Sanmina Corporation (NASDAQ: SANM) — A Thin-Margin Contract Manufacturer Re-Rated on One Customer’s AI Servers

Independent equity research · Date: July 3, 2026 Price (07-02-2026): ~$219.77 · 52-wk range: $95.79–$282.72 · ATH: $282.72 (06-03-2026), −22% off high Diluted shares: ~55.1M · Market cap: ~$12.1B · Net debt: ~$0.6B · EV: ~$12.7B Sector: Information Technology — Electronic Manufacturing Services (EMS)/ODM · GICS sub-industry: Electronic Manufacturing Services Segments: IMS — Integrated Manufacturing Solutions (incl. Sanmina AI Group / ZT Systems) + CPS — Components, Products & Services · FYE: late Sept/early Oct (52/53-wk) · Reporting currency: USD · CIK: 0000897723

⚡ Claude’s Take

This block is the author’s own subjective opinion and general information only — not investment advice and not a recommendation to buy or sell any security. The analysis that follows carries no recommendation and no price target; the sole directional view in this article is contained in this block.

Verdict: HOLD / trim-into-strength for existing owners · AVOID adding at ~$220 · NOT-A-SHORT. Accumulate only on a cycle-driven pullback into the ~$120–150 zone (≈ 12–15× a normalized ~$10 of blended EPS, ~0.6–0.7× EV/sales). Conviction: medium.

Tag: “The pick-and-shovel guy, re-rated on one miner’s gold.”

Sanmina is a genuinely well-run, 45-year-old contract manufacturer that just pulled off the acquisition of its life — buying ZT Systems’ data-center server manufacturing business from AMD for $3B and, in one stroke, doubling revenue toward a $14B FY26 run-rate and putting a credible $16B-plus FY27 target on the board. Management (founder-CEO Jure Sola) has executed cleanly: the balance sheet is strong (net leverage ~0.56×), the integration is ahead of plan, core Sanmina is compounding ~7% with a 1.1+ book-to-bill, and the company has quietly retired ~22% of its shares over five years at single-digit multiples. That is the bull’s honest core, and it is why I am not short: this is a beat-and-raise AI-infrastructure name with visible, named revenue in a still-running capex cycle, and shorting that is how you get carried out.

But price is the whole argument. At ~$220 the stock trades at ~46× trailing GAAP earnings and ~5× book — its richest multiple in at least a decade (98th percentile of its own history), versus the 10–17× P/E and 0.3–0.5× sales it earned for the entire 2018–2024 window. The market is paying an AI-franchise multiple for what the evidence says is a thin-margin (~6% operating), no-moat, cyclical, working-capital-hungry assembler whose crown-jewel new business is ~100% dependent on a single customer’s — AMD’s — AI-accelerator share against an entrenched NVIDIA, is explicitly “the vast majority… consignment” (low-value pass-through revenue), and carries a ~6% margin that management concedes is “roughly in line with core Sanmina” — i.e., not accretive. Two tells sharpen the point: the “$10.75–11.35” EPS everyone quotes is non-GAAP and roughly double the ~$5.50–6.00 of GAAP the company will actually earn (the opposite, flattering direction from Celestica’s own-stock-swap gain), and ZT revenue is so lumpy that management guided it down 40%+ sequentially for the very next quarter after the Q2 blowout. The re-rating from ~$85 (May 2025) to a $283 June-2026 peak priced permanence onto a business whose defining feature is that it is re-competed at every silicon generation.

One fact from the filings sharpens the skepticism and is almost entirely absent from the bull narrative: on a pro-forma basis, combined revenue fell ~33% year-over-year in Q1 FY26 ($3.72B vs $5.54B) — Sanmina bought a business whose revenue had already collapsed ~65% off a ~$14B-annualized late-2024 hyperscaler-capex peak, and the celebrated “$5–6B” ZT target is a rebuild toward a fraction of a former peak, not organic growth off a stable base. And the purchase accounting confirms what was bought: of ~$1.9B, goodwill was just ~$276M — the rest is working capital. This is a levered, deal-driven bet on a volatile, single-customer, low-margin box-build operation, dressed up by the market as an AI franchise.

The asymmetry drives the call. Spot ~$220 sits near the top of my base case. A bear scenario — an AI-capex digestion year, or AMD’s MI-series simply not taking the GPU share the bulls extrapolate — compresses the multiple back toward the EMS base rate (~12–15× blended EPS) and lands the stock around ~$100–140 (≈ −35% to −55%), roughly where it traded eight months ago. The base case (ZT hits $6–7B, core keeps compounding, multiple settles ~18–22× forward non-GAAP) is roughly ~$200–250. The bull case (SANM diversifies ZT beyond AMD to NVIDIA/hyperscalers, margins tick up on vertical integration, super-cycle persists) is ~$300–360. You are being asked to pay for the bull and absorb single-customer-plus-cycle risk for the privilege. Bullish trigger: ZT wins named NVIDIA or a second/third hyperscaler platform at scale (breaking the AMD monopsony) and segment margin disclosed structurally >7%. Bearish trigger: AMD guides its AI GPU ramp down, a hyperscaler in-sources, or FY27 revenue slips below the $16B “floor.” Great execution, wrong price, late in the cycle.

📈 Stock Price Action — Five-Year Event Map

Factual price history; the price move is FACT, the attributed driver is INTERPRETATION. No recommendation, no target.

Over five years SANM has gone from a forgotten ~$36 EMS small-cap (early 2022) to a ~$283 AI-infrastructure momentum name (June 2026) — a roughly 7.5-fold move — before giving back ~22% in the last month. It sits today at ~$219.77, ~22% below its June 3, 2026 all-time high of $282.72 and ~2.3× its 52-week low of $95.79 (July 2025). The move is a two-part story: a slow, fundamentals-driven grind from 2022–2024 as buybacks shrank the share count and margins crept up, then a violent AI/ZT-Systems re-rating from mid-2025.

Cycle narrative. (1–2) For most of the period SANM was a value small-cap: it round-tripped the COVID supply-chain cycle and then compounded quietly — revenue drifting $6.7B→$8.9B→$8.1B, operating margin from ~4% to ~5%, and a share count falling from 69M to ~54M as management bought back stock at 6–12× earnings. The stock roughly doubled off the 2022 low on nothing more exotic than earnings and buybacks. (3–4) The regime changed in May 2025 with the announcement that Sanmina would buy ZT Systems’ data-center infrastructure manufacturing business from AMD for ~$3B, becoming AMD’s preferred new-product-introduction (NPI) manufacturer for cloud rack- and cluster-scale AI systems — instantly repositioning a sleepy assembler as an “AI infrastructure” play; the stock nearly doubled again into late 2025. (5) A winter pullback reflected macro/tariff noise and legitimate questions about how lumpy and how profitable the ZT revenue would be. (6) Those questions were steamrolled by the April 27, 2026 Q2 print — $4.01B revenue (+102% YoY), ZT at $1.88B (well above plan on pulled-forward AMD compute shipments), non-GAAP EPS $3.16 (+125%), and a reaffirmed march to $16B-plus in FY27 — which sent the stock from ~$130 to a $283 all-time high in six weeks. (7) The most recent leg down is profit-taking plus a sector-wide EMS selloff on July 2, 2026 (a “why-is-it-moving” note citing electricity-grid/data-center-power concerns), not a company-specific event. The tape today is a high-beta (β≈1.5) momentum name that has just had its first real air-pocket.

1. Executive Summary

Sanmina is a San Jose–based electronic manufacturing services (EMS) provider — a contract manufacturer that designs, builds, tests, and manages the supply chain for complex electronic hardware on behalf of original-equipment manufacturers (OEMs) across communications/cloud, industrial & energy, medical, defense & aerospace, and automotive end markets. Founded in 1980, it employs ~32,000–35,000 people across 21 countries. For most of its public life it has been exactly what EMS is: a thin-margin, low-return-on-sales, cyclical, working-capital-intensive business with gross margins of ~8–9%, operating margins of ~4–5%, and net margins of ~3%. The stock reflected that, trading at 10–17× earnings and 0.3–0.5× sales for the better part of a decade.

