High Stakes Advanced Node Poker: TSMC Goes All-in, Intel Folds
Understanding the insatiable demand for TSMC advanced nodes and packaging, Intel’s bleak results, the saving grace of EMIB, and the $11B CPU fumble, and AMD’s silicon fungibility advantage.
The earnings calls held weeks apart by the two major leading-edge fabs in the world could not be more different.
TSMC has a line going out the door for their chips, they can’t build factories fast enough to keep up supply, and their revenue and capex plan is through the roof. Quite literally, it is their own conservatism holding them back.
The Intel story is one of gloom and doom save for a few diamonds in the rough. 18A is not yielding well enough for a world-class foundry, they are afraid to go all in on 14A until they have customers, there are supply shortages on the chip side even when they are their own foundry, and their capex remains flat. Any glimmer of hope in the short term comes from the insatiable demand for CPUs in the agentic AI era (which they don’t have the capacity to deliver) and the potential $1 billion in revenue from advanced packaging. Long term potential still exists.
In the game of leading-edge fab poker, the flop is on the table and TSMC has a “full house.” If there was any time in history for TSMC to go all-in, it is now. TSMC realizes this and plans to significantly increase capex spend over the next three years to meet demand. Intel on the other hand is holding an offsuit 7 and 2 - statistically the worst hand in poker.
The launch of the Apple A8 in 2014, and the subsequent explosive demand for Apple’s leading-edge chips drove TSMC to spend $10-15B every year in capex for the next 6 years (SemiAnalysis has a great piece on this). In 2026 alone, TSMC is on track to spend $54B in capex to power the next generation of AI accelerator and mobile chips. However, it is not just Apple this time. TSMC has complete pricing power across mobile and AI, and even Apple - TSMC’s maker - is on the verge of being told “No soup for you! Come back 1 year!”
In this detailed report, we will discuss the following.
TMSC vs Intel Advanced Nodes: The extreme demand for N2 from both mobile and AI, N3 oversubscription, N5 lingering demand, and outlook for A16/A14 from mobile and HPC. Intel 18A demand, and uncertain future of 14A.
Advanced Packaging Dynamics: CoWoS shortage and increasing demand, TSMC repurposing facilities, rise of alternative advanced packaging solutions, and Intel EMIB as a diamond in the rough.
Server CPU demand for Agentic AI: Intel’s $11B CPU fumble, AMD silicon fungibility, and Intel’s questionable architecture choices.
If you’re not a paid subscriber, you can buy just this report in EPUB and PDF formats using the button below.
TSMC vs Intel Advanced Nodes
TSMC is clear that 70-80% of their capex spending will go towards leading edge logic nodes. Specifically towards N2 and A14 process nodes which are gate-all-around nodes that are getting increasingly expensive compared to FinFET nodes of the past. In their earnings call, they were clear that their investments in the near term are only going towards productivity improvements, and not towards fab buildouts which takes 2-3 years. As a result, the increased capex spend will do nothing to alleviate the supply shortage for TSMC chips in the near term.