A picks-and-shovels AI-infrastructure compounder hiding inside an EMS multiple — the bet is that liquid-cooling/power vertical integration + relentless buyback keep core EPS compounding 20%+, but at ~29x forward core EPS the re-rate is mostly done and a single 16% hyperscaler customer + AI-capex cyclicality are the trapdoors.
No Friday close is on the record for JBL yet. The weekly job prices the covered universe; a name it cannot price is listed as missing rather than estimated.
| Date | Type | What happened | Source |
|---|---|---|---|
| 2026-07-10 | editorial note | Valuation figure revised: $365 → $332.51Valuation moved from $365 (deep-dive-2026-06-29.md) to $332.51 (deep-dive-2026-07-10.md). | dossier |
| 2026-07-10 | editorial note |
| Margin figure revised: 8.9% → 6.1%Margin moved from 8.9% (deep-dive-2026-06-29.md) to 6.1% (deep-dive-2026-07-10.md). |
| dossier |
| 2026-07-10 | editorial note | Revenue figure revised: $29.8B → 48%Revenue moved from $29.8B (deep-dive-2026-06-29.md) to 48% (deep-dive-2026-07-10.md). | dossier |
| 2026-07-10 | editorial note | Verdict changed: A picks-and-shovels AI-infrastructure compounder hiding inside an EMS multiple — the bet is that liquid-cooling/power vertical integration + relentless buyback keep core EPS compound…Before (deep-dive-2026-06-29.md): A picks-and-shovels AI-infrastructure compounder hiding inside an EMS multiple — the bet is that liquid-cooling/power vertical integration + relentless buyback keep core EPS compounding 20%+, but at ~29x forward core EPS the re-rate is mostly done and a single 16% hyperscaler customer + AI-capex cyclicality are the trapdoors. After (deep-dive-2026-07-10.md): "The Q3 10-Q upgrades the thesis from web-narrative to hard fact: the AI segment is now Jabil's HIGHEST-margin business (6.1% seg income) and ~half of revenue — but the AI growth just rotated from cloud/DC (decelerating) to networking, the whole $8.8B AI franchise is financed by a 135-day payables stretch, and an Amazon warrant all but names the 16% customer. Better business than the market's ~26x forward core credits; more fragile balance sheet than the headline CFO shows." | dossier |
The verdict
"The Q3 10-Q upgrades the thesis from web-narrative to hard fact: the AI segment is now Jabil's HIGHEST-margin business (6.1% seg income) and ~half of revenue — but the AI growth just rotated from cloud/DC (decelerating) to networking, the whole $8.8B AI franchise is financed by a 135-day payables stretch, and an Amazon warrant all but names the 16% customer. Better business than the market's ~26x forward core credits; more fragile balance sheet than the headline CFO shows."
Primary sources
SEC filings
Source documents — open to read in full
What changed:
. Core EPS $3.16 remains (the EPS reconciliation isn't in the Form 10-QThe quarterly version of the annual report. Lighter, and not audited.), but the $504M core OI is now filing-confirmed., down from ~$365 (June 26). Now **~26.2x forward core EPS** ($12.70) vs ~28.7x in the prior dossier — the modest de-rate the prior seed wanted has partly arrived.What held: the structural thesis is intact and, if anything, better-evidenced — an AI-data-center infrastructure compounder inside an EMS multiple, vertically integrating into the cooling/power bottleneck, shrinking the float ~2.5%/9mo. The two trapdoors are unchanged: a single ~16% (Amazon) customer and AI-capex cyclicality. Regulatory/legal remains clean (re-fetched 2026-07-10: zero SEC LR/AAER; Note 17 ordinary-course only).
Model unchanged from the prior dossier (turnkey EMS: materials pass-through + a thin labor/overhead margin; run through dedicated customer business units), with two sourced updates:
(see delta #4). Customer concentration is unchanged in shape — top-5 ≈ **36%** of 9-mo revenue, 78 customers ≈ 90% — but we can now put a likely name on the dominant relationship, which sharpens the concentration risk (Jabil's AI growth is levered to one hyperscaler's capex).Everything else in Lens 1 carries from the previous dossier.