Two things changed the story. First, a multi-year program of disciplined buybacks shrank the share count from ~69M (FY2020) to ~54M (FY2026) — a ~22% reduction, executed almost entirely at single-digit-to-low-teens multiples — which, combined with slow margin improvement, roughly doubled the stock from its 2022 low on fundamentals alone. Second, and decisively, in May 2025 Sanmina agreed to acquire ZT Systems’ data-center server manufacturing business from AMD for ~$3B, funded largely with ~$1.4B of new term debt, and became AMD’s preferred NPI manufacturing partner for cloud rack- and cluster-scale AI infrastructure. That acquisition — branded internally the “Sanmina AI Group” — doubled the company’s revenue run-rate (from ~$2.0B/quarter to $3.19B in Q1 FY26 and $4.01B in Q2 FY26) and re-rated the equity from ~$85 to a $283 all-time high.

Management’s FY2026 guidance is for revenue of $13.7–14.3B (~+73% YoY), non-GAAP operating margin of 6.3–6.6%, and non-GAAP diluted EPS of $10.75–11.35, with a stated FY2027 target of $16B-plus in revenue. Execution to date has been strong and the balance sheet is healthy (net leverage ~0.56×; ~$3.7B of liquidity). On the evidence assembled here:

• Business quality: there is no durable competitive advantage in the Greenwald sense. EMS is a textbook no-moat industry; Sanmina’s edge is a real-but-modest bundle of scale, global footprint, and program-specific qualification/switching costs that reset at each design generation — not IP ownership, network effects, or captive demand. Its customers are larger and more capable than it is.
• Concentration: the entire re-rating rests on the ZT/AMD business, which is ~100% AMD today — management confirmed it shipped zero NVIDIA product in Q2 and “all… AMD technology.” This is a leveraged bet on AMD’s AI-accelerator share against a dominant NVIDIA, not a diversified franchise (yet).
• Financials: margins are structurally thin (~6% operating even with the AI mix) and ZT is explicitly “the vast majority… consignment” — low-value pass-through revenue that inflates the top line more than the profit. FY2025 free cash flow ($473M) was real but flattered by a large working-capital inflow (customer advances tied to the ZT ramp) and by an $850M receivables-factoring facility; EMS cash conversion is lumpy.
• The growth is acquired, and off a peak: the “+73%” is the optic of consolidating ZT. On a pro-forma basis, combined revenue fell ~33% YoY in Q1 FY26 — Sanmina bought a business whose revenue had already collapsed ~65% off a ~$14B-annualized 2024 peak, and the purchase-price allocation (goodwill just ~$276M of ~$1.9B; the rest working capital) confirms it paid for a low-margin box-build operation, not franchise value — funded by taking net-cash to $2.2B of secured debt.
• Earnings quality: the widely-quoted “$10.75–11.35” is non-GAAP and roughly 2× the ~$5.50–6.00 of GAAP EPS the company will actually report (add-backs: SBC, acquisition/integration, amortization). Trailing GAAP P/E is ~46×; forward GAAP P/E ~37–40×; forward non-GAAP P/E ~20×.
• Industry/cycle: AI-datacenter hardware assembly is late-boom in Marathon capital-cycle terms — every EMS/ODM (Celestica, Jabil, Flex, Foxconn, Quanta, Wistron, Supermicro) is adding server capacity into the same demand, precisely the condition that mean-reverted the 2000–02 telecom/optical build.
• Capital allocation: above the EMS base rate — self-funded growth, a value-accretive (so far) transformational acquisition, heavy buybacks at historically cheap prices, no dividend — with one flag: the newly-authorized $600M buyback is being deployed at the richest multiple in the company’s history.
• Valuation: priced as an AI compounder (~46× trailing GAAP, ~20× forward non-GAAP, ~5× book — all ≥97th-percentile of its own history) at a ~2–4× premium to its decade-long multiple. The embedded expectations require ZT to reach ~$6–7B, margins to hold-or-rise, and a premium multiple to persist — a stacked bet with asymmetric downside.

The analysis that follows takes no position and sets no price target; the single position in this article is the labeled opinion block above. It lays out the evidence so the reader can weigh durability against price.

2. Business Overview

What Sanmina does. Sanmina is a global EMS provider. In the classic EMS model, an OEM — a networking-equipment vendor, a medical-device maker, a defense prime, or a hyperscale cloud operator — owns the product design and outsources the physical build to a contract manufacturer. Sanmina buys the components (often with the customer dictating the key parts, such as a GPU or switch ASIC), fabricates printed circuit boards and backplanes, assembles and tests full systems, manages the supply chain, and ships. Historically this is a low-value-added activity: the economic rent in an electronics value chain sits in the silicon and the software, not in the assembly. Sanmina earns a thin conversion margin on the labor, engineering, and supply-chain value it adds.

Sanmina distinguishes itself within EMS in three ways worth naming. First, it is more vertically integrated than most peers — it owns component operations (advanced printed circuit boards and backplanes in North America and Singapore, precision machining, cable assemblies, plastics, optical modules, RF, memory/storage, microelectronics) that let it capture more of the bill of materials in-house. Second, it has a deep, high-mix, regulated-market franchise in medical, defense & aerospace, and industrial/energy — sticky, qualification-gated programs with long design-in cycles. Third, since May 2025 it owns ZT Systems, a design-and-integration-led data-center server business that pushes it up the value chain toward ODM/JDM (original/joint-design manufacturing) for AI infrastructure.

Two reportable segments.

• IMS — Integrated Manufacturing Solutions (Q2 FY26: ~$3.58B, ~89% of revenue). The core EMS business: printed-circuit-board assembly, system assembly and test, direct-order fulfillment, and — now — ZT Systems, the rack-scale AI server integration business. IMS revenue grew +123.5% YoY in Q2 FY26, of which Core Sanmina IMS was $1.70B (+6.0%) and ZT was $1.88B. IMS non-GAAP gross margin was 8.5%.
• CPS — Components, Products & Services (Q2 FY26: ~$461M, ~11% of revenue). Higher-value component and product operations: PCB fabrication, optical/RF/microelectronics, defense/aerospace products, memory and storage platforms, cloud-based manufacturing-execution software, and after-market repair/logistics. CPS non-GAAP gross margin was 11.6% — meaningfully above IMS — making it the margin-richer, though smaller and slower, half of the business. CPS margin dipped in Q2 on depreciation from new-program investment and a component-shortage timing issue.

End markets (Q2 FY26). Sanmina now reports demand by end market: Communications Networks & Cloud/AI Infrastructure ~$2.77B (~69% of revenue, +~280% YoY) — the ZT-driven engine, plus high-performance networking (IP switching/routing, optical systems and pluggables at 400G/800G and emerging 1.6T, broadband access, some 5G); and Industrial & Energy, Medical, Defense & Aerospace, Automotive & Transportation combined ~$1.24B (~31%, roughly flat). The concentration into Comms/Cloud/AI — from a historically well-diversified base — is itself a risk marker (the relevant section, the relevant section).

How it makes money. Sanmina earns a margin on value added — assembly, test, engineering, component fabrication, and supply-chain management. Gross margin runs ~8–9% (GAAP), operating margin ~5–6%, net margin ~3%. Revenue is program-based, not contractual annuity revenue: Sanmina wins a program, ramps it, ships for its life, and must re-win the next-generation program. It is sticky within a generation (you don’t re-qualify a manufacturer mid-program) but re-competed at each new silicon node — the crucial distinction from a recurring-revenue franchise. A meaningful and growing share of the AI business is consignment, where the customer owns the expensive components (GPUs, memory) and Sanmina is paid a conversion fee — which lowers reported revenue and inflates margin percentage while capturing less absolute profit per system.

