[PSTG] Pure Storage Thesis 2026: AI Training Demand Reshapes the Storage Addressable Market

Key Takeaways

Pure Storage's fiscal year 2025 (ended February 2, 2025) was a breakout year defined by the company's emergence as a primary beneficiary of AI infrastructure spending: revenue reached $3.10B, up 11% from $2.79B in FY2024, and more importantly the subscription services business — Pure's annualized recurring revenue (ARR) — crossed $1.7B, marking the transformation of a hardware-centric storage vendor into a software and subscription-driven franchise with expanding predictable revenue. Product revenue from FlashArray and FlashBlade all-flash storage appliances grew modestly, but the more significant development was the Evergreen//One storage-as-a-service model gaining enterprise traction as AI training and inference infrastructure operators prioritized predictable storage costs over CapEx. Non-GAAP operating income reached approximately $530M at a 17% margin, and adjusted EPS reached approximately $1.70. The thesis is straightforward: Pure occupies the highest-performance tier of enterprise storage with a differentiated software (Purity) and subscription model that creates retention and expansion economics, and AI training clusters — which generate and process enormous volumes of unstructured data — structurally expand the addressable market for all-flash storage at the performance tier where Pure has no credible competitor at scale.

Pure Storage was founded in 2009 by John Colgrove and John Hayes, commercializing the insight that NAND flash memory had reached a price crossover point where all-flash storage arrays could be cost-competitive with spinning disk while offering dramatically better performance, power density, and reliability. The company went public in 2015 and built its enterprise storage position through a combination of technology differentiation (Purity operating system, DirectFlash modules that use proprietary NVMe-based flash rather than off-the-shelf SSDs), a clean consumption model (Evergreen subscription that upgrades storage in place rather than requiring forklift replacements), and a go-to-market focused on enterprise accounts displacing incumbent vendors (Dell EMC, NetApp, HPE). CEO Charlie Giancarlo, who joined in 2017 from Silver Lake, transformed the commercial strategy toward subscription and services; current CEO Rob Lee (as of 2024) has maintained this trajectory.

The AI infrastructure wave has provided a second structural tailwind beyond Pure's original enterprise replacement opportunity. AI training and inference workloads require high-throughput, low-latency storage for training datasets, model checkpoints, and inference serving. GPU clusters from NVIDIA (H100, H200, B200) are only as productive as the storage feeding them — storage that cannot keep up with GPU throughput creates idle GPU time that costs hyperscalers and enterprises thousands of dollars per hour. Pure's FlashBlade//S product line, with its parallel NFS/S3 architecture optimized for AI workloads, positions directly in this bottleneck. Several of the world's largest AI training clusters use Pure Storage, and the company has named AI infrastructure as its largest and fastest-growing incremental end market.

Business Structure

Pure Storage reports revenue in two categories: Product and Subscription Services.

Product (~$1.55B in FY2025, ~50% of total): All-flash storage hardware and embedded operating software. The primary products are FlashArray//C and FlashArray//X (block storage for databases and enterprise applications), FlashArray//XL (high-performance block for demanding workloads), and FlashBlade//S and FlashBlade//E (file and object storage for unstructured data and AI workloads). Product revenue is inherently lumpy — enterprise storage purchases are capital expenditure decisions that fluctuate with IT budget cycles.

Subscription Services (~$1.55B in FY2025, ~50% of total): Pure's subscription and support offerings include Evergreen//Forever (annual support and software upgrades), Evergreen//One (as-a-service storage with capacity-based pricing where Pure owns and operates the hardware), and Portworx (Kubernetes-native data services platform acquired in 2020 for $370M). Subscription services ARR reached approximately $1.73B in FY2025, growing at approximately 20-22% annually — faster than total revenue — and shifting the revenue mix toward recurring, predictable income. Subscription gross margin is approximately 73-75%, meaningfully above product gross margin of approximately 68%.