Unchanged — see the previous dossier, Lens 2. Upstream (Nvidia/AMD/Broadcom silicon Jabil integrates; single-source components as the flagged chokepoint; in-housed cooling via Mikros, power via Hanley) and downstream (hyperscalers led by the ~16% Amazon relationship; capital-equipment/networking OEMs; regulated OEMs; the shrinking consumer tail) are structurally the same. One reinforcing data point: Jabil now carries $2.8B of customer-directed components in prepaid (vs $1.1B at FY25) `` — physical evidence of how much accelerator/networking inventory it is warehousing on hyperscalers' behalf (revenue booked net on these).
The prior dossier argued Jabil is buying its way from no-moat commodity assembler toward a narrow-moat supplier of supply-constrained AI subsystems, but conceded the moat was "young" and unproven in the numbers. This refresh provides the first hard evidence it is working:
Moat verdict: upgraded from "asserted" to "early-validated," still narrow. Ground: positioning.md / bottlenecks.md (robotics wiki) frame the AI bottleneck as cooling/power-density — precisely where the 6.1% II margin is now printing.
The prior dossier's biggest gap — per-segment operating-income dollars "n/a" — is closed. Full Note-11 table, every figure ``:
Three months ended May 31 (Q3-FY26 vs Q3-FY25):
| Segment | Q3-FY26 rev | Q3-FY26 seg income | Q3-FY26 margin | Q3-FY25 rev | Q3-FY25 margin | Rev YoY |
|---|---|---|---|---|---|---|
| Regulated Industries | $3,181M | $180M | 5.6% | $3,056M | 5.5% | +4.1% |
| Intelligent Infrastructure | $4,169M | $256M | 6.1% | $3,433M | 5.3% | +21.4% |
| Connected Living & DC | $1,401M | $68M | 4.9% | $1,339M | 5.3% | +4.6% |
| Total | $8,751M | $504M | 5.8% | $7,828M | 5.4% | +11.8% |
Nine months ended May 31 (9M-FY26 vs 9M-FY25):
| Segment | 9M-FY26 rev | 9M-FY26 seg income | 9M-FY26 margin | 9M-FY25 rev | 9M-FY25 margin | Rev YoY |
|---|---|---|---|---|---|---|
| Regulated Industries | $9,280M | $502M | 5.4% | $8,754M | 5.0% | +6.0% |
| Intelligent Infrastructure | $12,050M | $689M | 5.7% | $8,576M | 5.1% | +40.5% |
| Connected Living & DC | $4,008M | $203M | 5.1% | $4,220M | 5.2% | −5.0% |
| Total | $25,338M | $1,394M | 5.5% | $21,550M | 5.1% | +17.6% |
What the newly-visible data says:
Blended gross margin improved to 9.46% (Q3) / 9.14% (9mo) from 8.70% / 8.65% YoY `` — mix + leverage lifting even the pass-through-diluted GM line.
Q3-FY2026, quarter ended 2026-05-31 — all `` unless noted:
Nine-month FY26 anchor : revenue **$25,338M** (+17.6%), GAAP diluted EPS **$6.01** (vs $3.94, **+52.5%**), core operating income **$1,394M** (core OM 5.5%), net income $644M. Implied 9-mo core EPS ≈ **$8.71** (1H $5.55 + Q3 $3.16) , leaving Q4 core ≈ $4.0 to hit the $12.70 FY guide ``.
Guidance (unchanged since Q3 print, still ``): FY26 revenue ~$35B, core OM ~5.8%, core diluted EPS ~$12.70, AI revenue ~$13.6B (walked up from $9.0B FY25 → $13.1B → $13.6B); FY27 core OM guided >6%. This was the third consecutive FY26 raise; no new guidance since (next update = Q4 print, late Sept 2026).
Balance-sheet flags `` — materially worse-looking than a year ago, and this is the key watch-item:
Market reaction: JBL $332.51 (2026-07-09) ``, ~9% below the prior dossier's ~$365 (June 26) and ~22% off the 52-wk high ($428.93). The beat-and-raise is three weeks old; the stock has drifted lower into a market that already prices continued execution.
transcripts/ remains empty — sentiment is still web/print-reconstructed, and this is now the single largest remaining gap on the shelf (flagged again below). No new call has occurred since the prior dossier (Q3 was 2026-06-17; Q4 lands late Sept). The arc is unchanged from the previous dossier: progressively more confident on AI, serial raises, lexicon fully migrated to liquid cooling / power / AI factories / "Intelligent Infrastructure," with "consumer/Mobility" retired. One nuance the Q3 filing adds to watch on the next call: management will have to explain the cloud/DC sequential deceleration (contribution +31pts 9mo → +8pts Q3) against the still-raised $13.6B AI number — whether that reads as "networking taking the baton" (bull) or "compute-rack digestion" (bear) is the most important tonal read of the Q4 call. Sentiment verdict: bullish and tightening, now with a specific stress-test looming.