Corporate. Headquartered in San Jose, CA; founded 1980; IPO’d April 1993. Founder Jure Sola is Chairman & CEO; Jon Faust is CFO. The company operates an India manufacturing joint venture (the minority-interest line in the P&L). It is a US domestic filer (10-K/10-Q), not an ADR, and pays no dividend.

3. Industry Dynamics

The value chain and where EMS sits. Electronics manufacturing is a layered value chain, and the economic rent is distributed extremely unevenly across the layers. At the top sit the silicon/IP owners — NVIDIA (60%+ operating margins), Broadcom, Arista, AMD (40–65% gross margins) — who own the scarce, defensible technology. Below them sit the system designers, OEMs, and hyperscalers who own the product architecture, brand, and customer relationship. At the bottom sit the EMS/ODM assemblers — the ~3–6% operating-margin tier that Sanmina occupies — who buy components (largely pass-through) and earn a thin conversion fee on assembly, test, and supply-chain labor. Sanmina’s Q2 FY26 gross margin of ~8.8% and operating margin of ~5.7% sit squarely in the middle of that band: the financial signature of a business with no pricing power. Because materials flow through at cost, revenue is a poor proxy for economic value in EMS — a doubling of revenue on consignment work adds far less than a doubling of profit.

Greenwald test — no durable moat. Running the industry through the Competition Demystified framework:

• Supply/cost advantage: absent. Foxconn/Hon Hai is ~$210B+ of revenue (≈26× Sanmina) with structurally lower Asian cost; Sanmina has no proprietary process technology that peers cannot replicate.
• Customer captivity (switching costs): partial and shallow. Qualification, new-product-introduction (NPI), and dedicated-line switching costs are real mid-program, but they reset at every product generation, and large customers multi-source by policy. The interconnects are open standards (Ethernet, OCP), not proprietary lock-in.
• Economies of scale + captivity (the only genuinely strong moat type): absent for Sanmina — it lacks the scale leg entirely (it is sub-scale even among Western EMS) and its captivity is generation-limited.
• Share-stability / ROIC test: fails. EMS revenue is re-competed program-by-program; Sanmina’s own ZT book is one AMD re-bid away from a step-down. A low-teens ROIC in a fragmented industry with no share stability is the profile of a commodity, not a franchise.

Marathon capital-cycle read — late-boom, pre-digestion. The AI-server ODM/EMS segment is a textbook capital-cycle top in the making: every player is simultaneously adding AI-server/rack capacity — Foxconn (~40% of NVIDIA rack share, targeting 17,000–30,000 racks in 2026), Quanta (~25–30% of NVIDIA AI-server orders), Wistron/Wiwynn (Wiwynn revenue +164% in 2025), Supermicro, plus Celestica, Jabil, Flex, and now Sanmina/ZT. High returns in a commoditized industry are the precondition for mean reversion, not evidence of a moat — capital floods to the high returns and competes them away. The bottlenecks that do protect incumbents (CoWoS packaging, HBM memory, advanced optics) protect the chip owners, not the assembly layer. Industry data tracks hyperscaler capex rising from ~$256B (2024) → ~$443B (2025) → ~$600–725B (2026), but with capex growth decelerating sharply on the forward path (~+51% → +13% → +5% across 2026–28) and the capex-to-AI-revenue gap wider than the pre-2001 telecom bubble. The current AI margin/revenue surge is a cyclical tailwind riding a debt-financed capex peak, not a structural step-up — the assembly layer is precisely what de-rated 70–90% in the 2000–02 telecom/optical bust even though the end-demand was real.

Structural verdict: a structurally bad industry — fragmented, cyclical, low-margin, materials-pass-through, working-capital-intensive, with negligible barriers to entry and competition on price/scale/footprint. Sanmina occupies a slightly worse corner of it than Celestica: CLS’s design-led HPS networking earns ~8% operating margin and captures some content rent; Sanmina’s ZT is largely low-gross-margin server integration for a single customer. The one mitigant is Sanmina’s higher-mix legacy franchise in defense/aerospace, medical, and industrial — genuinely stickier, qualification-gated, higher-margin programs — but that is the small, slow half of the company.

4. Competitive Position

The moat question, answered directly: there isn’t one — there is a modest, real, but non-durable bundle of advantages. Sanmina’s competitive position rests on three things, none of which rise to a Greenwald moat: (1) scale and global footprint (21 countries, 4 continents, one integrated IT/MES backbone) that is real but dwarfed by the Asian giants; (2) program-specific qualification/switching costs that are genuine mid-program but reset each generation; and (3) vertical integration — Sanmina fabricates more of its own bill of materials (advanced PCBs and backplanes in North America and Singapore, precision machining, cable, plastics, optical modules, RF, memory, microelectronics) than most pure assemblers. Vertical integration is the most defensible of the three and the crux of the margin-expansion thesis: if Sanmina can pull ZT’s rack-integration work together with in-house components, cooling manifolds, busbars, and optics, it captures more BOM per rack. But “we can make more of the box in-house” is a cost/content edge, not a pricing-power moat — it can raise Sanmina’s share of a thin margin, not create a fat one.

If the moat can’t be tied to a financial outcome that would deteriorate without it, it isn’t a moat. Sanmina’s ROIC (low-teens) and operating margin (~5–6%) are indistinguishable from the EMS base rate. There is no financial outcome — no premium margin, no pricing power, no share stability — that would visibly deteriorate if the “moat” vanished, because the numbers already look like a commodity. The 15–35-year customer relationships management touts are real and valuable (they lower customer-acquisition cost and smooth the core book), but they coexist with ~5% margins, which is the proof that relationships in EMS are retention, not rent.

The AMD/ZT concentration — the load-bearing risk of the entire thesis. ZT is the data-center-infrastructure manufacturing arm carved out of ZT Systems and sold by AMD to Sanmina (AMD kept ZT’s design business). In Q2 FY26, ZT revenue was $1.88B, 100% AMD-based accelerated compute, with zero NVIDIA product shipped. The entire re-rating therefore rests on a single customer’s single GPU architecture — a bet on AMD out-executing NVIDIA in AI accelerators:

• AMD’s AI-GPU position is small and unproven at scale. NVIDIA holds ~75–86% of data-center AI accelerator share; AMD is ~5–7% today, with bulls projecting ~12–15% by late 2026 if the MI400 generation ramps (2nm, HBM4, anchored by a reported large OpenAI commitment and Meta “Helios” racks). AMD’s own framing is “double-digit share within 3–5 years.” If AMD’s AI share stalls, ZT’s $5–6B→$7B revenue path stalls with it — and Sanmina has zero control over that outcome.
• Single-customer, single-architecture concentration is worse than the peer set. Celestica’s 3-hyperscaler book (58% of revenue) is already extreme; Jabil’s largest customer is ~16%. ZT is one customer, one GPU family. JPMorgan initiated Neutral, explicitly flagging “AMD rack concentration risk”; Susquehanna Neutral cited “balanced risk-reward.” (These sell-side price targets are noted only as market context; this memo sets none.)
• The far larger NVIDIA ecosystem remains locked up — by others. The NVIDIA AI-server rack pool is dominated by Foxconn (~40%), Quanta (~25–30%), Wistron/Wiwynn, and Supermicro. ZT built an NVIDIA GB200-class rack but shipped no NVIDIA product in Q2. Management claims new next-gen accelerated-compute wins “with multiple hyperscalers and OEM customers” targeting ~September-2026 production — unproven and undelivered as of this writing. NVIDIA is also centralizing rack assembly with a few favored ODMs, a structural headwind to a new Western entrant. And ZT’s manufacturing is not proprietary — AMD can dilute Sanmina’s share by multi-sourcing, exactly as it retained the ability to do on the design side.

