Everspin Technologies, Inc. (MRAM)

The dominant pure-play MRAM vendor — the instant-state-save memory inside factory robots — with a genuine industrial-automation core and a new defense deal, but a tiny, lumpy P&L on the most extended tape in this batch.

Investment Research · Photoncap-style deep dive · v1 of "Everspin Technologies" · May 14, 2026

Two-layer framing — read this first

This name demands the two-layer discipline up front, because the tape and the story diverge sharply. Layer one — the tape: MRAM sits +173.7% above its 50-day moving average with an RSI of 78.0. The +173.7% figure is among the most extreme extensions in the entire coverage universe — a near-tripling relative to the 50-day mean. Extensions like that are momentum, but they are also the setups that mean-revert most violently when the catalyst flow pauses. Layer two — the thesis: the underlying business has a genuine kernel — Everspin is the real dominant pure-play in a real (if small) memory category, industrial automation is a confirmed core end-market, and the recently announced $40 million defense subcontract is a tangible, dated piece of news, not vapor. The job here is to take that kernel seriously while refusing to let it excuse chasing a parabolic chart. The conclusion, built out below, is Bucket D: the move has played out, the fundamentals are too small and too lumpy to support the current price, and the only defensible posture is a deep-pullback watch stance. Anyone tempted to buy MRAM near $39 because "robots need state-save memory" should re-read this paragraph first.

What Everspin physically does

Everspin makes MRAM — magnetoresistive random-access memory — a non-volatile memory technology that stores each bit in the magnetic orientation of a tiny structure called a magnetic tunnel junction rather than in an electrical charge. The practical consequences of that physical difference are the whole investment case: MRAM is non-volatile (it keeps its data with the power off, like flash), but it is also fast and effectively endless in write-endurance (like SRAM or DRAM), and it is radiation-tolerant. Everspin ships two families — Toggle MRAM, the established, extremely robust generation, and STT-MRAM (spin-transfer torque), the higher-density modern generation — plus it has announced the UNISYST MRAM family aimed at the high-density standalone NOR-flash-replacement market.

Why does this matter for robotics specifically? The killer application in a factory robot or an industrial automation system is instant state-save on power loss. When a programmable logic controller, a robot arm controller or an automated machine loses power — a fault, an e-stop, a brownout — it has milliseconds to write its current operating state somewhere non-volatile so it can resume cleanly rather than restart from zero, potentially mid-motion, in a way that could damage product or hurt a person. Conventional flash is too slow to write in that window; battery-backed SRAM works but adds a battery that ages and fails. MRAM writes fast enough to capture the full machine state in the available window and holds it with no battery. That is why MRAM has a genuine, defensible home in industrial automation — and as factories deploy more robots, the attach grows. The honest caveat, developed below, is that this is a small, niche, lumpy memory market, and Everspin's whole company does roughly $60 million of annual revenue — the robotics-and-automation thesis is real at the use-case level but the scale is tiny.

Product roadmap

Everspin's roadmap runs across three product directions. The established core is Toggle MRAM — the extremely robust, well-qualified generation that has been the company's bread and butter in industrial, automotive and aerospace/defense applications for years, valued for reliability and radiation tolerance over density. The modern density leg is STT-MRAM, spin-transfer-torque MRAM, the generation that competes on capacity and is qualified into automotive and industrial reliability standards on 22nm and 28nm flows. The newest direction is the UNISYST MRAM family, which Everspin has said is expected to expand its addressable market by roughly $3 billion by targeting high-density standalone NOR-flash replacement — a deliberate push beyond the company's niche-state-save heartland into a much larger memory category.

Alongside the product families, the strategic-roadmap event of 2026 is the defense subcontract: in its Q1 2026 disclosures Everspin announced a new 2.5-year, $40 million agreement under which it acts as a subcontractor on an existing prime contract, providing Toggle MRAM process-technology capabilities and engineering services for U.S. defense-industrial-base customers. That is a roadmap item as much as a revenue item — it deepens Everspin's defense entrenchment, where MRAM's radiation tolerance is genuinely differentiated.

