Now Playing
TSMC's 2nm Bet, Inference Surge & the $1.5tn Chip Forecast
Audio is available on Spreaker — see link below.
TSMC's deputy COO has put a $1.5 trillion chip revenue target on the board by 2030, with AI inference overtaking training as the dominant demand driver. From 2nm factory ramp rates to CoWoS packaging yields and the Nvidia H200 export freeze, this episode unpacks what the forecast means for semiconductor investors and engineers.
Audio is available on Spreaker — see link below.
TSMC's deputy chief operating officer just told the market something that reframes the entire semiconductor investment thesis. Global chip revenue is heading to one point five trillion dollars by twenty thirty, and fifty-five percent of that growth comes from AI.
Here's the detail that matters most inside that forecast. Right now, AI training and AI inference split chip demand roughly fifty-fifty.
To meet that curve, TSMC isn't adding capacity at a normal pace. They're running at a seventy percent annual compound growth rate for two nanometre production across five factories through twenty twenty-seven and twenty twenty-eight.
TSMC is also launching a new photonics platform called COUPE in twenty twenty-six. The compact universal photonics engine is designed to reduce coupling loss in chip-to-chip optical connections, which directly addresses energy efficiency in large data centre deployments.
Now, pull back to the geopolitical layer, because it complicates the growth story in ways that matter to investors. Nvidia H200 chips have been formally approved for export to major Chinese firms.
The honest read on TSMC's positioning is this: the strategic logic is sound, the execution risk is real, and the geopolitical variable is genuinely unresolved. A seventy percent compound annual growth rate for two nanometre requires flawless yield scaling across five simultaneous fabs.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.