Direct peer contrast. Against the relevant set, Sanmina is a sub-scale #4–5 Western EMS: smaller than Foxconn/Quanta/Flex/Jabil, comparable to Celestica by legacy revenue but with lower margins and lower business quality. On EV/sales (~0.9–1.1×) it screens close to Jabil (~0.95×) and far below Celestica (~2.6–3.9×) — the market is not paying Sanmina a franchise multiple; it is pricing it as an assembler that bought a customer’s captive line. That is the correct read: ZT is volume, not rent. Verdict: no durable competitive advantage — a crowded, commoditized market with weak differentiation, in which Sanmina is a well-run but sub-scale participant now carrying an extreme single-customer concentration.

5. Growth History and Forward Opportunities

History. Sanmina’s revenue has been cyclical and roughly flat-to-modestly-growing over the pre-ZT window: $6.95B (FY20) → $6.74B (FY21) → $7.92B (FY22) → $8.94B (FY23, a cyclical peak) → $7.57B (FY24, a downcycle) → $8.13B (FY25). That is a ~4% five-year revenue CAGR with meaningful cyclicality — the signature of a business geared to its customers’ capex and inventory cycles, not a secular grower. What did grow steadily was per-share value: EPS rose from $1.88 (FY20) to $4.46 (FY25, diluted GAAP), driven about equally by margin improvement (operating margin ~3.6%→~4.9%) and the ~22% reduction in share count. This is a high-quality-capital-allocation growth story layered on a low-quality-organic growth base.

The ZT inflection. The acquisition changes the top-line trajectory abruptly: FY26 guided revenue of $13.7–14.3B is ~73% above FY25, and the FY27 target of $16B-plus implies the company roughly doubles in two years. Core Sanmina is guided to grow high-single-digits (7.3% in Q2 FY26, with a 1.1+ book-to-bill), while ZT contributes $5–6B in FY26 and a targeted $6–7B in FY27. Management frames a three-phase ZT plan — (1) integration (largely complete), (2) securing/continuing customer business (underway; new accelerated-compute wins already shipped), and (3) growth via vertical integration and platform expansion.

Forward opportunities (genuine). (a) AI compute — winning next-generation AMD platforms (the “300-series” and successors) and, critically, diversifying beyond AMD to other hyperscalers, OEMs, and neo-clouds; (b) vertical integration — pulling ZT’s system-integration work together with Core Sanmina’s in-house PCBs, enclosures, liquid-cooling manifolds, busbars, optical modules and memory to capture more BOM per rack (the real margin-expansion lever); © AI-adjacent infrastructure — power generation/distribution, transformers, and energy systems for data centers (industrial/energy segment); (d) high-performance networking — 800G→1.6T optical and switching; (e) defense/aerospace & satellite — a stable, growing, high-barrier franchise; (f) medical — steady, regulated, sticky.

Quality of the growth — mixed, and flattered by the acquisition-accounting optics. The core organic growth is real but modest and cyclical (high-single-digits at best). The ZT growth is enormous as reported but low-quality in four senses: (1) it is ~100% single-customer (AMD) today; (2) it is largely consignment (revenue-inflating, thin in absolute profit); (3) it is lumpy — management guided ZT revenue down from $1.88B (Q2) to $1.0–1.2B (Q3) purely on hyperscaler scheduling; and (4) most importantly, it is not organic growth at all — it is an acquired business that was itself in sharp decline. On a pro-forma combined basis, revenue fell ~33% YoY in Q1 FY26; ZT’s own quarterly revenue appears to have dropped from ~$3.5B (late 2024) to ~$1.2B (late 2025) before the Q2 FY26 rebound. The headline “+73% FY26 growth” is the accounting optic of consolidating an acquisition, layered over an underlying business that had contracted. High reported growth built on a thin-margin, pass-through, single-customer, recently-contracting acquired book is precisely the kind of growth that deserves a cyclical multiple, not a compounder multiple. Verdict: high-magnitude but low-durability, low-margin, acquisition-driven growth bolted onto a modest, cyclical core.

6. Financial Quality

Revenue & margins. The five-year P&L (GAAP) shows the EMS signature clearly:

Gross margin has crept from 7.4% to 8.8% over five years — real but glacial, and structurally capped by the pass-through nature of EMS. Tellingly, gross margin held ~8.8% even as revenue doubled with ZT — proof that the incremental ZT revenue is thin-gross (consignment) flow-through. Operating margin of ~5% (GAAP) / ~6.5% (non-GAAP) is good for EMS but low in absolute terms; ZT is roughly margin-neutral-to-slightly-accretive at the operating line (Q2 non-GAAP op margin rose ~80 bps YoY to 6.4% on operating leverage), not the fat-margin uplift the equity re-rating implies — management itself describes ZT margins as “roughly in line with core Sanmina.” ROIC on ROIC.ai’s calculation runs 9–15% (GAAP basis); management’s non-GAAP pretax ROIC of 34.7% (Q2 FY26) is a legitimately strong figure but rests on a denominator net of customer-funded working capital and on the flattering non-GAAP numerator — read it as “healthy, above WACC,” not as evidence of a moat.

Earnings quality — the GAAP/non-GAAP gap is the headline. Unlike Celestica (whose GAAP EPS is above adjusted because of an own-stock swap gain), Sanmina’s non-GAAP EPS is roughly double its GAAP EPS — the ordinary, flattering direction. FY26 non-GAAP diluted EPS is guided to $10.75–11.35; H1 FY26 GAAP diluted EPS was only $2.59 ($0.89 Q1 + $1.70 Q2), pointing to full-year GAAP diluted EPS around $5.50–6.00. The bridge is SBC (~$63M/yr and rising), acquisition/integration costs, and amortization of ZT intangibles. Neither number is “wrong,” but an investor quoting the ~20× “forward P/E” is using the non-GAAP figure; on GAAP the stock is ~37–40× forward. The ~$99M of “other non-operating income” in Q2 and ~$39M in Q1 (partly FX/other) further muddy the GAAP net line and should be normalized out of run-rate.

Cash flow. FY2025 operating cash flow was strong at $620M and free cash flow $473M (FCF/share ~$8.77), but flattered by a ~$189M working-capital inflow — including a ~$663M swing in “other operating assets/liabilities” that reflects customer advances/deposits funding the ZT inventory ramp. That is a real, recurring feature of the consignment/advance model (it lowers Sanmina’s own capital intensity), but it makes a single year’s FCF a poor run-rate guide: FCF has bounced from $44M (FY23) to $229M (FY24) to $473M (FY25) on working-capital timing. Q2 FY26 FCF was $342M on just $57M of capex. Capex is light (~1.5–2% of sales, ~$110–150M/yr), which is the one genuinely attractive structural feature — EMS is capital-light in fixed assets even as it is capital-heavy in working capital.

Balance sheet — strong, and the “negative equity” myth debunked. As of Q2 FY26: cash ~$1.58B; total debt ~$2.17B (up from ~$0.5B pre-ZT, reflecting the $1.4B acquisition financing); net debt ~$0.6B; net leverage ~0.56× (management basis); ~$3.7B liquidity (incl. an undrawn $1.5B revolver). Book equity is positive ~$2.42B (ex-minority), or ~$45/share — note that ROIC.ai reports a negative price-to-book, an artifact of a large accumulated-deficit line ($2.32B) offset by $6.75B of paid-in capital and $2.1B of treasury stock; the real book equity is positive, and AZI’s P/B of ~5.0× is the correct read. Tangible book is also positive (goodwill is only ~$359M). Inventory is elevated ($3.0B, net of customer advances) — the working-capital tell of a fast-ramping consignment business — with turns of 6.9× (improved YoY). Verdict: economics improve modestly with scale, but remain structurally thin, working-capital-intensive, and pass-through; the “AI margin uplift” thesis is not yet visible in the consolidated GAAP margins.

