Everspin is becoming a more interesting business because the reasons customers choose its memory are becoming more commercially relevant. Enterprise storage needs to protect unfinished writes. Industrial equipment needs to preserve operating data. Satellites need dependable boot and recovery functions. In each case, losing a small amount of information can be far more expensive than the memory installed to protect it.
That is the economic argument behind magnetoresistive random-access memory, or MRAM. Everspin has spent years commercializing it. The investment question now is whether a broader product portfolio, stronger manufacturing arrangements and better access to customers can turn that technical position into sustained growth.
I am bullish because several developments are beginning to reinforce one another. IBM provides a documented example of adoption in enterprise storage. Amentum connects Everspin to a substantial defense development program. Microchip is adding a second US manufacturing location, while Teledyne HiRel is preparing to take newer MRAM products to demanding aerospace customers. MaxLinear introduces a more direct opportunity in AI server architectures.
Each relationship has a different commercial role. Understanding those roles explains considerably more about Everspin than a collection of familiar company logos.
Why customers pay for persistence
MRAM stores information through magnetic states. Its practical attraction is the combination of fast access, repeated writes and the ability to retain stored data after power disappears. Everspin offers different products for different workloads, including established Toggle MRAM and denser spin-transfer torque MRAM, or STT-MRAM. Their performance and retention specifications vary by device.
Consider a storage controller that has accepted data but has yet to finish writing it to flash. A persistent buffer can protect that information during an interruption. In a factory controller, the valuable information might be a machine setting or an event log. Preserving it can simplify recovery and reduce the work needed to reconstruct the previous operating state.
The commercial comparison therefore includes the surrounding system. Battery backup, capacitors, protection circuitry and maintenance all carry costs. MRAM can reduce some of those requirements where its capacity, endurance and price fit the application. The memory itself still needs to be integrated into a correctly designed system.
Everspin can build an attractive business by winning these specific jobs. Its opportunity depends on how much customers value reliable persistence, how easily they can adopt it and how much of that value Everspin can retain in its margins.
IBM provides a concrete customer reference
In April 2024, Everspin announced that IBM selected its PERSYST 1Gb STT-MRAM for FlashCore Module 4. Everspin described the relationship as spanning successive generations of IBM storage products. The specified device provides 2.7GB per second of read and write bandwidth together with persistent storage.
The useful detail is what the component does. It helps protect critical information when power is lost inside an enterprise storage product. That is a clearly defined engineering requirement, with a named customer and an identified product generation.
For investors, this provides evidence that Everspin can meet demanding commercial requirements and remain relevant as a customer's architecture evolves. It also gives other storage designers a reference point when considering the technology. The announcement does not disclose IBM's current purchasing volume or guarantee a place in future generations.
AMD and Phison make adoption easier
A memory device becomes more useful when the surrounding controllers and development tools support it. Everspin's work with Xilinx, now part of AMD, addresses that integration problem. Its published design guide explains the use of 1Gb STT-MRAM with a Xilinx DDR4 controller in the Vivado development environment. Everspin's current partner page also identifies its work with AMD on controller support.
Phison adds another route into storage designs through native support for Everspin's 1Gb STT-MRAM in its enterprise SSD controller lineup. Cadence provides MRAM controller and verification IP for customers developing custom chips.
My interpretation is that these relationships reduce the engineering effort required to evaluate and adopt Everspin. A smaller supplier benefits when its memory can be used through tools and controllers that customers already know. The revenue still depends on customers selecting the memory and shipping the resulting systems. These relationships establish integration support; they do not establish undisclosed purchases of Everspin memory for AMD processors or GPUs.
GlobalFoundries adds a separate way to earn revenue
GlobalFoundries has a different role in Everspin's business. The companies have worked together on STT-MRAM process development and manufacturing, including discrete products and embedded memory. Their development agreement was extended in 2020 to cover a future 12nm project; that historical announcement alone does not establish a current production ramp at that node.
Everspin's annual filing describes royalties on qualifying wafers sold or transferred by GlobalFoundries that use specified Everspin design information. This gives Everspin a way to earn revenue when its technology is incorporated into another customer's chip.
That model could increase the economic value of the company's research beyond its own packaged products. The important variables are which implementations qualify, how much production they reach and the royalties generated. A growing embedded MRAM market can help Everspin when its own intellectual property participates in the resulting designs.
MaxLinear gives the AI opportunity a clearer purpose
The August 2026 memorandum of understanding with MaxLinear is interesting because it names the work the companies intend to pursue. They will evaluate Everspin MRAM alongside MaxLinear's compression, encryption and hardware acceleration technology for AI servers. The proposed applications include metadata management, logs, write buffers and cache functions.
These are places where latency, frequent updates and recoverability can matter. The investment case rests on MRAM earning a useful position within the server architecture. The MOU is an evaluation framework, with commercialization still to be demonstrated.
There is also a relevant personnel connection. Vikas Choudhary, who leads MaxLinear's connectivity and storage business, joined Everspin's board in August. His responsibilities at MaxLinear give him direct experience with the customers and design decisions Everspin hopes to address.
I regard that combination of technical collaboration and operating experience as encouraging. The next evidence to look for is a validated solution, followed by a customer program and disclosed commercial traction. Until then, AI server revenue from this collaboration belongs in the potential upside of the investment case.
Space and defense provide demanding applications
Astro Digital offers another named example of product selection. In July 2026, Everspin announced that its 64Mb PERSYST STT-MRAM had been chosen as the primary fail-safe boot memory for an upcoming Raven bus mission in geosynchronous Earth orbit. The role is specific: preserve the information needed to initialize spacecraft electronics and support recovery.