EMS / AI-EMS peers. JBL data +; peer multiples `` with source/date or n/a:
| Company | Ticker | Mkt cap (USD) | Fwd / P/E | As-of | Notes |
|---|---|---|---|---|---|
| Jabil | JBL | ~$34.8B `` | ~26.2x fwd core (=$332.51 / $12.70 ); 41.6x trailing GAAP | 2026-07-09 | The subject; de-rated ~2.5 turns vs prior dossier |
| Celestica | CLS | n/a this pass | ~33.7x fwd | 2026-07-03 `` | Still the richest, "purest" AI-EMS comp |
| Flex | FLEX | ~$48.9B `` | ~28.8x fwd `` | 2026-07-08 | Larger, similar AI tilt |
| Sanmina | SANM | ~$11.2B `` | 46.1x trailing (fwd n/a) | 2026-07-08 `` | Smaller; IMS +72% YoY per Q1 print `` |
| Hon Hai (Foxconn) | 2317.TW | n/a | n/a | — | Largest by far; the scale threat |
| Plexus | PLXS | n/a | n/a | — | Smaller, regulated-heavy |
Read (updated): the whole EMS group remains re-rated into the high-20s/30s on the AI narrative, and JBL got ~2.5 turns cheaper on the pullback — now ~26x forward core, the cheapest large-cap AI-EMS name (below Flex ~29x and Celestica ~34x; Sanmina's 46x is trailing, not comparable). The GARP-slot framing strengthens: JBL is compounding core EPS ~20%+ (FY26) at the lowest forward multiple in the large-cap AI-EMS set. The sector-multiple risk is unchanged — if the AI-capex narrative cools, all of these de-rate toward the teens together. Clean cross-peer 5-yr ROE remains n/a (buyback-distorted equity base).
New since the prior dossier ``:
; consensus PT ~$453 . The gap between ~$332 spot and ~$453 consensus is now ~36% — either the sell side is late or the pullback is an entry. (Note: some retail-forecast feeds show garbled ~$251/$265 figures — disregarded; verified spot is $332.51 ``.)Pattern (unchanged, reinforced): JBL is a high-beta AI-capex proxy — it will trade with the hyperscaler-capex narrative (and now, specifically, with anything that moves Amazon's capex expectations) far more than on its diversified-industrial roots.
Leadership unchanged: CEO Mike Dastoor (finance-trained operator, serial beat-and-raise), CFO Greg Hebard; Mondello + two directors exit the board at the Jan 2026 annual meeting (clean succession completion). Capital-allocation update, now ``:
Management verdict (unchanged): credible, disciplined capital allocator; the watch-item shifts slightly — from "buyback flatters EPS" toward "buyback is now buying at ~2x the old price while net debt doubles and working capital is payables-financed." Still a compounding steward; the balance-sheet stretch deserves the next question.
Acting as forensic analyst, `` unless noted:
Regulatory findings ``:
Anchor = FY26 guidance (core EPS ~$12.70, revenue ~$35B, core OM ~5.8%), diluted shares ~104.8M and falling. 9-mo core EPS ~$8.71 → Q4 core ~$4.0 to hit the guide. Base/bull/bear core-EPS paths **FY2026 → FY2028** (FYE August); output, inputs labeled.
**FY2026 (nearly complete, 3 of 4 quarters ):** **core EPS ≈ $12.70** — treat as near-actual.
Base case:
Bull case: networking leg accelerates, cloud/DC re-accelerates, II margin → ~6.5%+, $1.5B/yr buyback → FY27 ~$16.3, FY28 ~$19.5 ``.
Bear case: hyperscaler (Amazon) capex digestion in FY27 — II decelerates toward mid-single-digits, the payables/consignment working-capital tailwind reverses (CFO < earnings), new-capacity fixed costs drag margin to ~5.5% → FY27 ~$12.5–13 (flat), FY28 ~$13 ``, with simultaneous multiple compression (double hit).