7. Capital Allocation

The historical record — genuinely good. For a decade Sanmina ran a disciplined, shareholder-friendly capital-allocation program that is the best thing about the equity. It generated steady free cash flow, made no empire-building acquisitions, and returned essentially all of it through buybacks executed at cheap prices: diluted share count fell from ~69M (FY2020) to ~55M (FY2025) — a ~20% reduction — bought overwhelmingly at 6–13× earnings and 0.3–0.5× sales. There is no dividend (a defensible choice for a cyclical). SBC is moderate (~$63M/yr, ~26% of GAAP net income) and more than offset by buybacks, so the count genuinely shrinks. On the Marathon/Greenwald scorecard this is well above the EMS base rate — management behaved like owners.

The ZT acquisition — transformational, and a departure in kind. In October 2025 Sanmina closed the largest deal in its history, and it changes the capital-allocation character of the company:

• Consideration: ~$2.4B cash + $150M in stock (1.15M shares) + up to $450M contingent earnout over three years — ~$3B headline. The purchase-accounting consideration booked was ~$1.9B (net ~$1.6B of cash acquired).
• What was actually bought: the purchase-price allocation is the tell. Of ~$1.9B, the vast majority is working capital (acquired AR ~$1.29B, inventory ~$1.30B) — goodwill was only ~$276M and identifiable intangibles only ~$51M (customer relationships $31M/10yr, software $16M/5yr, trade name $4M/1yr). Sanmina paid ~$1.9B mostly for a working-capital-heavy, low-margin box-build operation, not for franchise value. That is the accounting confirming the strategic read: ZT is volume, not rent.
• How it was financed — the balance sheet changed character. Sanmina put in place a $3.5B senior secured facility ($1.5B revolver + $2.0B Term Loan A + $800M incremental Term Loan B) and at close drew $2.2B ($1.4B TLA + $800M TLB), taking long-term debt from ~$300M (a near-net-cash company) to ~$2.17B, now secured by first-priority liens on substantially all assets. Net leverage is still modest (~0.56× on management’s annualized basis) because EBITDA rose too, but the company went from unsecured near-net-cash to secured-levered in one quarter — a material reduction in balance-sheet quality that the equity re-rating ignores.
• An off-balance-sheet feature to watch: ZT operates an $850M receivables-purchase (factoring) agreement with PNC, selling hyperscaler trade receivables — which flatters reported working capital and cash conversion. A legitimate tool, but a reason to treat the headline FCF as partly financed by receivables sales.

Buying back stock near all-time highs while levering up — the one clear flag. In Q2 FY26 Sanmina repurchased ~1.08M shares at an average ~$150 (~$162M); it spent ~$239M on buybacks in H1 FY26; and after the quarter it authorized a new $600M buyback (replacing a nearly-exhausted $300M program). So management is simultaneously (a) adding $2.2B of secured debt to fund an acquisition and (b) buying back stock at $150–265 — its richest-ever valuation, after a 5× run. The decade of cheap buybacks earned enormous credit; deploying a fresh $600M at ≥46× trailing earnings is the opposite discipline, and worth flagging even if the amounts are modest against the market cap.

Incentives — sensible, but now mis-calibrated to the new balance sheet. CEO Jure Sola earned ~$18.2M in FY2025 (~93% performance-based); CFO Jon Faust ~$4.8M. The annual incentive keys on revenue and non-GAAP operating margin, with a cash-flow-from-operations modifier — reasonable metrics for an operator, but there is no return-on-capital, no leverage, and no TSR-based metric. That is a notable gap now that the company carries $2.2B of new secured debt: management is paid to grow revenue and non-GAAP margin — exactly the metrics the ZT deal maximizes — with no incentive tied to the returns on, or the risk of, the capital deployed. Long-term PSUs vest on multi-year financial targets; ownership is aligned but light (all insiders ~3.1%).

Insider behavior — net selling into the run; no conviction buying. The Form 4 record over the past two years shows broad, heavy insider selling and essentially zero open-market accumulation. The only discretionary open-market purchase (code P) in two years was a token 700-share (~$100K) buy by a director. Founder-CEO Jure Sola sold ~$27M of stock in a single May-2026 clip at ~$229 (near the highs; no 10b5-1 flag visible), on top of steady option-exercise-and-sell activity; CFO Faust sold near the highs under a 10b5-1 plan adopted February 2026; multiple directors sold into the April–May 2026 strength. Insiders own only ~3.1% as a group (Sola ~2.8%). This is a neutral-to-negative signal — the people who know the business best are monetizing the re-rating, not adding to it. Verdict: a historically excellent capital allocator making one genuinely value-creating (if lower-quality) transformational deal, but now levering the balance sheet, buying back stock at record multiples, and paid on revenue/margin with no return-on-capital discipline — a step down in capital-allocation quality, even as the strategic logic is sound.

8. Changes and Headwinds — Last Two Years

The transformation is entirely the ZT/AMD deal, and the timeline matters:

• May 18, 2025: Sanmina signs the Equity Purchase Agreement to buy ZT Systems’ manufacturing business from AMD (AMD, which had bought all of ZT for $4.9B in 2024, kept the design arm and immediately shopped the manufacturing operations). Stock gaps up ~15%.
• Jul 29, 2025: New $3.5B senior secured credit facility entered to fund the deal.
• Oct 27, 2025: Deal closes. $2.2B drawn; PNC $850M receivables facility established; ZT consolidated into the IMS segment.
• Nov 2025 – Apr 2026: Three quarters of integration and a rapidly-rising narrative — FY25 print, first ZT-inclusive quarter (Q1 FY26 revenue $3.19B), then the Q2 FY26 blowout (Apr 27, 2026): $4.01B revenue, ZT $1.88B, non-GAAP EPS $3.16, FY26 guide $13.7–14.3B and a reaffirmed $16B-plus FY27 target. New $600M buyback authorized.

The headwind hiding in the pro-forma — ZT was bought off a peak, and its revenue had already fallen sharply. This is the single most important, least-appreciated fact in the filings. On a pro-forma combined basis (as if ZT were owned for the full prior period), Sanmina’s net sales were $3.72B in Q1 FY26 versus $5.54B in Q1 FY25 — a ~33% year-over-year decline. Backing out standalone Sanmina (~$2.0B), that implies ZT was running roughly $3.5B in a single quarter (~$14B annualized) in late 2024 and had fallen to ~$1.2B by late 2025. In other words, Sanmina acquired a business whose revenue had already collapsed ~65% off a 2024 hyperscaler-capex peak, and the “$5–6B FY26 / $6–7B FY27” targets are a rebuild toward a fraction of that former peak, not organic growth off a stable base. The Q2 FY26 surge to $1.88B — partly a pull-forward of H2 orders — should be read against that volatile history, not extrapolated. This is the empirical core of the cyclicality/lumpiness risk.

Other changes and headwinds:

• Customer concentration intensifying: top-10 customers rose from ~52% of sales (FY25) to ~65% (Q1 FY26) with the ZT/hyperscaler mix; the ZT book is ~100% AMD.
• Component shortages (memory, custom ASICs) flagged by management as a constraint “through the rest of the year… potentially 2027” — capping shipments even amid strong bookings (1.1+ book-to-bill).
• Balance-sheet character change: from unsecured near-net-cash to $2.2B secured debt (above).
• Segment reporting opacity: ZT sits inside IMS and is not (yet) a separately-audited segment, so its standalone margin and profit are not cleanly visible — the investor is trusting management’s “in line with core” characterization.
• Macro/tariff/geopolitical: the July 2, 2026 EMS-sector selloff (grid/data-center-power concerns) is a reminder that the stock now trades on AI-infrastructure sentiment. US-China decoupling/reshoring is a genuine tailwind to Sanmina’s US and diversified footprint.
• India JV: the Reliance-partnered India joint venture (entered 2022) drives a growing minority-interest deduction (~$20M/yr) below the line.

Verdict: the changes are transformational and, executed as management hopes, thesis-strengthening on the top line — but they simultaneously weaken business quality (concentration up, balance-sheet quality down, revenue quality down) and rest on a business whose own recent history is one of extreme volatility. On balance the last two years have made Sanmina bigger and more exciting but lower-quality and more concentrated.

