TE Connectivity CPC and CMC Sealed Circular Connectors for AI Server Liquid Cooling: 2026 Sourcing Playbook
TE Connectivity CPC and CMC Sealed Circular Connectors for AI Server Liquid Cooling: 2026 Sourcing Playbook
The Thermal Wall: Why 1 kW+ GPUs Are Rewriting Connector Requirements
A single NVIDIA GB300 Blackwell Ultra GPU dissipates over 1,000 watts — a thermal load that air cooling cannot manage at the rack densities AI factories require today. By Q3 2026, direct-to-chip liquid cooling has moved from bleeding-edge to baseline architecture for hyperscalers and colocation operators building at the 100 kW-per-rack scale. Samsung's August 2026 commitment of $170 million to a new Gwangju CDU manufacturing facility is a clear signal that cooling supply is the binding constraint on AI infrastructure growth.
Every coolant tube junction, every CDU manifold port, every server-node quick-disconnect fitting needs an IP67-rated circular connector that holds 3–5 bar coolant pressure without leaking. When a connector in a 192-GPU rack fails, the downtime cost far exceeds the $3–67 price of the component. Yet cooling connectors are frequently under-stocked because they fall between the thermal team's BOM and the facilities group's procurement list.
What AI Server Liquid Cooling Demands of a Connector
The D2C cooled rack environment stresses connectors in ways industrial sensor applications do not:
- Pressure: CDU outlet manifolds typically run 3–5 bar; cold-plate circuits see pump-startup transients
- Temperature: Coolant sits at 20–45 °C but ambient air inside a failed-coolant scenario can reach 55 °C at the connector body
- Vibration: Server fans and pump pulsation create mechanical stress; bayonet-latching or threaded coupling resists disengagement
- Bulkhead penetration: Each server node chassis requires feedthrough connectors that maintain IP67 on both coolant-facing and air-facing sides
- Serviceability: CDU-to-rack connections must be mated under field conditions with positive retention
TE Connectivity's CPC (Circular Plastic Connector) Series 1 and CPC Miniature families were designed for industrial sensor and process-control — the IP67 / thermoplastic / bayonet combination that cooling circuits require. Their long catalog life means wide distributor stock, making them a practical first call for AI infrastructure procurement.
TE Connectivity CPC Series 1: Specs and Distributor Stock
The CPC Series 1 uses a size-16 contact insert (16 AWG shell) with bayonet or threaded coupling, thermoplastic housing, and crimp contacts (ordered separately). Sizes range from 4-position to 37-position. All versions carry IP67 sealing when mated, rated for 250 VAC/DC.
Catalog evidence:
| MPN | Description | Positions | Mounting | IP Rating | Price (USD) |
|---|---|---|---|---|---|
| 206060-1 | Plug housing, female sockets, coupling nut | 4 | Inline | IP67 | $3.91 – $6.45 |
| 206061-1 | Receptacle housing, male pins, flange | 4 | Panel/Flange | IP67 | $3.05 – $5.04 |
| 182649-1 | Plug housing, male pins, keyed, coupling nut | 14 | Inline | IP67 | $5.40 – $8.36 |
| 208459-1 | Receptacle, shielded, 24-position, nickel housing | 24 | Panel/Flange | IP67 (shielded) | $54.06 – $67.08 |
Size-16 contacts carry 7.5 A per contact (copper alloy, gold-plated). Thermoplastic housing rated -55 °C to +125 °C. Threaded fastening (206060-1, 182649-1) or bayonet latch (206061-1) provides retention against rack vibration.
CPC Miniature: Compact 4-Position for CDU Manifold Ports
Where space is constrained — inside a CDU manifold block or on a GPU cold-plate adapter — the CPC Miniature series delivers IP67 sealing in a smaller footprint.
Catalog evidence:
| MPN | Description | Positions | Mounting | IP Rating | Price (USD) |
|---|---|---|---|---|---|
| 1445389-1 | Receptacle housing, male pins, inline | 4 | Inline | IP67 | $7.76 – $11.59 |
| 1445390-1 | Plug housing, female sockets, coupling nut, inline | 4 | Inline | IP67 | $7.18 – $10.72 |
The CPC Miniature uses the same size-16 contact insert as Series 1 in a reduced-profile housing. Inline configurations mate as a plug-and-receptacle pair suitable for CDU manifold wall or GPU carrier plate penetration. All listed MPNs carry Active lifecycle status — critical for 5–7-year AI infrastructure deployments requiring serviceability into the 2030s.