What Everspin does not make is mainstream memory — it does not make DRAM, NAND flash or the high-volume commodity memory that fills phones and PCs and data centers. It is a specialty-memory company. Its roadmap risk, addressed in the terminal-risk section, is that the foundries (TSMC, GlobalFoundries, Samsung) offer embedded MRAM as IP within their own process design kits, which is a different go-to-market than Everspin's standalone-chip model.

The financial print

Everspin reported Q1 2026 on April 29, 2026: total revenue of $14.9 million, up 14% year-on-year and at the high end of guidance. The composition matters — MRAM product sales were $14.1 million, up 28% year-on-year, while the licensing, royalty and patent line fell to $0.8 million from $2.1 million in the prior-year quarter, which is the lumpiness in action. GAAP gross margin was 52.7%, up from 51.4%, and non-GAAP net income was $2.6 million, or $0.11 per diluted share. Management attributed the strength to industrial automation, transportation and data-center applications. Guidance for Q2 2026 is revenue of $15.5-16.5 million, with GAAP EPS guided to a net loss of $0.07-0.12 and non-GAAP EPS from breakeven to a $0.03 profit.

Put the scale in perspective: this is a company with roughly $60 million of annualized revenue and a market cap of about $918 million — a price-to-sales multiple around 15x and a forward P/E around 84.2x, for a business growing its product line in the mid-20s-percent range but with a total top line measured in tens of millions and a profit line that flickers between small profit and small loss quarter to quarter. The $40 million defense subcontract spread over 2.5 years is about $16 million of cumulative revenue — meaningful against a $60 million base, which is exactly why the stock reacted, but it does not transform the company into something the current valuation comfortably fits.

The binary event is Q2 2026 earnings, expected late July 2026 (Everspin reports on a roughly quarterly cadence; July 29, 2026 is the working estimate, to be confirmed). For a name +173.7% above its 50-day line, the next print is a high-stakes risk event — anything short of continued acceleration against that setup invites a sharp unwind.

Customer mix today

Everspin discloses by end-market application, not by named customer percentage. The structural picture from the Q1 2026 disclosures: the MRAM product business is driven by industrial automation, transportation and data center, with industrial automation the core and the most relevant to the robotics theme. Defense is a fourth and now-expanding leg via the $40 million subcontract. Everspin has not published a "Customer A is X%" breakout, and given the small revenue base, customer concentration is a real but undisclosed risk — a memory company doing $15 million a quarter very likely has meaningful exposure to a handful of large industrial and defense accounts.

The 2024-to-2026 change worth highlighting: the mix is tilting toward the higher-value, higher-margin applications. MRAM product sales grew 28% year-on-year in Q1 2026 while the lower-quality licensing/royalty line shrank — so the revenue is becoming more product-driven and arguably higher-quality, even as the absolute scale stays tiny. Within "industrial automation," factory robotics and automated machinery are a genuine and growing slice, but Everspin does not size robotics as a distinct line. The honest framing: industrial automation as a category is confirmed core and contains real robotics demand; robotics specifically is a sub-component the company does not break out; and the customer base is concentrated enough — though undisclosed — that the lumpy quarter-to-quarter revenue pattern is partly a concentration symptom.

What's actually happening in industrial automation and defense

The mechanism to watch is twofold. On the industrial-automation side, the driver is the secular build-out of factory automation — every additional robot arm, every additional PLC, every additional automated machine is a potential MRAM socket for instant state-save, and Everspin's Q1 2026 commentary explicitly named industrial automation as a strength. This is the cleanest robotics read-through: more factory robots, more state-save memory, and Everspin is the dominant pure-play supplier of it. The catch is that this is a slow-compounding, GDP-plus kind of growth, not a hypergrowth ramp — it shows up as the mid-20s-percent MRAM product growth, not as a step-change.

