top of page

📢 STT Premium is now open

Every free article stays free — nothing moves behind a paywall. Premium adds the weekly deep-dive, full earnings breakdowns and all long-form reports. Note: premium articles are published in Traditional Chinese.

Free keeps you current. Premium lets you judge.

NT$199/month · NT$2,000/year (about NT$167/month)

👉 Subscribe: vocus.cc/salon/simpletechtrend

Barclays 23rd Annual Global Technology Conference | Fabrinet | From Optical Leader to AI HPC: Reading the DCI Push and New AWS Orders

Oct 5
3 min read

This Barclays conference session featured Fabrinet (FN) CEO Seamus Grady.

Fabrinet has long been the undisputed leader in optical communications contract manufacturing (notably as a major contract manufacturer of Nvidia's optical modules), but in this interview management sent a clear signal: Fabrinet is replicating its precision manufacturing capabilities in High-Performance Computing (HPC) and Data Center Interconnect (DCI). Here are my key takeaways and analysis:

1. DCI (Data Center Interconnect): Must-Have Demand Under Power Constraints

Fabrinet decided to break DCI out of traditional telecom revenue and report it separately, reflecting a shift in market structure.

  • Drivers (The Why): Modern AI data centers consume enormous power, and a single site can't secure enough of it. That forces data centers toward a "distributed" architecture, spreading compute across different geographic locations.

  • Scale Across: Connecting these distributed data centers so they operate like one giant computer requires DCI technology.

  • Technical specs: Demand is concentrated in 400ZR and 800ZR (Coherent Optics).

  • Customer concentration: There are currently 5 customers, 2 of which contribute most of the revenue. This market is driven directly by AI compute expansion, not the cyclical demand of traditional telecom.


2. HPC (High-Performance Computing): The Key Win at AWS

This was the biggest highlight of the interview. Although Fabrinet usually keeps customers confidential, the interview directly confirmed its HPC collaboration with AWS.

  • Scope of work: Producing ultra-high-density, high-complexity PCBAs (Printed Circuit Board Assemblies) for AWS's custom AI chip/server programs (presumably related to Trainium/Inferentia).

  • Ramp timeline:

    • Revenue last quarter was about $15M (qualification stage only).

    • One production line is currently qualified, with two more in qualification.

    • The first phase of volume production is expected to peak in the June 2025 quarter.

  • Business model: Seamus stressed that Fabrinet is a Pure Contract Manufacturer (CM), not an ODM. That means the IP stays with the customer, and Fabrinet wins orders through superior manufacturing yield and cost control. Unlike the ODM model of Quanta or Wistron, this avoids potential competition with its customers.


3. AI Datacom: Demand Still "Insatiable"

On the Nvidia-related business (Datacom) that everyone cares about most, demand shows no sign of slowing.

  • The 1.6T era arrives: The product mix already includes 800G and 1.6T transceivers.

  • Technical detail: Products based on 200G per lane EML are in production.

  • Supply chain bottleneck: The current constraint isn't Fabrinet's capacity but shortages of specific upstream components. If supply eases, they could ship more. This suggests the tightness in the optical communications supply chain will persist.


4. Capacity Expansion: Building 10 and Plans for Buildings 11/12

To meet this demand, Fabrinet's expansion plans are very aggressive:

  • Building 10: Adds capacity for $2.5B of potential annual revenue (2 million sq ft). As a stopgap, part of the space (250,000 sq ft) has been pulled forward to open next June.

  • Future plans: The campus has room for Buildings 11 and 12, which could provide another $2.5B of capacity.

  • Financial discipline: This is Fabrinet's greatest strength. Building 10's CapEx is only $130M, and with $1B in cash and zero debt, the company can fully fund it internally. Even in the worst case (an idle factory), the gross margin impact is only 15 bps — very limited downside risk.


Conclusion

Fabrinet is proving it is more than "Nvidia's optical module assembler." By entering AWS's HPC supply chain and building an independent DCI growth curve, it is diversifying risk and expanding its TAM.

Notably, Fabrinet's Operating Margin rises as revenue scales (operating leverage), because the new businesses require almost no additional OpEx. In the AI hardware arms race, Fabrinet remains one of the steadiest "picks and shovels" players.

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page