9. Risk Analysis

Catastrophic-loss risk is low (strong balance sheet, positive tangible equity, capital-light fixed assets, real cash generation). The dominant risks are valuation and AMD-concentration — both of which are about the price paid and the durability of the AI narrative, not solvency.

10. Valuation Discussion — Embedded Expectations

The multiples, and the own-history context that is the whole story. At ~$219.77 (EV ~$12.7–12.9B), on trailing-twelve-month GAAP figures the stock trades at ~46× P/E, ~5.0× book, ~1.07× sales, and ~18× EV/EBITDA. Every one of those sits at the 97th–98th percentile of Sanmina’s own multi-year history (company multiples vs. its own 10-year history, as of 2026-07-02) — this is, unambiguously, the richest the stock has been in at least a decade. The contrast with its own past is stark: for the entire FY2018–FY2024 window Sanmina traded at 10–17× earnings, 0.27–0.50× sales, 1.1–1.6× tangible book, and 5–8× EV/EBITDA. As recently as FY2024 (ended Sept 2024) the stock changed hands around $68 at ~17× trailing earnings; it is now ~$220 at ~46×. Roughly two-thirds of the five-fold move from ~$45 (2022) to ~$220 is multiple expansion, not earnings.

Forward multiples — cheaper, but on a flattered numerator. On management’s FY26 guide (revenue ~$14B, non-GAAP diluted EPS $10.75–11.35), the forward non-GAAP P/E is ~20× and EV/sales ~0.9×; on the FY27 $16B-plus target, EV/sales falls toward ~0.8×. That “~20× forward” is the number bulls quote, and in isolation it does not look expensive for a company guiding ~73% revenue growth. But two adjustments matter: (1) the non-GAAP figure is ~2× GAAP — FY26 GAAP diluted EPS is tracking ~$5.50–6.00 (H1 was $2.59), so the forward GAAP P/E is ~37–40×; and (2) GAAP net income is itself flattered by large non-operating items — Q2 FY26 GAAP net income of $93.6M included ~$75.5M of “other non-operating income,” so trailing GAAP EPS (~$4.72) overstates the true operating earning power. Strip the non-operating gains and the operating earnings multiple is richer still. The honest summary: ~20× on flattering non-GAAP, ~37–40× on flattering-but-different GAAP, and higher on clean operating earnings — none of which is cheap for a ~6%-margin, no-moat assembler.

Peer comp set (approximate, mid-2026):

*Sanmina’s +73% is acquisition-driven (ZT), not organic — core Sanmina is guided to “high-single-digits.” Sources: ROIC.ai (2026-07-03);public peer data (CLS/JBL/FLEX/FN/TTMI filings); peer 8-Ks/macrotrends.

The table’s message: on EV/sales Sanmina (~0.9–1.1×) sits with the mid-pack Western assemblers (JBL ~0.95×), far below Celestica’s franchise multiple (~2.6–3.9×) — the market is not mistaking Sanmina for an AI franchise. But on trailing GAAP P/E and price-to-book it is at the top of both its own range and much of the peer set, and the Asian ODMs building the larger NVIDIA AI-rack pool trade at 12–18× forward for the same AI build. The market’s own cross-sectional verdict is that the assembly function does not deserve a franchise multiple; Sanmina’s premium-to-its-own-history rests on the bet that the ZT growth is durable and diversifiable.

Embedded-expectations analysis — what ~$12.7B EV requires. To justify ~$12.7B EV at a still-generous terminal ~15× P/E (a reasonable-for-EMS multiple, above the Asian ODMs), the market needs roughly ~$850M of sustainable net income, i.e., ~$15–16 of GAAP EPS at some out-year — about 3× FY25’s $4.46. Translated to operations, that requires ZT to reach ~$6–7B AND hold it (diversified beyond AMD), core Sanmina to keep compounding high-single-digits, consolidated operating margin to expand from ~6% toward ~7%+, AND a premium-to-legacy-EMS multiple to persist. The revenue leg is plausible on management’s own targets; the fight is margin durability, revenue quality (consignment vs. turnkey profit), the AMD-concentration tail, and multiple persistence — a stacked set of conditions, each individually reasonable but jointly demanding, with no room for an AI-capex digestion year at these multiples.

Scenario analysis (illustrative; not price targets):

• Bear (AI-capex digestion and/or AMD share stalls; ZT growth flattens; consignment caps profit; multiple re-rates toward the 12–15× EMS base rate on ~$8–10 blended EPS): a valuation zone roughly ~$100–140 (≈ −35% to −55%) — approximately where the stock traded eight months ago.
• Base (ZT reaches $6–7B, core compounds, margin roughly flat ~6–6.5%, multiple settles ~18–22× forward non-GAAP): ~$200–250.
• Bull (ZT diversifies to NVIDIA/multiple hyperscalers, vertical integration lifts margin >7%, super-cycle persists, premium multiple holds): ~$300–360.

Spot (~$220) sits at the top of the base case / bottom of the bull — the market is paying for the good outcome. (No price target, no BUY/SELL — these zones illustrate the embedded expectations only; the sole directional view in this document is the labeled Claude’s Take.)

11. Variant Perception

Consensus. SANM is an under-followed EMS name that has become a credible AI-infrastructure winner via the ZT/AMD deal — a “doubling in two years to $16B-plus” growth story with a strong balance sheet and a shareholder-friendly, buyback-heavy capital-allocation record, still “cheap” on ~20× forward non-GAAP earnings relative to other AI names. The Street has warmed rapidly (multiple new sell-side initiations in 2026, per the Q2 call’s roster of first-time questioners).

Bull case. The ZT acquisition is a genuine strategic coup that vaults Sanmina up the value chain into rack-scale AI at exactly the right moment; core Sanmina compounds high-single-digits with strong bookings; vertical integration lifts ZT margins over time; the company diversifies beyond AMD to NVIDIA and multiple hyperscalers; FY27 delivers $16B-plus and FY28 more; buybacks keep shrinking the count; and on FY27–28 non-GAAP EPS the stock is “only” mid-teens forward. In this world $220 is a reasonable entry into a multi-year compounding runway.

Bear case. This is a thin-margin, no-moat, cyclical assembler priced at its richest multiple ever on the back of a single-customer, consignment-quality, lumpy AI business that is a leveraged bet on AMD out-executing NVIDIA. If AI capex digests, or AMD’s GPU share disappoints, or the consignment mix caps absolute profit, ZT growth stalls and the multiple re-rates back toward the 12–15× EMS base rate — halving the stock. The GAAP earnings are half the quoted non-GAAP figure; insiders and the founder-CEO’s age add overhangs; and the whole re-rating happened in twelve months.

The 3–5 assumptions that matter most: (1) Does ZT reach and hold $6–7B — and diversify beyond AMD? (2) Is the AI-server capex cycle mid-game or late-game? (3) Can vertical integration actually lift ZT/consolidated margin above ~6–7%? (4) Does the multiple hold at a premium to legacy EMS, or mean-revert? (5) Is AMD’s AI-accelerator franchise durable against NVIDIA?

Factor-positioning read (factor-model analysis). SANM screens as a high-beta (β≈1.5), high-momentum name with a spectacular but late-stage risk-adjusted record: 1-year total return +117%, the trailing quarter (Apr–Jun 2026) roughly +69% (a ~+711% annualized m3 figure), but with a lifetime max drawdown of −96% and 5-year volatility ~45% — the profile of a violent, cyclical, sentiment-driven stock, not a low-vol compounder. FactorsToday’s ElasticNet model loads SANM on Market and DividendYield/Value in its base model with modest R² (~0.29–0.34), and its nearest factor-neighbors are the EMS peers (JBL, PLXS, FLEX) and momentum ETFs (PRN, XMMO) — statistical confirmation that the market is now trading SANM as a momentum vehicle. The −22% drop off the June high on a sector scare is the first crack in that momentum. This is evidence the consensus is offsides long the momentum, not that the fundamentals have broken — an input to the “priced for perfection, first air-pocket” framing, not a price call.