A single mission does not establish a large revenue stream. It does show why a customer might value persistence and reliability enough to qualify a specialized memory product. Successful use can strengthen Everspin's case in subsequent programs, although each new design must earn its own qualification.
The September partnership with Teledyne HiRel could broaden access to this market. Teledyne plans to offer Everspin's 256Mb PERSYST STT-MRAM, with the agreement also covering 64Mb and 128Mb options. Customer availability is expected in the fourth quarter of 2026.
Teledyne's contribution includes screening, qualification documentation, procurement support and management of long product lifecycles. Those services matter to customers who must approve a component for equipment expected to remain in service for years. My view is that this can make Everspin easier to purchase and specify, which is a practical source of competitive strength beyond the memory's technical specifications.
Amentum and Microchip strengthen the US business
Everspin's April announcement described a $40 million agreement with a US prime contractor over approximately two and a half years. The work covers Toggle MRAM process technology, engineering and foundry services for defense applications.
The subsequently filed contract identifies that contractor as Amentum Services. It supports a prime contract issued by NSWC Crane. The exhibit specifies a $40 million ceiling, with funding provided through approved milestone purchase orders. This is a development and manufacturing program whose financial contribution depends on authorized work and execution.
The manufacturing connection is particularly relevant. Everspin signed an initial ten-year foundry agreement with Microchip in April, with provision for extensions. It plans to establish its magnetic manufacturing process at Microchip's Oregon facility while retaining its intellectual property and continuing its Chandler operations. Initial shipments from the collaboration are expected in the second half of 2027.
The planned capacity covers MRAM and magnetic sensor products. It adds a US source and ITAR wafer-processing capability for applications where supply continuity and domestic production matter.
I see the two agreements as complementary. Amentum provides contracted development work, while Microchip offers manufacturing capacity that can support future programs. Customers evaluating long-lived products have a stronger reason to consider a supplier when its production arrangements can support those timelines. The remaining task is to bring the additional line into operation and translate that capacity into profitable demand.
UNISYST could expand the range of customer designs
The UNISYST roadmap extends the commercial argument into unified code and data storage. Announced in March 2026, the family targets densities from 128Mb to 2Gb, with engineering samples expected in the fourth quarter of 2026. Everspin is aiming at systems that need persistent storage, fast updates and dependable operation in demanding environments.
The larger capacities could make the product relevant to more demanding embedded workloads. OpenMV's contribution to the launch announcement highlighted the challenge of storing and frequently updating AI models on edge devices. That supports a plausible application, although sampling and customer evaluation must still lead to production purchases.
Quintauris provides another route into future designs. Its collaboration with Everspin seeks to integrate MRAM into RISC-V reference architectures and platforms for automotive, industrial and edge applications. Quintauris brings together semiconductor companies including Bosch, Infineon, Nordic Semiconductor, NXP, Qualcomm and STMicroelectronics. Participation gives Everspin access to an influential development ecosystem; individual supply agreements with those companies would require separate evidence.
There is already a named automotive example. In 2022, Rimac Technology selected Everspin's 256Kb MRAM for the Nevera. That design win demonstrates a use for the technology in a demanding vehicle. Its greater relevance to the investment case is as a customer reference, since a specialist hypercar program does not establish mass-market automotive volumes.
Product revenue is beginning to support the argument
Second-quarter 2026 revenue reached $18.7 million, up approximately 42% from a year earlier. Product sales rose about 38% to $15.3 million. That matters because the increase extended beyond licensing and engineering transactions, which can vary substantially between quarters.
The company reported a 53.9% GAAP gross margin and $43.9 million in cash. Its third-quarter revenue guidance of $19.5 million to $20.5 million indicates that management expected further sequential growth when it reported in August.
Profitability requires more care. Everspin reported a $3.6 million GAAP net loss alongside $2.9 million of adjusted net income. Exclusions included roughly $4.0 million of litigation costs, $1.1 million of engineering fees associated with the foundry agreement, and stock-based compensation. The commercial progress is visible; durable earnings still need to follow.
The risks that deserve attention
The patent dispute with Avalanche Technology is material. Everspin's August filing disclosed a US International Trade Commission investigation and related litigation. The company disputes the allegations. Legal costs are already affecting results, and an adverse outcome could affect the economics or availability of implicated products.
Customer concentration also matters. Everspin reported that its two largest end customers represented 33% of 2025 revenue. A broad customer count can therefore coexist with considerable exposure to a few purchasing decisions.
Execution remains the other major risk. Teledyne availability, UNISYST samples and the Microchip line have announced schedules that still need to be delivered. Evaluation agreements can take time to produce sales. MRAM must also justify its cost against established memory alternatives for each application.
Finally, a stronger business does not automatically make the shares attractive at every price. A sensible valuation must distinguish repeat product revenue from development income and place a probability on programs that have yet to reach production. Treating all potential opportunities as certain would overstate the case.
Why I remain bullish
Everspin has an increasingly credible way to expand. The existing business gives customers evidence that the technology works in commercial products. Controller support reduces integration work. Teledyne can assist with qualification and procurement. Microchip can improve supply continuity. The newer product roadmap gives those relationships more memory configurations to offer.
The financial opportunity comes from a higher rate of adoption across that network. If additional designs enter production while existing customers keep ordering, Everspin could spread its development costs across a larger recurring product business. Licensing and specialized engineering can add to that value, provided investors recognize their different revenue patterns.
I would watch repeat product growth, progress from evaluation to production and the cost of the legal dispute most closely. Those measures will determine whether the improving commercial position translates into returns for shareholders.
The reason I find Everspin compelling is specific: it supplies memory for functions customers need to trust, and its ability to serve those customers is expanding. Sustained execution would make the business substantially more valuable.
