Summary
- Everpure reported third-quarter fiscal 2027 revenue of $1.46bn, up 51%, with product revenue increasing 77% to approximately $948m.
- Remaining performance obligations reached $4.2bn, while full-year revenue guidance increased to $5.20bn–$5.21bn.
- The company reported new FlashBlade//EXA benchmark results, but its second top-five hyperscaler agreement was originally announced in August.
Everpure has reported a 51% increase in quarterly revenue to $1.46bn, as demand for storage infrastructure supports higher hardware sales and continued investment in systems designed for large computing environments.
The company, formerly known as Pure Storage, published results for the third quarter of its 2027 financial year on 18 November. The reporting period ended on 1 November 2026.
Product revenue increased 77% year on year to approximately $948m, while subscription services revenue rose 19% to $512m. The results reflect demand across Everpure’s wider storage business, including enterprise infrastructure and emerging applications in AI and hyperscale computing.
The company also reported $2.2bn in subscription annual recurring revenue, up 20%, and remaining performance obligations of $4.2bn, an increase of 44% from a year earlier.
Remaining performance obligations represent contracted revenue expected to be recognised in future periods. They are not equivalent to storage hardware already shipped, installed capacity or the electrical demand of operating data centres.
Everpure increased its full-year revenue guidance to between $5.20bn and $5.21bn, compared with its previous range of $5.03bn to $5.07bn. The revised outlook remains a company forecast rather than revenue already earned.
The quarter also included technical developments involving flash storage performance, including the company’s FlashBlade//EXA platform and software intended to improve access to data used by AI computing systems.
Flash infrastructure supports larger computing workloads
Storage performance affects how efficiently computing equipment can operate, particularly where workloads require frequent transfers of information between persistent storage, system memory and accelerators.
AI training systems may repeatedly retrieve large datasets while writing checkpoints used to recover processing after interruptions. These activities place different demands on storage infrastructure, combining sustained data transfers with periods of concentrated reading or writing.
Inference systems introduce other requirements. Models and supporting data must be made available to processing equipment, while the infrastructure responds to changes in customer requests and application activity.
The performance of a storage platform depends on its drives, controllers, software and network connectivity. A high-capacity flash system may still encounter limits if its external connections or internal processing resources cannot accommodate the demands of the workload.
Everpure’s DirectFlash architecture forms part of its approach to high-density storage. The company uses flash media and associated software to manage how information is stored, protected and accessed.
Flash systems can provide high throughput and relatively low access latency, although the economic and operational advantages depend on the devices used, application requirements and the system they replace.
Storage density is another consideration. Equipment capable of accommodating more information within the same rack space may reduce the footprint required for a given capacity, but its electricity consumption must be assessed under representative operating conditions.
Higher storage density does not automatically improve overall data centre efficiency. The supporting processors, network equipment and cooling systems contribute to the total energy requirement, while utilisation determines how much of the available capacity is used.
New results address checkpointing and inference
Everpure reported that its FlashBlade//EXA system ranked first across the checkpointing and key-value cache metrics it submitted to the MLPerf Storage v3.0 closed division.
The qualification is important. The company’s result applies to the submitted tests and configurations, rather than demonstrating that the system leads every storage benchmark or outperforms all competing platforms under every workload.
Checkpointing is particularly relevant to large training systems. Periodically storing the state of a workload allows processing to resume from an earlier point after an interruption, reducing the amount of work that must be repeated.
Large models can generate substantial checkpoint files, and the time needed to write or recover them depends on storage throughput, software behaviour and network performance.
The key-value cache tests address another component of AI infrastructure. Such caches retain information used during model inference, potentially reducing repeated processing and affecting response times.
Everpure has also introduced PureKVA on FlashBlade, a software capability intended to improve access to inference context and reduce the delay before a model begins returning results.
The actual benefit will depend on the model, memory arrangement, storage platform and serving architecture. Performance figures obtained with a specific test configuration should not be assumed to apply directly to every production deployment.
These technical developments extend the company’s business beyond conventional enterprise storage replacement. They are designed to support workloads where the performance of the data system can influence the utilisation of costly computing equipment.
Hyperscaler agreements remain subject to deployment
Everpure’s latest results refer to its design win with a second top-five hyperscaler, but that agreement was originally disclosed on 10 August. It is not a new customer contract announced during the November reporting period.
The earlier announcement established a design win and supply agreement for the company’s DirectFlash technology. The customer was not identified publicly, and the agreement did not disclose a confirmed number of storage units installed at operating facilities.
Everpure’s management indicated during its November earnings discussion that shipments to the second hyperscaler were expected during 2027. The company has not provided a precise facility-level deployment schedule.
The distinction between a design win and installed infrastructure is substantial. Hardware must be manufactured, delivered, integrated, tested and accepted before it can support commercial computing workloads.
Hyperscale customers also evaluate equipment against their wider operating requirements, including failure recovery, software compatibility, network architecture and the ability to manage large fleets of storage devices.
Demand for flash media can be influenced by semiconductor component availability and pricing. Everpure’s margins and future revenue may therefore be affected by procurement conditions even where customer demand remains strong.
The company’s quarterly revenue increase reflects the performance of its complete business rather than a separately disclosed 51% expansion in hyperscaler storage shipments. Product sales, subscription services and future contracted obligations describe different commercial measures.
The latest results establish higher recognised revenue, stronger hardware sales and additional reported storage performance evidence. The contribution from hyperscale customers will become clearer as planned shipments progress and further revenue is recognised.

