Optics takes the new links, copper keeps the short ones: both grow through 2027

The links that join AI chips are moving from copper to light, and the question is when. Copper is cheaper and uses less power over short distances; optics carries signals further as speeds rise. Today the links inside a rack mostly run on copper, and the links between racks on optics. The claim here is that the change is gradual: through 2027, optics takes the new and longer links on schedule, while copper keeps growing in the short ones, so both grow and neither displaces the other yet. The evidence on both sides is in the companies' own filings. Optics is arriving: Coherent's 4 November results guide its co-packaged and near-packaged optics line, PhotonLink, to more than $100M of revenue in the March 2027 quarter and $250M a quarter before the end of 2027, with a deal worth about $20B over five years starting in 2028; Lumentum reports more than $5B of commitments for the same kind of product; and Marvell, agreeing on 2 November to buy Lumentum in an all-stock deal valued at about $115B, cited copper's limits as a reason. Copper is growing too: Credo, the main maker of active copper cables, reported revenue of $479.0M in its quarter to 1 August, up 114.7%, at a 68.0% non-GAAP gross margin, and guided higher again. The newest rack designs from NVIDIA and Arista still use copper inside the rack. Conviction is MEDIUM. Both tests rest on single companies: Coherent's guidance for optics, and Credo's revenue for copper, which is a rough gauge because Credo also sells optical chips. Lasers are short, which could slow optics, and NVIDIA's next rack generation could move to optics sooner than its current designs suggest. It fits MenFem's calls as they stand: bullish on Credo and Arista, watching Coherent, Marvell and Astera Labs. BULLISH on the thesis, conviction MEDIUM, one year. This view was formed by MenFem's research process at Connor's request and adopted by him.