Copper will keep its near-term lead over optical links inside AI racks — resolves 2027-06-30

AI racks need links that let many chips exchange data. The choice is between copper cables and optical links that carry signals as light.

This choice affects rack costs and when optical suppliers receive large orders. SecondSource expects copper to retain its near-term lead, while keeping former Intel CEO Gelsinger’s opposing view of a rapid shift to optics on record.

SecondSource expects copper cables, supported by signal-processing chips and retimers that restore signals, to remain dominant in the near term. Co-packaged optics (CPO), which places optical components beside chips, is expected to arrive later. This is a forecast to be checked in the first half of 2027, not a claim that the outcome is already known.

Terms used below: Rubin is an NVIDIA AI-platform generation; scale-up here means links among chips inside a rack. DSP means signal-processing chip; a hyperscaler is a large cloud operator. Per-engine yield is the share of individual optical-signal modules that meet requirements; 99.5% is the level named in the original rule.

The original rule has an unresolved edge case: Rubin could remain mainly copper while a large cloud operator deploys CPO at scale. The “The receipt” and “What would overturn it” sections also use different conditions: deployment alone versus yield plus deployment. The rule does not say which condition takes precedence if they conflict. That needs clarification at assessment; this card is not settled.

The original rule also uses “void” to describe the copper view being overturned, but gives no separate no-score condition. How that wording maps to settlement status also needs clarification.

More Than Moore (Ian Cutress)
“I joke that NVL72 is an engineering marvel and a…”

Reviewed in H1 2027: if Rubin-generation scale-up is still mainly copper and DSP, the call stands. Any hyperscaler deploying CPO at scale, beyond pilots, and it fails.

What would overturn it

A CPO system yield breakthrough, on the order of 99.5% per engine, together with volume deployment at any major buyer, would void the near-term copper call.