GOOGL prints tomorrow. Celestica's Q1 2026 10-Q (filed yesterday) is best read in the 16 hours before that catalyst, because the most important number — adjusted ROIC of 49.8% annualized vs 31.5% Q1 2025 — is what reframes the trade. EMS companies don't print 49.8% ROIC. JBL trails at ≈18%, FLEX at ≈14%. Either Celestica is misclassified at the multiple level (12-15x EMS) and deserves rerating toward ODM/specialty (mid-30s), or the ROIC mean-reverts. The Q1 filing has the data either case needs.

What confirmed (and is now priced)

Enterprise revenue +101% YoY ($830M vs $414M) is the Google TPU Ironwood ramp expressed at the rack-integration layer. Communications +69% YoY ($2.41B) on three concurrent networking generations (400G/800G/1.6T). The supply chain corroborates loudly: AVGO Hock Tan named "seventh-generation Ironwood TPU" on March 4; AVGO's 8-K April 6 confirms a Google long-term agreement covering "future generations" through 2031; TSM CoWoS capacity expanding 4x by year-end; DELL ($50B FY27 AI guide), MRVL, CRDO, ANET all guide H2 acceleration. Five primary sources confirming the same direction = one signal, not five. Loud corroboration compresses edge. The H2 deceleration narrative that drove the -13% Jan 29 selloff is dead; the stock at +381% over 1Y has done most of that math.

What is not yet in models

The IG pathway. April 27 credit amendment expanded the revolver $750M → $1.75B and added an investment-grade collateral fallaway provision. Banks added that language because management has a path. ≈$719M term loans repricing 100-150bps lower at IG = ≈$7-9M annual interest savings, ≈$0.05-0.07 EPS. Not transformative on earnings; transformative on multiple. Five turns of multiple at $9 EPS = $45 of stock.

ROIC step-change. 31.5% → 49.8% in four quarters is regime change, not noise. CCS margin 8.6% (vs 8.0%); ATS 6.0% (vs 5.0%). If sell-side keeps modeling CLS at JBL's 12-15x forward, they're missing 5-7 turns of latent rerating. Whether the right multiple is 30x, 40x, or AVGO's 25-30x is the open argument — but it isn't 12-15x anymore.

Customer 1. This is the trade and the risk simultaneously. 35% of revenue (up from 28% Q1 2025), almost certainly Google. The TPU rack-ODM partnership is the engine of the +101% Enterprise print and the structural ROIC. It is also a single-customer binary. Generation transitions create gaps (Enterprise -19% FY2025 was the prior-gen wind-down). Supplier diversification on Google's side moves the stock 25-30%. Up 700bps of concentration in one year is the single largest forward risk and the single largest forward bull driver. Do not own this without sizing for that.

Counter-evidence engaged

• $3.4M acquisition advisory (5.7x YoY) looks like an active M&A signal. Possibly is. Also possibly recurring overhead at $16B revenue. The HZO hidden-M&A pattern required guidance withdrawal + zero buybacks + credit amendment; CLS has the credit amendment but the other two are missing. Probability-weighted: don't price this.
• Short interest 2.3% is a crowded-long tell. Almost nobody is willing to fight this tape. That cuts both ways — bullish on flows, bearish on what happens if positioning unwinds.
• The pair trade rests on JBL as a clean commodity-EMS short. JBL is not pure commodity — Apple, automotive, healthcare exposure. The discriminator alpha is real but the pair has basis risk; expect 20-30% of the trade to be JBL-specific noise.

Risks ranked

1. Customer 1 supplier diversification or generation transition. -25-30%, dominant single risk.
2. Momentum unwind. RSI 91, IV Rank 103%, +50.7% over 1 month. Sector rotation out of AI crashes the trade before idio plays out.
3. Thailand tax holidays expire 2027-2029 during $1B CapEx payback window. Holiday-rate vs Thailand statutory step-up could be 500-1500bps on segment ETR (not disclosed by company); material, unquantified.
4. AI capex deceleration broadly. DeepSeek-style efficiency gains reducing hyperscaler order volumes is the sector left tail.

Catalysts

What changes the read

Down: GOOGL TPU pause language; CLS Q2 sequential decline; Customer 1 disclosure drops below 30%; cohort-wide hyperscaler capex cuts ≥10%.

Up: Customer 2 or 3 revealed as a different hyperscaler (concentration narrative breaks); IG rating action announced; M&A target identified at HPS-adjacent capability layer; sell-side reclassification of CLS out of EMS coverage.

Cross-chain reads

The CLS print is a confirmation tape for the entire AI infrastructure stack: AVGO Ironwood ramp validated at rack level, TSM CoWoS demand confirmed, DELL/SMCI competitive positioning unchanged (CLS does ODM rack-level, SMCI does merchant-silicon assembly — different jobs). The discriminator that matters across the cohort: who has structural ROIC step-change vs who has revenue growth without margin discipline. CLS is in the first bucket. JBL/FLEX are in the second. The pair structure isolates that gap.

Naked long is ≈35% thesis / ≈65% factor noise; the cleanest expression is long CLS / short JBL, sized fractional.

Evidence