On the defense side, the $40 million, 2.5-year subcontract announced with Q1 2026 results is the concrete event. Everspin acts as a subcontractor to a U.S. prime, supplying Toggle MRAM process technology and engineering services for defense-industrial-base customers. This is where MRAM's radiation tolerance is genuinely irreplaceable — mission-critical defense, satellite and aerospace systems need memory that survives radiation and holds state, and that is structurally Everspin's strongest competitive position. Be specific and skeptical, though: the prime contractor is unnamed, the $16-million-or-so cumulative revenue is spread thinly over 2.5 years, and a single subcontract — however welcome — does not by itself justify a near-tripling of the stock relative to its 50-day average. The defense deal is real and good; the market's reaction to it is the part that requires discipline. The qualification-and-ramp timelines for both legs are gradual — this is not a business with a single dated catalyst that re-rates it; it is a slow grind that the tape has gotten far ahead of.

The competitive threat / the foundries and Avalanche

Everspin's competitive picture has two distinct fronts. The first is other MRAM suppliers. Avalanche Technology is the named direct pure-play competitor, and in March 2026 Avalanche reported progress scaling STT-MRAM magnetic-tunnel-junction cells toward 64Gb-128Gb space-grade products for defense, satellite and aerospace — directly contesting Everspin's strongest niche. The much larger structural competitors are the foundries: TSMC, GlobalFoundries and Samsung all offer embedded MRAM (eMRAM) as IP within their process design kits, so a chip designer who wants MRAM-like non-volatile memory can increasingly get it embedded on a foundry process rather than buying a standalone Everspin chip. The top of the broader STT-MRAM market by the market-research tallies is Samsung, TSMC, SK Hynix, Micron and Intel — giants for whom MRAM is a feature, not a company.

The bear-case-via-competitor is structural: Everspin is the dominant pure-play, but "pure-play" in a category that the foundries can offer as embedded IP is a precarious position. Everspin's defense is genuine — it holds deep MRAM-specific process IP, has the longest qualification track record (which matters enormously in industrial, automotive and defense), and the standalone-chip model still wins where a designer needs MRAM but is not building on a foundry process that offers eMRAM. There is no IP litigation defining the competitive timeline. But the competitive reality caps the multiple: Everspin can be the best pure-play MRAM company in the world and still be a sub-$100-million-revenue niche supplier, because the addressable market for standalone MRAM is itself bounded.

The terminal risk

The terminal risk is that MRAM stays a permanent niche — and that the standalone-pure-play slice of that niche gets squeezed from two sides. From below, conventional memory keeps getting cheaper and "good enough": for many applications, slower flash plus a supercapacitor, or battery-backed SRAM, remains the cost-competitive answer to state-save, and MRAM's premium is only justified where the timing and battery-free constraints are hard. From above, embedded MRAM on foundry processes (TSMC, GlobalFoundries, Samsung eMRAM offerings) absorbs the designs being built on those processes, removing them from Everspin's standalone-chip addressable market. The transition is not a single dated event — it is the slow, continuous evolution of where MRAM gets sourced. The named beneficiaries are the foundries and the conventional-memory incumbents.

Everspin's credible defense is that there is a durable, defensible core — industrial, automotive, aerospace and defense applications where qualification track record, radiation tolerance and the battery-free state-save guarantee are non-negotiable, and where Everspin's standalone parts and process IP genuinely win. The UNISYST family is the attempt to break out of the niche into the larger NOR-flash-replacement market, which is the bull's growth path. But the honest terminal-risk read is that MRAM has been "the memory of the future" for two decades and remains a small category — the risk is not that Everspin fails, it is that the company stays small and lumpy indefinitely while trading at a valuation that requires it to break out. That, combined with the +173.7% tape, is what drives the Bucket D classification.

Bull / Gap / Optionality (Photoncap framing)

1. Everspin is the genuine dominant pure-play in a real memory category. It is the recognized MRAM frontrunner with the broadest Toggle-and-STT portfolio and the longest qualification track record — in industrial, automotive and defense, where qualification history is decisive, that incumbency is a real moat within the niche.

