NVIDIA’s Bold Move: CEO Hints at Alternatives to TSMC Amid Soaring Chip Demand and Blackwell Challenges
2min read NVIDIA CEO Jensen Huang remarked at a Goldman Sachs technology conference this Wednesday that, if necessary, the company could shift away from TSMC as its primary chip manufacturer.
Huang emphasized that the adaptability and responsiveness of TSMC to NVIDIA's demands are exceptional, which is why NVIDIA currently relies on them for chip production. However, he acknowledged that if the situation demands, other suppliers could be considered. "Due to the majority of our technology being internally developed, it allows us to move orders to other manufacturers," Huang stated, cautioning that such a change might result in a decrease in chip quality.
The overwhelming demand for chips has strained relations with clients, many of whom are upset when they can't secure their desired quantities promptly. Huang noted, "The demand is so high that everyone wants to be in the first batch, and everyone wants the most. Our clients are understandably excited, and the situation is tense as we try to meet these demands." During the conversation, he also pointed out that "NVIDIA collaborates with every AI company in the world, and everyone depends on us."
Discussing the AI market's trajectory, Huang commented that although generative AI is still in its early stages, it is poised to expand beyond data centers. "What's astonishing now is that the first trillion-dollar market for data centers will happen through accelerated chip development. This evolution is inevitable," he said, adding that "generative AI is not just a tool, but a skill that will, for the first time, enhance human capabilities."
Since the release of its second-quarter financial results, NVIDIA’s stock had dipped over 13%. Investors have been anxious for updates on the development of the company's Blackwell chips, hoping for a catalyst to stabilize the stock.
Huang assured that the Blackwell chips, despite their delayed release, are in high demand, and suppliers are catching up with production schedules. He reiterated to Goldman Sachs analyst Toshiya Hari that Blackwell chips are now in full production and are expected to start shipping in the fourth quarter. As of Wednesday's close, NVIDIA's stock surged 8.15%, marking the largest single-day increase over the past six weeks and boosting its market value by $216.1 billion overnight.
During the conference, Huang highlighted that cloud service providers utilizing NVIDIA's accelerated computing technology can deploy infrastructure in the cloud, yielding remarkable returns on investment. He elaborated that every dollar spent on NVIDIA’s technology by cloud service providers translates to a $5 return. Huang projected that traditional software engineering will be supported by around-the-clock digital engineers, steering the industry's future direction.
Huang further shared that NVIDIA currently employs 32,000 people and plans to increase the number of digital engineers 100-fold in the near future to sustain the company's growth and technological advancements.
Regarding the postponed Blackwell chip, challenges related to its significant size and complexity were highlighted. Reports indicated that the Blackwell chip is roughly twice the size of the previous H100 chip, containing 2.6 times more transistors. These features necessitate substantial design modifications to enhance yield rates. Unlike a single monolithic silicon die, the Blackwell chip comprises multiple advanced-process NVIDIA processors and several storage components, forming a complex interwoven structure of silicon, metal, and plastic. Manufacturing such a chip demands near-perfect precision, as any serious defect in a component could result in failures, which becomes increasingly likely with the number of components involved.
The issue of coordinating different chip components and managing yield rates was underscored as critical. Industry expert Dan Hutchison from TechInsights noted, “The key challenge is ensuring different chip components work seamlessly together. When yield rates for individual components are not high enough, everything can quickly deteriorate.” This complexity is not unique to NVIDIA, as other chip manufacturers may also face similar issues while attempting to increase chip sizes to boost processing power.
According to AMD CEO Lisa Su, as companies seek to enhance performance by stacking more silicon layers together, the complexity will inevitably increase. She remarked, “These technological achievements demand significant expertise, and future chips will be more complex and larger. Moreover, in the context of AI data centers consuming substantial electric power, power supply issues are becoming increasingly concerning.”
Huang remains confident in tackling these technological hurdles. Reflecting on the challenges posed by the ambitious goals of the Blackwell chip, he stated earlier this year, “It's just a giant chip. When informed that Blackwell's ambitious targets surpassed physical limits, our engineers responded, ‘So what?’”