Valuation frame at $332.51: ~26.2x FY26 core, ~22x base FY27, ~19x base FY28 ``. The ~9% pullback moved JBL from "priced for continued execution" toward "reasonable for a ~17% core grower" — a re-rate to mid-20s on the base path implies upside to ~$375–400; the bear digestion path still points to the mid-teens multiple / ~$200s. The risk/reward is better than at $365, not yet cheap.
(Per --watchlist rule + wave boundary, no our model create logged. If committed at a future our position log pass: "JBL FY27 (Aug-2027) non-GAAP core EPS ≥ $15.0," p≈0.55 — unchanged base, marginally higher confidence given the sourced II margin.)
Bull case (strengthened on two fronts). Jabil is the GARP way to own the AI-data-center buildout without silicon-layer cycle risk — and the Q3 filing hardens it: the AI segment is now the highest-margin business (6.1% Q3 seg income, +80bps YoY), proving the Mikros/Hanley cooling-and-power vertical-integration is margin-accretive, not dilutive; the AI growth has a visible second leg (networking inflected to the #1 driver) just as the first (cloud/DC) matures; and the stock de-rated ~9% to ~26x forward core, the cheapest large-cap AI-EMS multiple, while core EPS compounds ~17–20%. Add a de-risked tariff picture (SCOTUS IEEPA strike-down + refunds) and a narrowing GAAP-to-core gap (restructuring done). Contrarian read: the market still prices JBL like a low-margin commodity EMS (~26x, below peers) even as its highest-growth segment now earns its best margin — the quality re-rate is incomplete.
Bear case (permanent-impairment risks, updated). (1) Concentration now has a face: the ~16% customer is almost certainly Amazon (warrant), so half the growth segment rides one hyperscaler's capex and Jabil has no pricing power over it. (2) Cloud/DC just decelerated: the compute-rack contribution fell from +31pts (9mo) to +8pts (Q3) — the exact "digestion" signpost, papered over by a still-raised $13.6B AI headline; if networking can't keep carrying the baton, the growth narrative cracks. (3) The balance sheet is financing the AI ramp: AP +$4.0B, receivables sold +73%, net debt doubled to ~$2.0B, CFO timing-flattered — a normalization of the 135-day payables stretch would drop CFO below earnings and expose how capital-hungry the AI franchise is (II assets doubled to $8.8B for $689M of 9-mo income). Pre-mortem (18 months out): Amazon dual-sources or trims capex growth in FY27, cloud/DC goes flat, networking can't offset, the payables tailwind reverses (CFO craters), and JBL de-rates from ~26x to mid-teens on a flat EPS — a ~40% drawdown requiring nothing to break at Jabil, only at its customer.
Is the multiple too high? At ~26x forward core (down from ~29x), still elevated vs JBL's own pre-2024 low-teens history, but cheap vs AI-EMS peers and now supported by a demonstrated highest-margin AI segment. The bet remains whether the AI-infrastructure secular call — and Amazon's capex — holds long enough for EPS to grow into it.
Dismantling the bull case with the new data:
The 15 questions from the previous dossier, Lens 14 carry forward (customer concentration, Mikros/Hanley content mix, digestion signposts, cash conversion, margin bridge, buyback-vs-price, II ceiling, Foxconn threat, networking, Regulated profile, incremental/decremental margins, tariffs, value-chain line, succession depth, "protect one thing"). Two are re-pointed by this refresh and jump to the top of the queue:
Every dossier we have written on Jabil, newest first.
The Q3 10-Q upgrades the thesis from web-narrative to hard fact: the AI segment is now Jabil's HIGHEST-margin business (6.1% seg income) and ~half of…
A picks-and-shovels AI-infrastructure compounder hiding inside an EMS multiple
| Industry | Robotics |
| Size | Public Company |
Where Jabil sits against the other names we cover on this beat. Each line is that company’s verdict, not a summary of it.
NOT ASSESSED — this refresh is descriptive.
Cash $44.7B
The cost-curve and volume king of humanoids (≈32% of 2025 unit shipments, profitable, BYD/DJI-style vertical integration)
The EKSO equity no longer exists as a medical-robotics bet
The best-run Chinese industrial-robot franchise, wearing a balance sheet (net-debt/EBITDA ~15x, receivables > cash) and a valuation (~100x TTM P/E) th…
The only eVTOL pure-play that will die of missed schedules, not empty coffers