Phase Space AI

Trade Construction

Celestica [CLS]

Celestica [CLS] — Trade Construction

Spot $328.43 (2026-07-29). The memo issues no position verdict — this section specifies how the trade would be constructed if the book allocates to it.

1. Vehicle: EQUITY — decided on a measurement, not by default

Required disclosure Value
Trailing 252-day realised volatility 76.31%
Measured ATM implied volatility, 387 days 83.56% (CLS270820C00330000, strike $330, expiry 2027-08-20)
IV − RV +7.26pp — UNFAVOURABLE
Quoted size at the intended strike bid $110.76 × 64 / ask $119.30 × 72
Bid-ask as % of premium 7.4% ($8.54 on a $115.03 mid)
Delta 0.685
Chain existence PASSES — 14 contracts between $300 and $360 across the 2027-08-20 and 2027-11-19 expiries

CLS passes the existence gate that FN failed — FN had no contracts within ±10% of spot at any expiration 350 days or more out, so no IV could be measured at all. Here the chain exists and the test moves to economics, where it fails: implied volatility sits 7.26 points above realised on a name whose realised volatility is already 76%. A long LEAP pays that premium and then a further 7.4% round trip in spread.

Surface confirmed across three strikes so the reading is not a single stale quote: $330 IV 83.56%, $340 83.18%, $360 84.00%. Open interest at the near-ATM strikes is thin (7–122 contracts, 357 at $360), so size beyond a few hundred contracts would move the market.

Per the vehicle ladder, absent a favourable IV − RV there is no argument for anything other than equity. Vehicle: EQUITY, ladder rung 1. Liquidity is not a constraint: $38.2bn market capitalisation, dual-listed NYSE and TSX.

2. Entry
Reference price $328.43
Position in own multiple range 25.4th percentile of the trailing-twelve-month EV/gross-profit history (median 24.99x, current 22.93x); 52.2nd percentile since 2025-01-01
Distance from recent high −10.0% from the 2026-06-30 close of $364.80
Post-earnings status Q2 2026 reported 2026-07-27, two days before the reference date — revenue beat the high end of guidance, adjusted EPS beat the high end

Entry is not being timed against a technical level. The relevant observation is that the multiple sits in the bottom quartile of its own post-regime range while consensus revisions are up on every horizon with zero downgrades in 30 days. That is the configuration the momentum criterion scores, and it governs timing only.

Scale in against the TRS reversal, not before it. On current marks Q3 2026 books a ~$44.8m TRS loss against a $104.1m gain in Q2 — a ~$149m swing, roughly 3.1pp of a single quarter's reported operating margin. Reported gross and operating margin will fall in Q3 for reasons that have nothing to do with the business. Any investor who has not made the adjustment in CLS_Valuation.md §2 will read that as deterioration. A starter position now and the balance after the Q3 print in late October is the construction that is paid for the analytical work rather than surprised by it.

3. Sizing constraints

Sizing is the book's decision. The constraints this memo hands it:

  1. Correlation, and it is severe. CLS is a levered proxy on the same hyperscaler AI capex that drives NVDA, TSM, AVGO, MRVL, ALAB, CRDO, FN, COHR, LITE, ANET and APH — all in or queued for this corpus. CLS is further down the value chain than any of them, at 11.2% gross margin, so it has the least pricing power and the highest operating leverage to the same single variable. Treat it as an amplifier of existing AI-hardware exposure, not a diversifier.
  2. Single-customer risk is existential and unnameable. The largest customer is 34% of 1H 2026 revenue and is not named. The top three are 63–65%. There is no backlog, no disclosed contract term and no volume commitment anywhere in the filings. Position size should reflect that the downside case is a single-counterparty decision, not a cycle.
  3. 76% realised volatility. Whatever the book's volatility-scaled sizing rule is, this name sits at the high end of it.
  4. Balance-sheet fragility is low but not nil. Net debt $274.7m — 0.7% of EV — against $1,146.5m of TTM operating cash flow. But that operating cash flow contains a +$2,096.7m payables swing. On a payables-neutral basis 1H 2026 operating cash flow is deeply negative while capex runs at $493.3m. A working-capital reversal would require the revolver.

4. Exit and invalidation

Target

$495 (+50.7%) — 22.00x clean EV/gross profit, held flat at today's level, on NTM clean gross profit of $2,627m. Full build in CLS_Valuation.md §7. The target implies 32.8x FY2027E adjusted EPS of $15.08 against 21.8x at spot, and this is disclosed as the aggressive reading: the entire target is a gross-profit-growth statement at a flat multiple.

Downside

$235.76 (−28.2%), probability 0.30. Named cause: the AI/ML compute program is lost or dual-sourced at the 34%-of-revenue hyperscaler. Full build in CLS_Research.md §8.

Invalidation — thesis is wrong, exit regardless of price

  1. Enterprise end-market revenue declines sequentially for two consecutive quarters while Communications grows. This is the direct confirming metric of the downside case: it is the AI/ML compute program rolling off. Enterprise was $1,157.1m in Q2 2026 against $830.4m in Q1 2026 and $433.1m in Q2 2025.
  2. The number of ≥10% customers falls from three to two, or the largest customer's percentage falls more than 5pp year on year in an annual 10-K. Concentration going down would normally be good news; here it means a program was lost, because nothing else in the business is growing fast enough to dilute it.
  3. Clean gross margin falls below 10.8% — the prior-TTM level — for two consecutive quarters. That would say the AI programs are being won on price, and it removes the 0.30pp of terminal gross-margin expansion the model underwrites.
  4. DPO reverses below 70 days. The 84.3-day DPO released roughly 27 days of working capital in a year. A reversal consumes cash at the same rate, into a capex ramp, and would force the revolver.

Not invalidations — expected, and pre-committed here so they cannot be re-read as news

5. What the position is actually expressing

That the rack-integration layer of the AI buildout is being paid 22x its own gross profit for growing that gross profit 53% a year, and that the required 5-year revenue CAGR of 30.0% is 17.3 points below what has already been delivered — with a flip point at a 5.98x exit multiple that no comparator in the class trades anywhere near.

Against it: the growth belongs to three unnamed customers at 63% of revenue, with no backlog, no contract term and no volume commitment disclosed anywhere; the reported margins that made the name screen well are 19.1% a swap on its own stock; the working-capital improvement is a supplier stretch; and capital intensity has changed by 7x in two quarters with two quarters of evidence on the return.

The bull case is arithmetic. The bear case is a counterparty. Size accordingly.