Seagate Q1 2026: AI Storage Boom, 130 Exabytes, HDD vs Flash Guide

Seagate shipped 130 exabytes of hard drive capacity in its fiscal Q3 2026 (January through March 2026), a record for the company. Revenue came in at $2.42 billion, up 27% year-on-year. Shares jumped 11% on the April 29 earnings beat and took Western Digital (WDC) and SanDisk (SNDK) up with them — both gained 5-7% on the Seagate read-across. The driver in every line of the earnings call: nearline hard drives for AI data centers.

This is the storage angle that has been buried under GPU and HBM coverage. While the tech press has focused on HBM4 shortages and custom AI silicon, the actual data that AI models train on — petabytes of internet text, images, code, and video — lives on magnetic hard drives in hyperscaler data lakes. Seagate's record quarter is a direct consequence of the $630 billion in AI infrastructure capex being deployed in 2026.

Seagate Q3 FY2026: The Actual Numbers

Seagate's fiscal year ends in late June, so its third fiscal quarter covers January through March 2026.

• Revenue: $2.42 billion, up 27% year-on-year from $1.91 billion in Q3 FY2025
• Gross margin: 34.2%, up from 28.1% a year ago (margin expansion driven by product mix shift toward high-capacity nearline drives)
• Exabytes shipped: 130 exabytes, a company record — approximately 14% above the prior quarter
• Average selling price per drive: up meaningfully quarter-on-quarter, driven by the mix shift to 20TB+ nearline drives
• Non-GAAP EPS: $1.87, beating consensus by $0.18
• Free cash flow: $380 million

CEO Dave Mosley on the earnings call: "The data center customer segments, particularly those building out AI training and inference infrastructure, are ordering at a velocity we haven't seen outside of the 2021 supply crunch. Lead times on our high-capacity nearline product are extending as demand continues to outpace our current build rate."

The read-across effect was immediate. Western Digital (which also produces nearline HDDs) gained 7% on the Seagate announcement. SanDisk (enterprise flash, now separate from WDC) gained 5%. Storage as a category is repricing for the AI data era.

Why AI Infrastructure Runs on Hard Drives (Not Just Flash)

The AI infrastructure narrative in 2025-2026 has focused on compute (GPUs, TPUs, custom ASICs) and memory (HBM3e, HBM4). Storage has been the neglected third leg of the AI infrastructure triangle — and Seagate's results make it impossible to ignore further.

Here is the practical reality of how data flows in an AI training pipeline:

Cold storage (nearline HDD): The raw training corpus lives here. For large language model training at the scale of GPT-5, Claude Opus, or Gemini Ultra, this is petabytes to tens of petabytes of text, image, and video data. Nearline HDDs at 20-32TB per drive offer the lowest cost per terabyte at scale. A single rack of nearline drives can hold 5-8 petabytes. Cost: approximately $15-25 per terabyte at current pricing.

Warm storage (high-density SAS or SATA SSD): The training pipeline reads batches from cold storage and stages them in warm storage for repeated access during training runs. Not everything needs to be in warm storage — only the batches scheduled for the next training cycle. Cost: approximately $50-120 per terabyte.

Hot storage (NVMe SSD, GPU-local VRAM or HBM): Active training data in the immediate compute pipeline. Extremely low latency, extremely expensive, smallest capacity. A100/H100 GPUs have 80 GB of HBM. Cost: thousands of dollars per terabyte.

The tiering decision is not arbitrary — it is an economic optimisation. Putting all training data in NVMe SSDs is prohibitively expensive. Using only HDDs creates an I/O bottleneck that starves GPUs. The correct architecture balances cost against the I/O requirements of the training job.

Seagate's boom is specifically in the cold and warm nearline tiers. Hyperscalers building out AI training infrastructure need petabyte-scale nearline storage before they can train anything.

The 20TB+ Drive Economics

The revenue per exabyte metric is the key to understanding Seagate's margin expansion. As hyperscalers order higher-capacity drives (20TB, 24TB, 28TB), each drive costs more but the per-terabyte price is actually lower than older smaller drives. This means:

• Seagate ships fewer physical drives to deliver the same capacity
• Manufacturing cost per drive is relatively fixed (not linearly proportional to capacity)
• Revenue per drive and margin per drive both increase as capacity increases

Seagate's HAMR (Heat-Assisted Magnetic Recording) technology enables 32TB and eventually 50TB+ drives that were not feasible with conventional recording. HAMR drives are now in volume production. The technology transition was painful (HAMR drives had higher failure rates early in production), but Seagate has resolved the yield issues and is shipping HAMR-based nearline drives to major hyperscaler customers.