CMC Series 1: 24-Position Shielded for Complex GPU Node Chassis
For GPU nodes that bundle power, signal, and cooling into a single chassis penetration — common in NVL72-style rack-scale designs — the CMC Series 1 24-position shielded receptacle (208459-1) provides a single bulkhead solution where separate unshielded connectors would create EMI leakage.
The 208459-1 has a nickel-plated zinc die-cast housing, keyed orientation, threaded panel mount, and EMI shielding at the chassis boundary. At $54–67 per unit, it costs more than standard CPC housings but replaces two or three unshielded connectors plus external shielding hardware.
Sourcing Reality for AI Infrastructure Buyers
Independent distribution handles most AI server ODM and colocation procurement. Distributor inventory on TE CPC/CMC Series 1 is replenished on 6–10-week cycles. There is no allocation reserve specifically for cooling applications. Buyers running large rack deployments should place 6-month rolling forecasts with their distributor to avoid shortage during ramp.
MOQ for OEM quantities starts at 100 units per MPN. Sample quantities (1–10 units) are available from franchised distributors without a formal PO.
IP67 is mandatory for any connector in the coolant path. A non-sealed connector will eventually leak coolant into the server node cavity, destroying GPU, PCB, and power delivery simultaneously. The $3–11 premium per IP67 CPC housing over a non-sealed alternative is among the highest-value line items on a GPU server BOM.
Flange-mount and inline versions serve different roles. Flange-mount receptacles (206061-1) attach to fixed server-node chassis panels. Inline plug-and-receptacle pairs (1445389-1 / 1445390-1) are correct for CDU-to-server hose connections that need float or must be mated in confined spaces.
Stock 90 days of growth buffer. AI infrastructure procurement teams who navigated the 2021–2023 connector shortage typically maintain 90-day safety stock on standard CPC housings. At 30–50% quarter-over-quarter rack deployment growth, keeping 500–1,000 units of each popular MPN (206060-1, 206061-1, 1445390-1) in regional warehouse eliminates a procurement emergency from a cooling connector shortage.
BOM for a 100 kW AI Rack with D2C Cooling
A typical 100 kW rack with direct-to-chip cooling contains 60–120 sealed circular connectors:
- CDU-to-rack manifold hose connections: 4–8 per rack (inline CPC Miniature 1445389-1 / 1445390-1)
- Server-node chassis bulkheads: 1–4 per GPU server node (flange-mount CPC Series 1 206061-1)
- Rack-level coolant busways: 2–6 per rack frame (CPC Series 1 206060-1 as feeder connectors)
- High-density GPU node chassis: CMC Series 1 208459-1 for combined power/signal/cooling penetration
At a blended average of $8–15 per connector, total connector BOM for a 192-GPU rack is under $2,000 — less than 0.05% of a $4–6 million rack hardware cost. Failure of any one of them, however, can take down the entire rack.
What to Watch in 2027
NVIDIA's Rubin architecture is expected to push GPU TDP to 1.5 kW and beyond by 2027, requiring either immersion cooling or hybrid two-phase systems in addition to D2C. Connector families rated for current CPC Series 1 conditions may need re-evaluation for higher coolant pressures and 80–100 °C coolant temperatures. Buyers locking in designs now should confirm thermal derating with TE's updated datasheets before committing to volume build.
Where to Source
TE Connectivity CPC and CMC Series 1 connector housings (206060-1, 206061-1, 1445389-1, 1445390-1, 182649-1, 208459-1) are available through independent distributors with no allocation constraints. Sample quantities accessible without formal PO. Standard distributor replenishment lead time is 6–10 weeks. For buyers scaling AI server infrastructure at hyperscale or colocation, adding a CPC housing order to the procurement queue before Q4 2026 deployments accelerate prevents a cooling connector shortage from blocking a GPU commissioning schedule.