Computex 2026: Has the HDD’s Real Bottleneck Been Hiding in Plain Sight?

What WD detailed, first at its Innovation Day earlier this year and again on the Computex floor, are two complementary mechanisms aimed at access density. High Bandwidth Drive Technology builds on triple-stage actuator precision to position read and write heads over more than one track at a time. A drive that today keeps a single head active can, with this approach, work multiple paired tracks at once, and this two-track operation is designed to roughly double sequential throughput. Dual Pivot Technology is the more structural change. It adds a second actuator at the far end of the drive, allowing independent sequential streams and independent random seeks.

The counterintuitive part is capacity. Earlier stacked dual-actuator drives bought performance by dropping a platter to make room; WD's opposite-end placement distributes the heads, which appears to let it tighten disk spacing and add platters rather than sacrifice them. Think of it less as a bigger warehouse and more as widening a loading dock that had fallen behind the size of the building. Combined, two-track HBDT and dual pivot are projected to lift throughput from roughly 300MB/s to about 1.2GB/s, a fourfold gain, while holding the cost structure that keeps HDDs as the backbone of bulk storage.

We would temper the enthusiasm on timing. The high-bandwidth drives are described as already in customer validation, while dual pivot reads as a 2028-class technology with sampling ahead of it. So this is a direction, architected and demonstrated, not a purchase order for this quarter. Since 2017, WD has grown CMR capacity 116% while sequential throughput rose only 18%. These designs are the company’s bid to close that gap.

Market Analysis

The competitive picture is three-sided, and each maker is telling a different story. Seagate runs the industry-standard capacity-first play, its HAMR-based Mozaic platform qualified and shipping up to 44TB at hyperscalers with a public path toward 100TB. Toshiba takes the conservative route, leaning on flux-controlled MAMR and treating early HAMR as a test vehicle rather than a volume product. Whether throughput-per-terabyte parity is sufficient remains an open question. AI workloads continue to increase data intensity, and maintaining today’s access characteristics may prove necessary but not sufficient for future AI-scale environments.

The industry’s first response to performance bottlenecks is often to add more systems. Improving bandwidth  per drive potentially reduces the number of drives, enclosures, ports and watts required to deliver a given level of bandwidth. WD is seeking differentiation in the space by competing on throughput-per-terabyte rather than capacity alone. The contra position for Seagate is strong, and we will state it plainly: if capacity per rack gates a hyperscaler's build, proven density today is worth more than an access performance roadmap that matures in 2028. WD's counter is that the workload mix is shifting toward the patterns where throughput-per-terabyte bites: object storage, data lakes, and training pipelines. With those in mind, the company claims that its design keeps the same interface and command set, so adoption does not force a rearchitecture.

The macro backdrop rewards whoever holds the cost line. McKinsey forecasts global data center spending could reach $6.7 trillion by 2030, with roughly 70% of new capacity demand tied to AI workloads. Storage has to scale inside that build without breaking the budget. We would also stress complementarity over conflict. WD's framing is not aimed at the compute layer; it argues compute and storage scale together.

WD is also staffing for the thesis. In May it added Manuvir Das to its board, the operator who ran enterprise computing at NVIDIA and held senior roles at Dell EMC and Microsoft before becoming an Operating Partner in Stonepeak's digital infrastructure group. That background sits at the exact seam WD is trying to own, where AI compute meets the data infrastructure beneath it. It signals a company that wants a seat at the design table as hyperscalers architect the next generation of AI data systems, rather than being perceived as a vendor waiting to supply the drives that fill them.

A drive-level gain only matters if the surrounding system carries it to the accelerators, and that is what WD's platform layer is for. The throughput technologies arrive with the building blocks meant to deliver them: the Ultrastar Data 3000 JBOD, OpenFlex EBOF, the RapidFlex NVMe-oF controller, and tiered architectures developed with Ceph, IBM Storage Scale, and XTAO. The aim is to match each tier of the AI data lifecycle to its own cost and performance profile, so faster drives raise GPU utilization rather than leaving bandwidth stranded. Without that plumbing, a faster drive is just a faster component inside the same bottleneck.

Comparative View: The Three-Way HDD Roadmap

• WD is betting on access density. Two-track HBDT plus dual pivot targets roughly 4x sequential throughput (about 300MB/s to 1.2GB/s), paired with a capacity path toward 60TB on ePMR and 100TB on HAMR by 2029. Dual pivot reads as a 2028-class technology.
• Seagate is betting on areal density. Its HAMR-based Mozaic 4+ platform is qualified and shipping up to 44TB at hyperscalers today on a 10-platter design, with Mozaic 5 targeting 50TB qualification in late 2027 and a public path to 100TB. MACH.2 dual-actuator covers its performance gap.
• Toshiba is betting on derisking. A conservative, step-by-step path centered on flux-controlled MAMR, with early HAMR positioned as a test vehicle through 2026 to 2027 rather than a volume product.