Reading Micron (MU) as “essential infrastructure for the AI era”: Can it sustain a leading-edge product mix while riding the waves of the memory cycle?

* This report is based on data as of 2026-03-19.

What this company does—in three lines a middle schooler could understand

Micron Technology (MU) makes and sells “memory” and “storage” components that sit inside electronic devices. In simple terms, it supplies memory (DRAM, etc.), which is like a “workbench” that typically gets cleared when the power goes off, and storage (NAND/SSD, etc.), which is like a “drawer” that keeps what’s inside even when the power is off. End markets are broad—PCs, smartphones, data centers, autos, factory equipment, and more—and MU sits in the “foundational capacity components” layer where required volumes tend to rise in an AI-driven world.

What it sells: two major product buckets

1) Memory (the “workbench”)

Memory is where a device temporarily holds data while it computes or “thinks.” Beyond server, PC, and smartphone memory, MU is also pushing deeper into ultra-high-speed memory for AI workloads.

• Data center memory (cloud, AI servers)
• PC memory
• Power-efficient memory for smartphones
• Ultra-high-speed memory for AI (HBM, etc.; discussed later)

2) Storage (the “drawer”)

Storage is where data is kept over longer periods, often delivered in SSD form factors. In AI use cases, read/write throughput can become congested, which increases the importance of high-speed data center SSDs.

• High-speed SSDs for data centers
• SSDs for PCs
• Storage components for embedded applications (automotive, industrial equipment, etc.)

Who buys: predominantly B2B customers

MU primarily sells to “companies,” not individual consumers. Data center operators, server OEMs, PC and smartphone OEMs, automakers and auto-parts suppliers, and industrial equipment makers purchase in large volumes.

How it makes money: scale manufacturing × specs that support ASP

As a B2B component manufacturer, MU produces memory and SSDs at scale and ships into large customer programs. Better specs—speed, power efficiency, capacity, reliability, and so on—generally support higher pricing. That said, memory and storage pricing is extremely sensitive to supply-demand: oversupply tends to pressure prices, while shortages tend to lift them. In recent years, higher-performance AI products have been areas where “technical capability” and “co-development with customers (design-in)” matter more, making differentiation more achievable than in purely commodity, scale-driven segments.

Earnings pillars by end market (what matters most today)

• Data center memory: A core pillar. As AI advances, not only the server count but also “memory per server” tends to rise.
• Data center SSDs: A growing pillar. In AI training and inference, read/write throughput can become a bottleneck, supporting demand for faster SSDs. MU has been increasingly highlighting next-generation offerings such as PCIe Gen6.
• PC and smartphone: A mid-sized pillar. The market is large, but it can be structurally harder than AI data centers to “sell higher performance at meaningfully higher prices.”
• Automotive and industrial equipment: A more stability-oriented pillar. Durability and long-term supply commitments matter, which tends to support steadier demand.

Strategic direction: shifting toward AI-oriented “advanced products,” and aligning the organization around that

MU is moving forward with a reorganization that brings product development and go-to-market closer to each customer segment, reflecting AI-era demand—especially data center-driven demand. Because performance requirements and commercialization paths differ by application (data center, mobile, automotive, etc.), this reads as an effort to run co-development and supply execution through a structure that stays close to customers.

Next growth pillars (important even if still small today): the “second floor” of the playbook

1) HBM (ultra-high-speed memory for AI)

HBM is a “special memory that can feed data to AI extremely quickly,” addressing a key AI performance bottleneck (data delivery speed). MU continues to highlight HBM3E adoption and volume production of next-generation HBM4. Also important: once HBM is designed into next-generation systems, relationships often become longer-duration.

2) New memory form factors such as SOCAMM (server integration innovation)

SOCAMM is “a new form factor that changes how memory is mounted, making it easier to use in AI servers.” MU has reported developments related to mass production of SOCAMM2 modules, which ties into environments where large capacity and power-efficiency requirements are especially demanding.

3) Next-generation data center SSDs (PCIe Gen6, etc.)

AI inference often requires rapidly ingesting large datasets, which can make storage I/O the limiting factor. MU has been positioning newer-standard SSDs early—an area that tends to be tightly linked to early customer validation and adoption races.

