TSMC executives described a semiconductor industry growing faster than earlier forecasts, with AI demand driving a striking share of the expansion. The figures also point to pressure on chip design, memory and power efficiency, not just a race to build more processors.
TSMC Watch analysis
What happened
At TSMC’s Open Innovation Platform North America Technology Symposium, executives presented forecasts and technology updates, according to Semiconductor Engineering. The publication says TSMC North America CEO Sajiv Dalal forecast that the semiconductor industry would reach about $1.7 trillion by the end of 2026, with more than $1 trillion from AI alone. Those are forecasts, not confirmed year-end results.
The report also cites TSMC executives on a fourfold rise in new tape-outs at N5, N3 and N2, measured in the second year of each technology. Another executive said inference token use had grown 500-fold over three years and was becoming a driver of AI growth.
Our top picks
- AI’s share of chip spending
TSMC’s forecast puts AI at more than $1 trillion of a projected $1.7 trillion semiconductor market in 2026. - More designs at advanced nodes
TSMC’s reported fourfold tape-out increase across N5, N3 and N2 signals a growing pipeline of chip designs. - Inference is taking a larger role
A TSMC executive cited 500-fold growth in inference token use over three years and called inference a growth driver. - A14 performance and power targets
The report says TSMC’s A14 Nanoflex is projected to improve speed by 14% and reduce power by 23% versus N2P. - Ambitious memory forecasts
TSMC’s five-year projections include bandwidth increases of 34 times for HBM, 115 times for 3D DRAM and 830 times for SRAM.
Why it matters
These projections sketch a supply chain challenge broader than producing more AI accelerators. More inference use means pressure on compute, memory bandwidth and power, while the rise in new designs suggests chipmakers and customers are planning for further demand. Whether forecasts translate into actual products and revenue remains to be seen; this is a view from a company symposium, not a tally of future results.
Our read
The useful signal is the combination: AI demand is shaping both the volume of chip designs and the technologies being developed to feed data to them efficiently. The eye-catching numbers deserve attention, but they are targets and forecasts, not delivered gains. The spreadsheet is ambitious; manufacturing still has to do the difficult bit.
What to watch
- Whether reported industry and AI revenue forecasts are borne out by results.
- How the fourfold tape-out increase develops into products and manufacturing demand.
- Whether the projected memory and power improvements appear in commercial technologies.
- How much of future AI chip demand comes from inference rather than training.
Discussion spark: Should chipmakers prioritise more compute for AI, or invest more heavily in memory bandwidth and power efficiency to make existing compute go further?
Sources and evidence
- TSMC OIP: Chip Industry Growth Blows Past Forecast – Semiconductor Engineering (29 September 2026, 07:33 UTC)
- People of HAI | Stanford HAI (29 September 2026, 07:33 UTC)
Independent WittyWires tracker for public updates about TSMC. Not affiliated with or endorsed by TSMC; this is not an official account.