12. Fact vs. Interpretation Table

13. Open Questions

1. What is ZT’s absolute gross profit dollar contribution (not revenue), given the consignment mix? The revenue is disclosed; the profit is not cleanly separable.
2. How concentrated is ZT within AMD — and how quickly can a second/third hyperscaler or NVIDIA platform be won at scale? Management targets 12–18 months; no named non-AMD wins yet disclosed.
3. What is the real, normalized run-rate FCF once the ZT working-capital ramp stabilizes and customer advances stop growing?
4. What is the founder-CEO succession plan? Jure Sola has led since 1980; no public successor.
5. How much of the “$16B-plus FY27” is consignment vs. turnkey — i.e., how much profit does it actually carry?
6. Where does the AI-server capital cycle stand — how much of the current EMS/ODM order strength is genuine end-demand vs. double-ordering/pull-forward (the Q2 pull-forward is a live example)?

14. What Must Be True (Bull and Bear, each with a falsification test)

Bull case — what must be true:

• ZT reaches $6–7B in FY27 and diversifies beyond AMD (NVIDIA and/or ≥2 hyperscalers at scale).
• Vertical integration lifts blended operating margin toward 7%+ and holds.
• The AI-server capex cycle has multiple years to run without a digestion year.
• The market continues to award a premium-to-EMS multiple.
• Falsification test: If, by FY2027, ZT is still >85% AMD, or consolidated operating margin is not above ~6.5%, or a hyperscaler in-sources a flagship program — the bull thesis is broken.

Bear case — what must be true:

• The AI-server assembly layer mean-reverts (capacity floods in; margins/multiples compress) or AMD’s AI-GPU share stalls.
• ZT growth proves lumpy and consignment-capped, so profit lags revenue.
• The multiple re-rates toward the 12–15× EMS base rate.
• Falsification test: If ZT diversifies to multiple non-AMD hyperscalers, consolidated margin expands past 7%, and FY27 revenue clears $16B with visible profit growth — the bear thesis is broken and the premium multiple is earned.

15. Source Appendix

See SANM_source_appendix.md (Appendix B in the combined report).

APPENDIX A — Standard Diligence Questionnaire

SANM — Standard Diligence Questionnaire Appendix

Sanmina Corporation (NASDAQ: SANM) · Report date 2026-07-03 Supplemental to the memo. Fact / Interpretation / Assumption labeled where it matters.

General

What thoughtful questions have other investors asked about this company? The recurring institutional debates: (1) Is the ZT/AMD business durable and diversifiable, or a single-customer bet on AMD’s AI-GPU share war with NVIDIA? (2) How lumpy is ZT revenue — and how should we read the ~33% pro-forma YoY decline and the fact that ZT was bought off a ~$14B-annualized 2024 peak? (3) How much profit (not revenue) does the consignment-heavy ZT book actually carry, given goodwill was only ~$276M of the ~$1.9B PPA? (4) Is the ~46× trailing GAAP multiple (98th percentile of its own history) sustainable for a ~6%-margin EMS? (5) Why is non-GAAP EPS ~2× GAAP, and what is the clean run-rate? (6) What is the founder-CEO succession plan? These map to the memo’s the relevant section, the relevant section, the relevant section, the relevant section, and the relevant section.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Interpretation: At a cyclical high — reported revenue, margins, and (especially) the multiple are all elevated by the AI-capex up-cycle and the ZT consolidation. The AI-server assembly layer is late-boom in capital-cycle terms.

Driven by the external environment or internal actions? Both: internal (the ZT acquisition, disciplined buybacks, margin creep) and external (the hyperscaler AI-capex wave, AMD’s accelerator ramp). The durable part is internal execution; the cyclical part — which the valuation extrapolates — is external and outside Sanmina’s control.

How stable are revenues? Low stability. Core Sanmina is cyclical (revenue $8.9B→$7.6B→$8.1B across FY23–25); ZT is extremely volatile (pro-forma −33% YoY; guided down 40%+ sequentially into Q3 FY26). Program-based, re-competed at each generation — not annuity revenue.

Outlook for products/services? How big will this market be? The AI-infrastructure end market is large and growing (hyperscaler capex ~$600–725B in 2026), but capex growth is decelerating and the assembly layer captures the thinnest slice. Core end markets (comms, industrial/energy, medical, defense, auto) are stable-to-slow. International and domestic; US footprint is a reshoring tailwind.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — every EMS/ODM (Celestica, Jabil, Flex, Foxconn, Quanta, Wistron, Supermicro) is adding AI-server capacity into the same demand.

How profitable is the business (ROIC, ROE)? Modest. GAAP-basis ROIC low-teens (ROIC.ai 9–15%); management’s non-GAAP pretax ROIC of 34.7% (Q2) is real but rests on a working-capital-financed denominator and flattering numerator. Operating margin ~5–6%. This is the EMS base rate, not a franchise return.

How profitable is the industry — competitors, barriers to entry? Structurally low-margin (3–6% operating), fragmented, with negligible barriers to entry. Competition on price/scale/footprint. No durable moat (Greenwald): no supply/cost advantage, only shallow generation-resetting captivity, no scale+captivity combination.

Can the business be easily understood? Yes — it is a contract manufacturer. The complications are the ZT purchase accounting, the GAAP/non-GAAP wedge, the consignment revenue model, and the India JV minority interest.

Can it be undermined by foreign low-cost labor? Yes — this is the core structural risk of Western EMS; Asian giants (Foxconn ~26× Sanmina’s revenue) have a cost edge. Sanmina’s counter is its US/diversified footprint (a reshoring tailwind), higher-mix regulated franchises, and vertical integration.

Do brands matter? Nature of competition? Switching costs? Brands do not matter to end demand; Sanmina’s “brand” is reliability/qualification with OEMs. Competition is on price, footprint, engineering, and qualification. Switching costs are real mid-program but reset each generation — retention, not rent.

Financial Condition & Balance Sheet

Assets not fully recognized / off-balance-sheet liabilities? The $850M PNC receivables-purchase (factoring) facility at ZT is the key off-balance-sheet item — it flatters working capital and cash conversion. The $450M contingent earnout (booked at ~$111M fair value, re-measured +$59M in H1 FY26) is a real, growing liability. Operating leases are standard.

How conservative is the accounting? Mixed. GAAP is arguably conservative on EPS (non-GAAP is ~2× GAAP, the ordinary direction — GAAP absorbs SBC, ZT step-up amortization, and acquisition costs), but GAAP net income is flattered by large non-operating income (~$75.5M in Q2), and the ZT PPA is preliminary (measurement period open). The consignment model and factoring make revenue and cash quality harder to read. Not aggressive, but not clean.

How CapEx-hungry is the business? Light on fixed capex (~1.5–2% of sales, ~$110–150M/yr) — an attractive feature — but heavy on working capital (inventory $3.0B; the ZT ramp consumes cash), which is the real capital intensity of EMS.

Capital Allocation & Management

How much FCF does the business generate, and how is it used? FY2025 FCF ~$473M (lumpy: $44M→$229M→$473M across FY23–25), used almost entirely for buybacks (share count 69M→55M over five years) and, in FY26, the ZT acquisition. No dividend. Philosophy (management-stated priorities): organic investment → strategic M&A → balance sheet/investment-grade goal → opportunistic buybacks.

Significant acquisitions recently? Yes — the transformational ZT Systems deal (~$3B, closed Oct 2025), Sanmina’s largest ever, funded with $2.2B of new secured debt.

Buying back shares? Issuing to insiders? Both — heavy buybacks (new $600M authorization Apr 2026, deployed at record multiples = a flag); modest SBC (~$63M/yr) more than offset by buybacks. 1.15M shares issued as ZT consideration.