The “plumbing” behind future competitiveness: fabs and the supply network

For memory manufacturers, when demand is strong, “how much you can make” becomes a direct competitive advantage. In recent news flow, there has been discussion of a new fab in Singapore to expand 3D NAND production capacity, suggesting MU sees AI and data center demand as a long-term engine. At the same time, new fabs and expansions take time to ramp, so they are not, by themselves, a way to immediately relieve near-term supply tightness—an important nuance to keep in mind.

An analogy to make it intuitive

One way to think about MU is as a company that helps AI-era “compute factories” run smoothly by making the workbench (memory) bigger and the drawers (storage) faster—so the whole factory doesn’t stall.

Lynch-style “type”: MU is a Cyclical, with AI adding growth characteristics

Using Peter Lynch’s framework, MU most closely fits Cyclicals (economic cycle / supply-demand cycle). The reason is straightforward: profits swing materially with memory pricing and supply-demand dynamics, and annual EPS can move from losses to profits depending on the year.

At the same time, the recent recovery has been meaningfully supported by AI and data center demand, making it more reasonable to view MU as a hybrid of cyclicality and growth elements. The key isn’t “high growth” in the abstract; it’s where we are in the cycle and execution on ramping and supplying advanced products to protect its position in the next wave.

Long-term fundamentals: expanding, but not “smooth” (the waves are the point)

Revenue and EPS: long-term growth, but frequent sign flips

Revenue CAGR over the past 5 years is +11.8% per year, and over the past 10 years is +8.7% per year, pointing to long-term expansion. EPS CAGR is +26.1% per year over the past 5 years and +11.8% per year over the past 10 years, which can look like high growth at first glance. However, MU also includes years with negative annual EPS (for example, annual EPS of -5.34 in 2023). That pattern—“the average rises, but losses show up along the way”—is a defining feature of cyclicality.

Profitability (ROE and margins): recurring peaks and troughs

ROE in the latest FY is 15.76%. Annual operating margin, net margin, and FCF margin typically rise in upcycles and can fall into negative territory in downcycles, and the latest data support a “bottom → recovery” pattern. For example, on an annual basis, 2025 shows an operating margin of 26.4%, net margin of 22.8%, and FCF margin of 4.46%, consistent with recovery-phase profitability.

Free cash flow (FCF): it’s normal to swing from positive to negative

FCF CAGR is +82.2% per year over the past 5 years, but only +3.5% per year over the past 10 years. In other words, the picture changes dramatically depending on the time window—consistent with an industry where FCF can move sharply with the cycle. On an annual basis, there is a large negative in 2023, a smaller positive in 2024, and a positive in 2025 (annual FCF of $16.68bn), reflecting a “trough → rebound.”

Recent momentum (TTM and 8 quarters): acceleration consistent with a recovery phase

In the latest TTM, you see the “very large YoY growth” that typically shows up coming out of a trough. TTM EPS is 21.15, EPS growth is +408.38% YoY, revenue growth is +85.55%, and FCF growth is +675.19%.

The key is that while the latest TTM can make MU look like a “high-growth stock,” cyclicals often post extreme growth rates from “bottom → recovery.” So it’s more appropriate to read this not as a permanent re-rating of the business type, but as evidence that MU is in a recovery phase.

As a supplemental look at the last two years (8 quarters), EPS, revenue, net income, and FCF are all moving higher with strong consistency. That points less to a one-time spike and more to an ongoing uptrend.

Keep in mind that FY (annual) and TTM cover different time periods, so the same metric can look meaningfully different. For example, annual FCF margin is 4.46% in 2025, while TTM FCF margin is 19.41%; this should be treated as a period-timing effect.

Financial health: for a cyclical, can it survive the trough?

For cyclicals, because profits and cash flow are expected to compress in downturns, leverage, interest coverage, and cash buffers matter a lot. In the latest FY, D/E is 0.28, net debt / EBITDA is 0.27, and the cash ratio is 0.90. Interest coverage is 21.26x, indicating ample capacity to service interest.

These figures suggest that, at least today, the recovery does not look “borrowed” in the sense of being driven by heavy leverage, and from a bankruptcy-risk perspective they point to a relatively solid financial cushion. That said, memory is a capex-heavy industry with large investment cycles, and if the market turns after capex ramps, cash flow can deteriorate quickly. The ongoing monitoring focus, therefore, is not just “current health,” but “resilience during an investment expansion phase.”

