🕸️

CLS

💡
Thesis. Explosive AI infrastructure growth — revenue up 62%, the cloud segment up 84% — with unusually strong forward visibility, since management guides 2027 to accelerate beyond the 65% pace of 2026 on the OpenAI and AMD rack programmes. But this is a fee business at around 11.5% gross margin, carrying a $1.5B inventory build.
Reviewed 11 September 2026, after Q2 2026 results. Next review after Q3 earnings on 26 October 2026.

What it does

Celestica provides electronics manufacturing services. It builds networking switches, AI compute racks and storage hardware to hyperscalers' own designs.
Two segments. CCS — Connectivity and Cloud Solutions, 81% of revenue — is the AI business. ATS — Advanced Technology Solutions — is aerospace, defence, industrial and healthcare, and moves on a completely different clock. Inside CCS, HPS (Hardware Platform Solutions) is the higher-value work where Celestica contributes design content rather than just assembly. CEO is Robert Mionis.
The distinction that matters is between building someone else's design and helping design it. HPS is the second kind, and it's why Celestica earns more than a pure assembler — though still far less than anyone who owns a scarce input.

Bull case

  • Q2 2026, reported 27 July: revenue $4.70B, up 62%, above the high end of guidance. Adjusted EPS $2.54, up 83%
  • The cloud segment is where it's happening. CCS $3.81B, up 84%, with Communications up 62% on 800G and 400G switch programmes and Enterprise up 167% on an accelerated hyperscaler AI compute ramp
  • HPS $1.9B, up 58% — 800G switch programmes across multiple hyperscalers, with 1.6T ramping
  • Returns are genuinely good for this business. Record adjusted operating margin 8.2% (+80bps) and adjusted ROIC around 55%, up about 20 percentage points year over year
  • Guidance raised twice this year — FY26 to $20.5B from $19.0B in April, with the free cash flow outlook raised to $600M
  • 2027 is guided to accelerate beyond 65% growth, with EPS growing faster than revenue. Unusually strong visibility for a contract manufacturer
  • Two named multibillion-dollar 2027 programmes — OpenAI custom racks with deliveries from late 2026, and the AMD Helios scale-up platform

Bear case

  • Adjusted gross margin is around 11.5%. This is a contract manufacturer. It earns a fee rather than scarcity pricing, and that ceiling is permanent
  • The headline EPS was flattered. GAAP EPS of $3.17 included a $0.90 per share total-return-swap gain. The adjusted $2.54 is the real number
  • Cash conversion is the weak point. Free cash flow was only $147M on $4.7B of revenue, with inventory up $1.5B year over year and capex jumping to $264M — 5.6% of revenue against 1.1% previously
  • No buybacks in the quarter — reinvestment took priority over returns
  • Concentration among a few hyperscalers, with supply chain constraints explicitly flagged as a growth limiter
  • ATS is a persistent drag at 8% growth and 6.3% margin
  • The OpenAI programme is expected to be consigned, meaning material costs won't flow through revenue. Good for margin optics, but it makes revenue comparisons harder to read
  • The stock fell around 35% from its June 2026 high of $474.03 before recovering

Major customers

Individual customer percentages are not disclosed. Concentration among a few hyperscalers is explicitly flagged as a risk.
  • OpenAI — custom AI racks with initial deliveries later in 2026, and a multigenerational custom accelerator roadmap with Broadcom starting with the Jalapeño accelerator, in mass production from 2027
  • AMD — the Helios scale-up platform, described by management as a multibillion-dollar 2027 opportunity
  • Multiple hyperscalers — 800G switch programmes across several accounts with 1.6T ramping, plus the AI compute ramp that drove Enterprise up 167%
  • ATS customers — aerospace and defence, industrial, healthcare and capital equipment. A separate, slower base
🔗
Read-through.
🕸️
AVGO is a partner here rather than a competitor — the Jalapeño roadmap is a joint Broadcom, Celestica and OpenAI effort. That makes Broadcom's XPU commentary a direct read on Celestica's 2027 revenue, and vice versa.
Against
🕸️
ANET the relationship inverts: Celestica competes for hyperscaler switch spend, so when a hyperscaler moves to white-box designs, Celestica wins what Arista loses. Watching both pages together tells you which way that decision is going at any moment.
🏛️
AMD's Helios timing is the other half of the 2027 story. Slippage in either the OpenAI or AMD programme is the main downside risk, and neither is within Celestica's control.

What would change the view

  • Whether the 2027 acceleration beyond 65% is reaffirmed at the next print — the entire forward case rests on it
  • OpenAI rack deliveries starting on schedule in late 2026
  • Free cash flow conversion — the weakest part of an otherwise excellent quarter
  • Inventory unwind as the programmes ramp against the $1.5B build
  • 1.6T mass production ramp through Q3
  • ATS segment recovery from 8% growth

Update log

  • 11 Sep 2026 — Reviewed. Verdict held at Constructive.
  • 7 Aug 2026 — Full review after Q2 2026 results. The 2027 acceleration guide and the OpenAI and AMD programmes became the centre of the thesis; the 11.5% gross margin ceiling and $1.5B inventory build flagged as the limits.
Research and education only — not investment advice.