Compensation / motivations of management? CEO Sola ~$18.2M FY25 (~93% performance-based); incentives on revenue + non-GAAP operating margin + CFO modifier — no return-on-capital, leverage, or TSR metric, a mis-alignment now that the balance sheet carries $2.2B of secured debt. Insiders own ~3.1% and are net sellers into the run (CEO sold ~$27M May 2026). Founder-CEO key-person/succession risk is unaddressed in filings.

Valuation & Market Data

ADR, MLP, or K-1 issuer? No — US domestic C-corp, common stock, NASDAQ.

Dividend policy? None.

How profitable is the business? Thin (net margin ~3% GAAP); see above.

Is net income diverging from cash from operations? Periodically yes — FY2025 OCF ($620M) far exceeded net income ($246M) on a working-capital inflow (customer advances); conversely, working-capital build for the ZT ramp will pressure future OCF. CFO/NI is volatile — read multi-year, not single-quarter.

Risks & Downside

What factors would cause the stock to decline? A multiple de-rating toward the EMS base rate; an AI-capex digestion year; AMD’s AI-GPU share stalling; a hyperscaler in-sourcing; ZT revenue proving lumpy/consignment-capped; component shortages; disappointment vs the $16B FY27 target.

Risk of a catastrophic / total loss? Low. Strong (if now-levered) balance sheet, positive tangible equity, real cash generation, capital-light fixed assets, diversified legacy franchise. The risk is valuation and concentration, not solvency — a large drawdown is plausible; a permanent impairment of capital is not.

Recent News & Events

Has the business environment changed recently? Dramatically — the ZT/AMD acquisition (Oct 2025) doubled the company and re-rated the stock 5×; the Q2 FY26 blowout (Apr 2026) took it to an all-time high $283; a broad EMS-sector selloff (Jul 2, 2026) pulled it to ~$220 (−22% off high).

Significant acquisitions? ZT Systems (above).

Change in accounting policies? No policy change, but ZT purchase accounting (preliminary), the receivables-factoring facility, and the growing non-GAAP wedge materially change the reported optics.

Recent changes — new markets, facilities, management? New rack-scale AI/data-center capacity (ZT plants; incremental power, liquid-cooling, test-cell investment); Ireland site expansion (~150 jobs); no management change (founder Sola remains Chairman/CEO — a monitorable given his tenure/age).

APPENDIX B — Source Appendix

SANM — Source Appendix

Sanmina Corporation (NASDAQ: SANM) · Report date 2026-07-03 Primary sources prioritized. Third-party aggregated data (ROIC.ai, AZI, FactorsToday) reconciled to filings where material. All access dates 2026-07-03 unless noted.

Primary — SEC filings & company disclosures

• Sanmina Corporation Form 10-K filings, FY2020–FY2025 (fiscal years ending late Sept/early Oct). SEC EDGAR CIK 0000897723. https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000897723
• Sanmina Form 10-Q filings, FY2024–FY2026 (incl. Q1 FY26 ended 2025-12-27 and Q2 FY26 ended 2026-03-28).
• Sanmina Q2 FY2026 earnings press release and slides (quarter ended 2026-03-28), filed via 8-K. https://www.sec.gov/Archives/edgar/data/0000897723/000089772326000024/sanmina_exx991xmarch282026.htm
• Sanmina Q2 FY2026 earnings conference call transcript, 2026-04-27 (via ROIC.ai). Management: Jure Sola (Chairman & CEO), Jon Faust (EVP & CFO), Paige Bombino (SVP IR).
• Sanmina 8-K filings (last ~24 months): ZT Systems acquisition announcement (May 2025) and closing (Oct 27, 2025); $600M buyback authorization (Apr 2026); debt facilities ($600M delayed-draw + $800M incremental term loan); quarterly earnings 8-Ks.
• Sanmina DEF 14A proxy statements (executive compensation, incentive metrics, ownership).
• Sanmina Form 3/4/5 insider-transaction filings (trailing ~24 months).
• Sanmina corporate/IR site: https://www.sanmina.com (Investor Relations).

Primary — counterparty / transaction

• AMD press release, “AMD Announces Agreement to Divest ZT Systems’ Data Center Infrastructure Manufacturing Business to Sanmina,” 2025 (deal terms: ~$3B total, $2.55B cash+equity + up to $450M contingent; AMD retains ZT design business). https://ir.amd.com/news-events/press-releases/detail/1252/
• Manufacturing Dive, “AMD to sell ZT Systems’ manufacturing business to Sanmina,” 2025. https://www.manufacturingdive.com/news/amd-sanmina-zt-systems-acquisition-deal-ai-data-center-infrastructure/748587/

Quantitative data helpers (third-party; reconciled to filings)

• ROIC.ai MCP — income statement, balance sheet, cash flow, profitability/credit/liquidity/working-capital ratios, enterprise value, valuation multiples, company profile, earnings-call transcripts (accessed 2026-07-03). Third-party aggregated; EDGAR/10-K remain primary. NB: ROIC.ai reports a negative price-to-book for SANM — an accumulated-deficit artifact; real book equity is positive (~$45/share). ROIC “34.7%” figure is management’s non-GAAP pretax ROIC, not a comparable GAAP ROIC (GAAP-basis ROIC is low-teens).
• Market data feed — valuation_index own-history percentile ranks (P/E 46.5x = 98.2 pctile; P/B 5.02x = 98.3; P/S 1.07x = 97.2; composite 97.9), price CSV (adjusted/unadjusted OHLCV, EMAs, beta/alpha), and news feed (accessed 2026-07-02/03). https://azitrading.com
• FactorsToday factor model — stock-loadings, leaderboard (risk-adjusted returns/drawdowns by horizon), stock-info (beta 1.51, RS), related-stocks (nearest neighbors: JBL, PLXS, FLEX + momentum ETFs PRN/XMMO). https://www.factorstoday.com/api (accessed 2026-07-03)

Industry, peer & market context

• Peer context drawn from public filings and market data for Celestica (CLS, 2026-06-05), Jabil (JBL, 2026-06-19), Flex (FLEX, 2026-06-12), Fabrinet (FN, 2026-06-14), TTM Technologies (TTMI, 2026-06-19) — EMS industry structure, capital-cycle framing, peer comps.
• Silicon Analysts, “AMD vs NVIDIA AI GPU Market Share 2026.” https://siliconanalysts.com/analysis/amd-vs-nvidia-ai-gpu-market-share-2026
• Digitimes, “NVIDIA AI server ODM / Foxconn production,” 2025-11-13. https://www.digitimes.com/news/a20251113PD201/
• Tech-Insider, “Foxconn Q1 2026 AI server revenue,” 2026. https://tech-insider.org/foxconn-q1-2026-ai-server-revenue-66-billion/
• TIKR, “Sanmina stock jumps after Q2 2026 earnings beat — is the $16B revenue target within reach,” 2026. https://www.tikr.com/blog/
• Simply Wall St, SANM valuation/analyst coverage (JPMorgan Neutral $145; Susquehanna Neutral $135), 2026-06. https://simplywall.st/stocks/us/tech/nasdaq-sanm/sanmina
• Sahm Capital, “Sanmina valuation check after strong returns and ZT Systems acquisition,” 2026-06-07. https://www.sahmcapital.com/news/content/
• Seeking Alpha, “Sanmina’s ZT Systems gamble is paying off faster than anyone expected,” 2026. https://seekingalpha.com/article/4914157
• Benzinga (via AZI feed), EMS-sector selloff note, 2026-07-02. https://www.benzinga.com/wiim/26/07/60258883/

Analytical frameworks

• Investment-research-frameworks skill (Greenwald & Kahn, Competition Demystified; Marathon/Chancellor, Capital Returns) — moat taxonomy, share-stability/ROIC tests, capital-cycle placement.

All sources above are public. This article is independent research and reflects no position in the securities discussed.