Cash flow quality: how closely EPS and FCF line up, and where the investment load shows up

Latest TTM free cash flow is $11.279bn and FCF margin is 19.41%. Cash conversion is strong alongside earnings, which reduces the risk that the momentum is purely “on paper.”

At the same time, semiconductor memory is an industry where FCF can swing due to capex and working capital. As a recent indicator, CapEx/operating CF is 0.45, implying investment is ongoing but currently within a range that operating cash flow can support. To separate temporary investment-driven volatility from true business deterioration, it remains important to track FCF margin, the investment burden, and working-capital movements such as inventory.

Capital allocation: dividends are a supporting act, and the current burden looks light

MU pays a dividend, but it’s best viewed as unlikely to be the centerpiece of the investment case. Latest TTM dividend per share is $0.46228, and the report-date share price is $461.73. The latest TTM dividend yield cannot be calculated due to insufficient data, so it’s not possible to conclude here whether the yield is high or low.

That said, the “weight” of the dividend is clear in the numbers: latest TTM payout ratio is 2.19% on an earnings basis and 4.67% on an FCF basis, and FCF dividend coverage is 21.40x. At least in the latest TTM, the dividend does not appear to be crowding out growth investment (capex), so it’s reasonable to treat it as a supplemental component of shareholder returns.

On dividend growth, the 5-year CAGR of dividend per share is 18.46%, while the 10-year CAGR is difficult to assess due to insufficient data. The latest TTM dividend growth rate (YoY) is +0.22%, which is modest and suggests a period where dividend growth is not keeping pace with operating performance.

As a track record, the number of years with dividends is 16, consecutive dividend growth years is 1, and the most recent year in which a dividend cut (or an effective cut/cut-like outcome) occurred is 2024. For a cyclical name, this suggests dividends may be more exposed to the earnings cycle (without asserting the specific cause).

Regarding peer comparison, this material does not include data on differences versus peers, so no conclusion is drawn. As an industry characteristic, unlike stable high-dividend sectors, capex and market cyclicality tend to be major drivers.

In terms of investor fit, it’s unlikely to be a primary vehicle for income-focused investors, while for total-return investors it can be interpreted as “a light dividend load that preserves investment capacity.”

Where valuation stands today (benchmarked only to its own history)

Rather than declaring MU “cheap” or “expensive,” this section simply maps where it sits versus its own historical ranges in a neutral way (no peer comparison).

PEG (valuation relative to growth)

PEG is currently 0.05x, within the past 5-year normal range (0.03–0.09x) and roughly around the 5-year median. It is also within the past 10-year normal range and sits below the 10-year median (0.08x). Over the last two years, it has remained within the normal range while currently skewing toward the lower end.

P/E (valuation relative to earnings)

TTM P/E is 21.83x, above the past 5-year normal range (5.44–19.44x), putting it on the higher side within the past 5 years (around the top 20%). Over the past 10 years it remains within the normal range (6.15–30.89x), but sits above the 10-year median (13.22x). For cyclicals, P/E can look elevated in recovery phases and depressed near peaks, so it’s best interpreted alongside “where we are in the cycle,” not in isolation.

Free cash flow yield

TTM free cash flow yield is 2.17%, within the past 5-year normal range (-3.16%–10.88%) and around the median. It is also within the past 10-year normal range and is above the 10-year median (1.80%). Over the last two years, it has remained within the normal range while currently leaning toward the lower end.

ROE (capital efficiency)

ROE in the latest FY is 15.76%, within the past 5-year normal range (-1.27%–16.09%) and near the high end. It is also within the past 10-year normal range and slightly above the 10-year median (14.55%). The last two years show an upward trend (recovery direction).

Free cash flow margin (cash-generation quality)

TTM FCF margin is 19.41%, above the upper bound of the past 5-year normal range (9.06%) and also above the upper bound of the past 10-year normal range (14.25%). Relative to historical ranges, the simple fact is that current cash-generation strength stands out. The last two years are characterized by an upward trend.

Net Debt / EBITDA (financial leverage: lower generally means more flexibility)

Net debt / EBITDA in the latest FY is 0.27x, within the past 5-year normal range (-0.10–0.90x) and around the 5-year median. It is also within the past 10-year normal range, but above the 10-year median (0.08x)—i.e., somewhat higher net debt versus that longer baseline. Over the last two years, it has been trending downward (toward a smaller number).

