Sources: TSMC is in discussions with Nvidia to make its Blackwell AI chips at TSMC's new Arizona plant starting in early 2025, expanding on production in Taiwan
Taiwan Semiconductor Manufacturing Co (2330.TW) is in discussions with Nvidia Corp (NVDA.O) to produce its Blackwell artificial …
Context & Ripple Effects
TSMC’s Arizona build-out had already been upgraded toward 4nm production at the urging of major chip customers in the earlier plan for 4nm output in Arizona. Nvidia’s discussions would put a flagship AI processor among the prospective workloads for that capacity.
The talks also sit against signs that the Nvidia–TSMC relationship was being tested by reported Blackwell manufacturing flaws during the Blackwell ramp. Subsequent coverage records Blackwell production beginning in Phoenix, making this report an early marker of that shift.
First-order effects
- Nvidia gains a potential second geography for Blackwell fabrication alongside Taiwan, while TSMC can seek to fill new Arizona capacity with a high-value AI-chip customer.
- The reported move would require Nvidia and TSMC to coordinate Blackwell production across sites, adding an operational path beyond Taiwan even though the discussions were not yet a confirmed commitment.
Second-order effects
- Locating a leading AI-chip workload in Arizona raises the importance of nearby packaging, testing, and other production support; the practical benefit depends on whether those stages can keep pace with wafer output.
- Other large chip designers may view Nvidia’s proposed use of Arizona as a signal that TSMC’s U.S. capacity can serve advanced products, increasing competition for available local capacity.
Third-order effects
- If advanced AI chips are manufactured across Taiwan and the U.S., supply assurance becomes a differentiator in chip procurement rather than a back-end manufacturing detail.
- This is one data point in a broader split between designing leading AI hardware and securing geographically diversified capacity to make it.
The trend: AI-chip leaders are increasingly pairing performance roadmaps with manufacturing-footprint diversification, extending AI infrastructure demand into regional semiconductor supply chains.