HAMR drives are not priced at a discount to conventional drives — they command a premium. The margin expansion Seagate is reporting (gross margin up from 28% to 34% YoY) is largely explained by this product mix shift.

Lead Times Are Extending: What This Means for DevOps Teams

Seagate's CEO flagged extending lead times on high-capacity nearline drives. In practical terms: if you are building or expanding a data center that will host AI training workloads, you cannot assume you can order nearline storage and receive it within the standard 4-6 week window. Lead times on 24TB+ nearline drives are currently running 10-16 weeks for non-hyperscaler customers.

For teams building on-premises AI infrastructure (private cloud, on-prem GPU clusters for training): Order nearline storage early and over-provision. The cost of having excess cold storage capacity is low (drives are relatively cheap per TB); the cost of a GPU cluster sitting idle waiting for training data is very high.

For teams running managed storage on AWS, Azure, or GCP: You are insulated from drive lead times — the cloud providers pre-procure at scale. However, the Seagate lead time pressure will likely translate into higher S3, Blob Storage, and GCS pricing in H2 2026 as cloud providers face the same supply constraints and pass costs through.

For teams evaluating HDD vs all-flash arrays (AFA): At current pricing, all-flash is still 4-8x more expensive per raw terabyte than nearline HDD for cold/warm storage. The NVMe price curve is declining, but at the petabyte scale required for AI training data, the economic case for HDD nearline remains strong. All-flash makes sense for hot storage, active databases, and low-latency inference caches — not for the training data lake itself.

The Power and Heat Reality

Nearline HDDs are not power-efficient compared to SSDs on a per-operation basis, but they are extremely efficient on a per-terabyte stored basis. At petabyte scale:

• A 24TB SATA HDD: approximately 4-7 watts idle, 7-9 watts active
• An equivalent-capacity NVMe SSD array (24TB of enterprise NVMe): approximately 35-50 watts
• Per petabyte of stored data: HDDs consume approximately 3-5x less power than equivalent NVMe capacity

For AI data centers already under extreme power pressure (the $630B capex wave is straining grid capacity across every major data center market), the power efficiency of nearline HDD for cold storage is not just a cost consideration — it is a physical constraint. A data center serving a 100-exabyte AI training corpus from all-flash would require roughly 3-5x more power for storage alone versus HDD nearline. This is not feasible given current grid availability.

Seagate's storage is, in a non-trivial sense, enabling the AI training infrastructure buildout precisely because HDDs make petabyte-scale training data manageable from a power perspective.

Key Takeaways

• Seagate Q3 FY2026: $2.42B revenue (+27% YoY): record 130 exabytes shipped; gross margin 34.2% (up from 28.1%); non-GAAP EPS $1.87 beat consensus by $0.18; shares +11%, pulling WDC +7% and SNDK +5%
• AI nearline HDD is the demand driver: hyperscaler AI training data lakes require petabyte-scale cold/warm storage; nearline HDD at $15-25/TB is the only cost-viable answer at that scale
• Lead times extending to 10-16 weeks: on-prem AI infrastructure teams should over-order now; cloud teams are insulated but expect S3/Blob/GCS price pressure in H2 2026
• HAMR technology driving margin: 24TB+ HAMR drives at premium pricing explain gross margin expansion from 28% → 34%; 32-50TB drives in production pipeline
• Storage tiering reality: cold (HDD nearline) → warm (SAS/SATA SSD) → hot (NVMe/HBM) is the correct AI pipeline architecture; all-flash for petabyte training corpora is 4-8x more expensive and 3-5x more power-hungry than HDD
• Developer action: price storage tiers before ordering; lead-time-extend drives now for H2 2026 build-outs; on managed cloud, model S3/GCS price increases for large training data workloads in H2

For the GPU and HBM supply chain context driving AI infrastructure demand, read TSMC Q1 2026: 58% Profit Jump, 4.17M Wafers, HBM4 Sold Out. For the hyperscaler capex that is driving storage demand at this scale, read Big Tech Q1 2026: Meta +31%, Google Cloud +50%, Amazon Chips $20B.