Reading the six metrics together

• On multiples, P/E is above the past 5-year range, while PEG is within the past 5-year range and near the median.
• On cash generation, TTM FCF margin is above the past 5-year and 10-year ranges.
• Financial leverage (Net Debt/EBITDA) is within historical ranges, currently around the 5-year median.

This section is not a conclusion; it is simply a map of “where MU stands today relative to its own historical ranges.”

Success story: why MU has won (the essentials)

MU’s core value is its ability to supply—at scale, with high quality, and over long periods—the memory (DRAM) and storage (NAND/SSD) needed to keep computing systems running. End markets are broad and the products are highly essential, but standardized products can also circulate easily. As a result, advantage tends to come less from brand and more from supply capacity (fabs and equipment), scale and yield, speed of generational transitions, and manufacturing execution such as quality, power efficiency, and reliability.

For AI-oriented advanced products, co-development and qualification tied to customers’ next-generation platforms become even more important. That makes the ability to plug into customers’ design cycles and the ability to mass-produce and deliver as committed central to the winning formula. To borrow the framing in the material, MU is less “a company that monetizes narrative” and more “a company that earns its seat in the next standard through execution.”

Story durability: do recent developments reinforce the “winning formula”?

The market narrative has been shifting from “AI = GPU story” toward “AI = memory bandwidth/capacity and storage I/O are bottlenecks.” MU’s positioning fits that shift, which elevates the importance of data center memory and data center SSDs.

Just as importantly, attention is moving from specs alone to “volume ramp and supply.” The value of advanced products isn’t merely that they can be built—it’s that they can be ramped and shipped on schedule. In that context, MU’s emphasis on supply, quality, and manufacturing execution reads as broadly consistent with the facts organized here: “cyclical, but profitability and cash generation are strong in the recovery phase” (even though the industry’s cyclicality itself does not go away).

Quiet structural risks: what to watch most closely when things look strong

MU’s main risk is less “demand goes to zero” and more the “hard-to-see failure modes” that can emerge when “cycle × advanced-product competition × supply constraints” overlap. This section lays out those mechanisms (not as a claim of deterioration, but as a structural checklist).

• Upside risk of customer concentration: The more MU leans into AI-oriented advanced products, the more exposure can concentrate in mega-customers or specific platforms. If specs change, qualification slips, or pricing/terms reset, earnings volatility can increase.
• Fast-moving dynamics as advanced products become the main battlefield: In areas like HBM, delays can translate directly into “design wins” (inclusion in next-generation systems), not just share shifts. As competitors concentrate investment on AI, gaps can close quickly.
• Loss of differentiation (drift back toward commoditization): If MU fails to differentiate in high value-added areas, the commodity mix rises and the business can revert to “oversupply → price competition → margin compression.” The stronger the current phase, the larger the downside if differentiation erodes.
• Supply chain dependence: Bottlenecks in specific steps—such as advanced packaging or equipment—can create situations where MU “can’t ship despite demand.” Expansions like the new Singapore fab take time and do not automatically solve near-term tightness.
• Cultural deterioration shows up first as “delays”: Within the scope of this review, no primary information has been confirmed as a decisive sign of cultural deterioration. Structurally, however, cultural issues often appear in development, ramp, and customer-qualification delays before they show up in reported numbers.
• The “ceiling” in a recovery phase: Profitability can reverse as supply rises, demand cools, and prices fall. A common pattern is that as AI-related supply constraints ease, scarcity premiums fade and margins compress.
• Re-emergence of financial strain: Even with flexibility today, cash flow can weaken quickly if the market turns during a period of rising capex. Whether the company can withstand “investment expansion × reversal” is a less visible risk.
• Shortening of transaction practices: If shorter contract durations and more frequent price resets become more common, they can help in upcycles but hurt in downcycles—potentially amplifying earnings volatility.

Competitive landscape: oligopolistic structure, but outcomes are decided by each generation

Memory and storage have limited new entry, but results are driven less by brand and more by “generational transitions,” “yield,” “advanced packaging,” “customer qualification and ramp,” and “supply certainty.” Commodity products tend to commoditize, while in advanced products “only a limited set of companies can ship.” At the same time, delays can translate directly into missed revenue—creating a two-layer competitive structure.