2. Industrial automation is a confirmed core end-market with a real robotics read-through. Management explicitly named industrial automation as a Q1 2026 strength, and the instant-state-save use case in factory robots and automated machinery is genuine and grows as factory automation scales — this is the cleanest robotics-thesis link in the name.

3. The $40 million defense subcontract is concrete, dated and entrenching. A 2.5-year subcontract supplying Toggle MRAM process technology to the U.S. defense industrial base deepens Everspin's position exactly where MRAM's radiation tolerance is irreplaceable — and at roughly $16 million cumulative against a ~$60 million revenue base, it is materially additive.

4. The revenue mix is becoming higher-quality. MRAM product sales grew 28% year-on-year in Q1 2026 while the lower-quality licensing/royalty line shrank — the business is becoming more product-driven, with GAAP gross margin improving to 52.7%.

5. The UNISYST family is a credible attempt to break out of the niche. Everspin's stated ~$3 billion addressable-market expansion via standalone-NOR-flash-replacement is the growth optionality — if UNISYST gains traction, the "permanent small niche" terminal risk is partly answered.

Gap

1. The tape is +173.7% above the 50-day moving average — among the most extreme in the universe. A near-tripling relative to the 50-day mean is a setup that mean-reverts violently when catalyst flow pauses. Buying here is buying at the most dangerous point of a parabolic move — the entry-timing risk alone disqualifies a full position.

2. The business is tiny and lumpy. Roughly $60 million of annualized revenue, a profit line that flickers between small profit and small loss quarter to quarter, and a licensing line that swung from $2.1 million to $0.8 million year-on-year. A ~$918 million market cap and ~84x forward P/E on that base prices a breakout that has not happened.

3. The foundry-eMRAM and conventional-memory squeeze is structural. TSMC, GlobalFoundries and Samsung offer embedded MRAM as IP; cheaper flash-plus-supercapacitor and battery-backed SRAM remain "good enough" for many state-save needs. Everspin's standalone-pure-play addressable market is bounded from both sides.

4. Customer concentration is a real, undisclosed risk. A memory company doing ~$15 million a quarter almost certainly leans on a handful of large industrial and defense accounts; Everspin does not disclose the breakout, and the lumpy revenue pattern is partly a concentration symptom.

5. MRAM has been "the memory of the future" for two decades and stayed niche. The terminal risk is not failure — it is permanent smallness. The valuation requires a breakout the category's twenty-year history has not delivered.

Optionality

The trade

MRAM is a Bucket D name: the near-term move has, on every technical measure, played out, and the fundamentals — a ~$60 million-revenue, lumpy, niche memory company trading at ~84x forward earnings — do not justify chasing a stock sitting +173.7% above its 50-day line. The disciplined posture is watch-and-wait. The entry zone is set deliberately deep at $29-34 — roughly 13-26% below the current $39.15 — because the two-layer framework requires it: a name this extended with no near-term catalyst to defend the price is precisely where the entry zone must be a real reversion target, not "current ± 5%." Size at 0.5% of risk capital at most, and only on a genuine pullback into that zone — a probe position in a tiny, volatile specialty-memory name, never a core holding. Set a stop near $26 (below the deep-entry zone, with room for the volatility this name carries). The catalyst is Q2 2026 earnings around July 29, 2026 — which, for a name this extended, is as much a risk event as an opportunity. There is no cleaner listed pure-play substitute for the MRAM-in-robotics thesis — Everspin is the pure-play — so the honest pivot is that the broader, cleaner way to own factory-automation-driven robotics demand is through the automation and compute names in the theme rather than through a sub-$100-million-revenue memory niche on a parabolic chart. MRAM's role in a robotics book is a deep-pullback-only watch item, not an initiate-now position. Conviction: 4 / 10.

Sources referenced inline throughout. Reference v1 of this template format: _Watchlist/hanmi-photoncap-style.md.

\newpage