Taiwan Semiconductor Manufacturing Co. (TSMC) is in talks with NVIDIA to produce its Blackwell AI chips at its Arizona facility, Reuters reported. The company unveiled the chips in March and said they were being manufactured at its Taiwan facilities.
The publication, quoting anonymous sources, said TSMC is preparing to start production early next year at its Arizona site.
The plant, part of TSMC’s multibillion-dollar investment in Phoenix, is scheduled to begin production in 2024. The sources also added that if the deal is finalised, NVIDIA would join Apple and AMD as customers at the Arizona facility.
NVIDIA’s Blackwell chips are in high demand for generative AI and accelerated computing applications. According to the giant, the chips are 30 times faster at tasks like answering chatbot queries.
While the Arizona plant would handle the front-end process of Blackwell chip production, the chips would need to return to Taiwan for packaging due to the lack of chip-on-wafer-on-substrate (CoWoS) capacity in Arizona. All of TSMC’s CoWoS capacity is currently in Taiwan.
NVIDIA did not respond to AIM’s queries.
TSMC is also building three facilities in Phoenix with support from U.S. government subsidies aimed at strengthening domestic semiconductor production. The talks with Nvidia highlight the Arizona plant’s growing role in meeting the global demand for advanced AI chips.
In October 2024, Nvidia began shipping Blackwell chips to key partners, including OpenAI and Microsoft. OpenAI reported that the new DGX B200 systems, powered by Blackwell, provide three times faster training speeds and fifteen times greater inference performance compared to previous models.
The demand for Blackwell chips has propelled the company’s financial performance. In the third quarter of fiscal 2025, the company reported record revenue of $35.1 billion, a 94% increase year-over-year, controlling 95% of the AI chip market.
The data centre segment contributed $30.8 billion, up 112% from the same period last year. NVIDIA’s CEO, Jensen Huang, noted that Blackwell production is “running at full steam” to meet the staggering demand.
In Q1 2024, NVIDIA also led all vendors in component revenues, accounting for nearly half of the reported figures, as supplies of its H100 GPUs improved for both cloud and enterprise markets.
The capabilities of Blackwell chips have also enabled advancements in AI infrastructure, breaking the rise of Moore’s Law. In September 2024, Oracle unveiled the world’s first zettascale AI supercomputer, incorporating 131,072 Blackwell GPUs to deliver 2.4 zettaFLOPS of peak performance.
This development underscores the growing importance of high-performance AI hardware in supporting large-scale AI workloads. NVIDIA’s ongoing efforts to enhance AI chip performance and production capacity reflect the increasing demand for advanced computing solutions across various industries.
The post TSMC, NVIDIA in Talks for AI Chip Production in Arizona appeared first on Analytics India Magazine.