Main competitors (specific names)

• Samsung Electronics (a massive player across memory)
• SK hynix (reported to have strong presence in AI-oriented areas such as HBM)
• Kioxia (a major NAND player)
• Western Digital (a major NAND/SSD player, also in the context of collaboration with Kioxia)
• Solidigm (a meaningful presence in data center SSDs)
• YMTC (China NAND; even under constraints, production increases and share targets have been reported, making it a potential exogenous variable for supply-demand)

Multiple sources also discuss the view that AI-oriented high-bandwidth memory such as HBM tends to become a de facto three-way competition among MU, Samsung, and SK hynix.

Winning vs. losing paths by domain (investor-oriented translation)

• High-bandwidth memory (HBM): Competitive axes include customer qualification, yield, advanced packaging, supply certainty, and power efficiency/thermal design. If multi-sourcing across the three suppliers strengthens, it may become less about a runaway leader and more about “allocation and ramp execution quality.”
• Server DRAM: Competitive axes include cost/GB, power efficiency, supply volume, and ramp aligned with generational updates.
• Data center SSDs: Competitive axes include controller/firmware maturity, reliability, performance consistency (QoS), customer qualification, and supply capacity. Mass production and shipment in the PCIe Gen6 generation tends to tie directly to adoption races.
• NAND (more commodity-like): Competitive axes include manufacturing cost, layer count and generational updates, supply discipline, and investment discipline. As an exogenous variable, supply increases from players such as YMTC can be a source of volatility.

Moat (Moat): real, but not “permanent”—the question is whether it carries across generations

MU’s moat is driven less by network effects or brand and more by a combination of massive investment and manufacturing know-how (entry barriers), advanced packaging and ramp execution (implementation barriers), and participation in customers’ design cycles (relationship assets).

However, because memory retains meaningful commodity characteristics, the moat is less a static castle wall and more about whether it can be sustained through generational transitions. Put differently, durability is determined by “whether it can keep winning seats in the next standard through on-schedule mass production.”

Structural position in the AI era: a tailwind, but the winning path isn’t software-like

• Network effects: Not a business built on network effects like social platforms. The winning path is closer to “staying designed into next-generation GPU/server platforms (ecosystem alignment).”
• Data advantage: Not about controlling user data; the advantage tends to come from accumulated manufacturing capabilities—process, quality, yield, and power efficiency.
• AI integration level: Not a software layer that uses AI to raise product value, but a component supplier embedded in AI compute infrastructure. Parallel rollout of HBM, AI-inference SSDs, and new modules is a move to deepen integration.
• Mission criticality: AI servers are often constrained by memory bandwidth/capacity and storage I/O, making advanced memory and advanced SSDs critical components where “performance can’t be achieved without them.”
• Barriers to entry: Rooted in massive capital requirements and advanced manufacturing execution. For advanced products, “mass-produce on schedule and hit design timelines” is the core of durability.
• AI substitution risk: Relatively low because MU isn’t a service AI replaces; required volumes tend to rise as AI proliferates. The main risks are cycle reversals and mix deterioration if MU loses advanced-product adoption battles.
• Position in the AI stack: Not the application layer, but the physical layer (hardware infrastructure). Efforts to align with next-generation ecosystems and ship HBM, SSDs, and modules together reinforce that positioning.

In short, MU isn’t “on the side being replaced by AI,” but “on the side that enables AI compute factories.” That said, the return profile isn’t as one-directional as many software businesses; it depends heavily on consistent mass production, supply, and adoption of next-generation products.

Management and corporate culture (the execution engine that matters for long-term investors)

CEO messaging: AI and data centers, with an emphasis on “volume production, supply, and quality”

MU’s management narrative is oriented around “continuing to supply memory and storage as essential infrastructure, with AI and data centers at the center.” Near-term messaging also places more emphasis on practical constraints—volume production, supply, and quality—than on aspirational spec targets, with a clear focus on repeatable execution.

Persona → culture → decision-making linkage

As products become more advanced, outcomes are increasingly determined by “repeatable execution that doesn’t miss generational transitions.” If leadership’s execution-first persona translates into culture, it typically produces an organization that treats manufacturing, quality, and supply certainty as core strategic KPIs. In periods of supply constraint, decisions about where to allocate limited supply (focus and prioritization) become more consequential—and those tradeoffs tend to be felt most acutely on the ground.

General patterns often seen in employee reviews (not claims, but structural context)

• Positive: Broad opportunities for technical learning; frequent cross-functional collaboration.
• Negative: High workload and pace, with uneven burden across teams; top-management evaluations tend to be polarized.