Key Core Metrics Performance

Revenue and ARR Growth (FY2021–FY2025)

Revenue has grown consistently above the broader enterprise storage market, driven by flash share gains in block storage and new category creation in AI-optimized file/object storage.

The ARR re-acceleration from 17% in FY2024 to 21% in FY2025 reflects the AI infrastructure demand surge, as new Evergreen//One contracts for AI training environments contributed a larger-than-expected proportion of new business.

Non-GAAP Operating Margin (FY2021–FY2025)

Margin expansion has been the secondary narrative — Pure has consistently argued that investment in go-to-market and R&D (Portworx, AI-optimized FlashBlade) would yield operating leverage as the subscription base scaled.

Margin plateaued at approximately 17% in FY2025 as the company increased Evergreen//One investment (CapEx to deploy as-a-service hardware reduces near-term margins but builds the ARR base) and Portworx go-to-market spending. The path to 20%+ margins requires either continued ARR growth or a decision to slow as-a-service investment, which management has not signaled.

Non-GAAP EPS and Free Cash Flow (FY2021–FY2025)

FCF margin is structurally below operating margin because Evergreen//One growth requires capital expenditures for hardware deployment that precedes subscription revenue recognition — a transitional dynamic as the as-a-service book grows. On a steady-state basis (when Evergreen//One capex equals Evergreen//One depreciation), FCF margin should converge toward non-GAAP operating margin.

Market Evaluation

Pure Storage trades at approximately 25-30x forward non-GAAP earnings entering FY2026 (fiscal year ending February 2026), reflecting the combination of above-average growth expectations and the premium multiple assigned to subscription-model infrastructure. The primary bull argument is that AI training workloads are structurally capacity-expanding: each new AI model generation requires larger training datasets and more frequent checkpoint storage, perpetually expanding the petabyte demand for high-performance file storage. The FlashBlade//S platform has no close competitor at the throughput/capacity combination required for 100-petabyte-scale AI training clusters — NetApp's A-Series and Dell PowerScale are both credible alternatives at lower performance points but lack Pure's parallel architecture advantage for the largest workloads. The bear case is margin: Evergreen//One growth suppresses near-term FCF margin, and if the transition to consumption-based storage accelerates, Pure bears the CapEx risk and hardware depreciation before subscription revenues catch up. Competitor price competition from Dell's APEX program and NetApp's StorageGrid service could also compress Pure's subscription pricing power in the SMB and mid-market segments.

AI Infrastructure and FlashBlade//S Positioning

The defining commercial development of FY2025 was the acceleration of hyperscaler and large enterprise AI training cluster purchases of FlashBlade//S. The product's architecture — which uses a single-namespace file system distributable across hundreds of blades with parallel data delivery — achieves throughput levels (terabytes per second sustained) that spinning-disk or conventional flash alternatives cannot match at equivalent cost density. AI training jobs on GPU clusters need to feed training data and checkpoint model states to storage continuously; a 1,000-GPU cluster training at 300GB/second aggregate throughput requires storage capable of sustaining that throughput without creating bottlenecks. FlashBlade//S achieves this at a price per TB that has fallen steadily as NAND costs declined through FY2023-FY2025.

Several public hyperscalers and AI-native companies have disclosed Pure Storage deployments, and Pure itself cited AI infrastructure as the fastest-growing end-market in FY2025 earnings commentary. The Evergreen//One model is particularly well-suited to AI operators who prefer operational expense over capital expenditure — start-ups and mid-sized AI labs frequently prefer a predictable $/petabyte/month subscription over a $5-10M CapEx purchase, and Pure's as-a-service model accommodates this preference while building recurring revenue.

The key FY2026 variable is whether this AI-driven demand is durable or cyclical. If the AI infrastructure buildout pauses — as occurred briefly in early 2023 — Pure's product revenue is vulnerable to a significant quarterly swing. The subscription ARR base provides a buffer (recurring revenue does not disappear in a quarter), but approximately 50% of Pure's revenue remains project-based product purchases, which are subject to the same cyclicality as enterprise IT spending broadly.