TSMC is also basically the only supplier, which is a reason the US and EU push so much for their own production lines, although it looks like the US wants to stop theirs.
NVIDIA used Samsung for one generation, people are saying because of the deal they got, but went back to TSMC, apparently because of yield issues.
Intel was behind schedule for a long time, and even used TSMC for their current line up, but I think their new 18A process is supposed to come this year, who knows how that will turn out.
For NVIDIA specifically I’ve also heard that the HBM chips for the high-end AI cards are also a bottleneck, otherwise we might get even fewer consumer GPUs, but I never followed up on that.
Sources similar to yours, and I think that’s been the case for years: https://www.digitimes.com/news/a20250109PD237/tsmc-54nm-3nm-capacity-2025.html https://www.tomshardware.com/tech-industry/semiconductors/tsmcs-arizona-chip-fab-production-is-sold-out-through-late-2027
TSMC is also basically the only supplier, which is a reason the US and EU push so much for their own production lines, although it looks like the US wants to stop theirs.
NVIDIA used Samsung for one generation, people are saying because of the deal they got, but went back to TSMC, apparently because of yield issues.
Intel was behind schedule for a long time, and even used TSMC for their current line up, but I think their new 18A process is supposed to come this year, who knows how that will turn out.
For NVIDIA specifically I’ve also heard that the HBM chips for the high-end AI cards are also a bottleneck, otherwise we might get even fewer consumer GPUs, but I never followed up on that.