This should be read less as a “good/bad” verdict and more as a structural reality: in a business where supply-demand waves and advanced-product ramps coexist, employee experience is unlikely to be uniformly stable.

Governance changes (facts) and how long-term investors might interpret them

At the January 2025 shareholder meeting, MU announced a transition to a structure where the CEO also serves as Chair of the Board. In addition, multiple director retirements have been announced ahead of the 2026 annual meeting. CEO-Chair duality can improve speed and consistency of decision-making, while also putting more focus on how oversight is perceived; the conclusion depends on an investor’s governance philosophy.

Fit with long-term investors is described as strongest for those who believe in the business’s essential role while accepting earnings volatility, and who can monitor whether management maintains supply discipline, investment discipline, and alignment with customer design cycles.

The long-term investment backbone “in two minutes” (Two-minute Drill)

The foundation of a long-term view on MU is whether you buy the premise that “as AI servers expand, not only compute but also memory and storage will remain persistent bottlenecks—and the importance of advanced memory (HBM, etc.) and data center SSDs will continue to rise.” From there, the key isn’t demand strength by itself; it’s whether MU can not only “launch” next-generation products, but also “mass-produce and supply them on schedule,” while sustaining a high value-added mix.

The memory-specific supply-demand cycle still exists, and results can be volatile. So the investment case isn’t “straight-line growth,” but rather: “there are waves, and if MU can keep winning next-generation seats in each wave, it can participate in AI-era expansion.” Mispricing tends to show up in two directions—either “ignoring the cycle and valuing it like perpetual growth,” or “overweighting the cyclical label and underestimating the shift toward advanced-product mix”—depending on sentiment.

KPI tree: what drives enterprise value, and where bottlenecks can form

What we want to see as ultimate outcomes

• Earnings power (can it compound despite volatility?)
• Cash generation (can it fund investment and returns?)
• Capital efficiency (ROE, etc.)
• Cycle resilience (can it endure the trough without stopping investment and generational transitions?)

Intermediate KPIs (Value Drivers)

• Revenue scale (can it supply into strong markets?)
• Shipment volume (supply capacity) and supply certainty
• Product mix (commodity ↔ high value-added)
• Unit profitability (price vs. cost spread, volatile with supply-demand)
• Manufacturing execution (yield, generational transitions, stable mass production)
• Capex burden and timing
• Working capital efficiency (inventory turns, etc.)
• Financial cushion (debt, cash, interest-paying capacity)

Business-level drivers (Operational Drivers)

• Data center memory (revenue scale and mix, alignment with generational updates)
• Ultra-high-speed memory for AI (HBM: high value-added mix and mass-production execution)
• Data center SSDs (relieving I/O bottlenecks, reliability and qualification)
• PC and smartphone (revenue base, commodity characteristics)
• Automotive and industrial (stickiness via long-term supply and reliability)
• Manufacturing capacity expansion and supply-chain strengthening (foundation for supply certainty and generational transitions)

Constraints

• Price volatility driven by supply-demand cycles
• Supply constraints (advanced products, mass-production capacity, supply network)
• Friction during generational transitions (customer qualification, compatibility, validation burden)
• Dependence on specific steps such as advanced packaging and equipment
• Capex burden (large scale)
• Customer concentration (volatility when specs change, qualification delays, or terms change)
• Shortening of contract/price reset cycles (changes in transaction practices)
• Organizational execution load (localized overload leading to delays)

Bottleneck hypotheses (investor monitoring points)

• Whether advanced-memory mass-production ramp and sustained supply are on plan (signs of delays or shortages)
• Whether the share of high value-added products is being maintained (signs of reversion toward commodity products)
• Whether friction around supply allocation is intensifying (surfacing customer dissatisfaction)
• Whether customer qualification/validation is getting stuck during generational transitions (stalled adoption refresh)
• Whether cash generation is being excessively pressured during capex upswings (tempo of investment vs. payback)
• Whether inventory turns are deteriorating (whether supply-demand distortion is showing up first in working capital)
• Whether the financial cushion is being maintained (durability in a reversal phase)
• Whether dependence on large customers/specific platforms is rising too far (volatility when terms change)
• Whether constraints in advanced processes (packaging/equipment) are becoming a bottleneck to supply certainty