endurance failure mid-inference is measured in model downtime, not just replaced hardware.

Based on reporting from WAIC 2026: DapuStor Highlights Enterprise SSD Innovation for AI Infrastructure, originally published 2026-07-31 08:01:00.

TAGGED:
Share This Article