NVIDIA Could Manufacture Its Blackwell AI Chips at TSMC's Arizona Plant
2min read Multiple sources have reported that TSMC is in talks with NVIDIA to discuss the production of NVIDIA's Blackwell AI chips at TSMC's new factory in Arizona. The sources say that TSMC is preparing for production to begin early next year. Both TSMC and NVIDIA have declined to comment on this matter.
The Blackwell is an AI chip released by NVIDIA in March of this year. NVIDIA has stated that there is high demand from customers in generative AI and accelerated computing for its Blackwell chips, which are currently in short supply. It is said that the Blackwell chip is up to 30 times faster and 25 times more power-efficient in tasks such as providing answers for chatbots.
If the agreement is finalized, it will secure another customer for TSMC's Arizona factory, which is scheduled to begin mass production next year. Currently, Apple and AMD are existing customers of TSMC's new factory in Arizona.
The report mentions that although TSMC plans to produce the front-end process for NVIDIA's Blackwell chips in Arizona, these chips will still need to be shipped back to Taiwan for packaging. This is because the Arizona facility currently does not have the capability for chip-on-wafer-substrate (CoWoS) packaging, a key technology required for the Blackwell chips. All of TSMC's CoWoS packaging capacity is currently concentrated in Taiwan.
TSMC is the world's largest contract chipmaker and is investing tens of billions of dollars in building three factories in Phoenix, USA. NVIDIA is currently ramping up production of the Blackwell chips and striving to expand its capacity for next year, but it will still be in short supply.
NVIDIA is a major beneficiary of the AI boom. This year, the company's stock price has nearly tripled, with a total market value exceeding $3.5 trillion, making it the second most valuable publicly traded company in the world, just $100 billion behind Apple, the number one.
Currently, NVIDIA's next-generation flagship AI chip, the Blackwell, is being actively delivered. In the third quarter, many end customers, including Microsoft, Oracle, and OpenAI, have begun to receive the company's next-generation AI chip, the Blackwell. NVIDIA's founder and CEO, Jensen Huang, recently stated that the Blackwell is fully ramped into production.
Recently, there are also reports that Musk's AI startup, xAI, has ordered $1.08 billion worth of NVIDIA's GB200 AI servers and has obtained priority delivery rights. It is expected that NVIDIA will begin delivering these servers in January 2025, with Foxconn as the contract manufacturer.
The report states that Musk directly contacted NVIDIA CEO Jensen Huang to discuss the GB200 server order for xAI. Priority access to NVIDIA's AI servers will help xAI achieve its goals better.
Last month, there was news that Musk was discussing a potential investment in xAI with NVIDIA. At the time, Huang refused to comment on the rumors but acknowledged the hard work of the xAI team.
In late November, after completing a $5 billion financing round, xAI's valuation reached about $50 billion, nearly doubling in value within six months. According to insiders, the latest round of investors includes the Qatar Investment Authority, Valor Equity Partners, Sequoia Capital, and Andreessen Horowitz. Recently, there have also been reports that Musk's xAI company plans to launch a standalone app for its Grok chatbot as early as December to compete with OpenAI.