본문 바로가기
bar_progress

Text Size

Close

NVIDIA Blackwell Mass Production Begins... AI Triangular Alliance Grows Stronger

Q3 Sales Surge 94% Year-on-Year
Blackwell, SK Hynix and TSMC Collaboration Essential

NVIDIA, which posted a 'surprise performance' in the third quarter (August to October) of this year, announced that it will proceed with the launch of the next-generation AI chip ‘Blackwell’ as scheduled, strengthening the 'AI triangular alliance' among NVIDIA, TSMC, and SK Hynix.


On the 20th (local time), NVIDIA announced that it recorded revenue of $35.08 billion (49 trillion KRW) and earnings per share of $0.81 (1,134 KRW) in the third quarter of this year. Both revenue and earnings per share exceeded Wall Street expectations by 5.8% and 8%, respectively. Fueled by the AI boom, revenue surged 94% year-on-year, and net profit soared 106% during the same period to $9.24 billion.


Jensen Huang, CEO of NVIDIA, said during the third-quarter earnings conference call, "The AI era has already begun, and AI is changing the landscape of all industries, companies, and countries," adding, "Expectations for Blackwell, which has entered full-scale production, are at an astonishing level."


NVIDIA Blackwell Mass Production Begins... AI Triangular Alliance Grows Stronger Jensen Huang, CEO of NVIDIA. Provided by NVIDIA

Blackwell is the successor to the previous generation Hopper architecture and is a high-performance graphics processing unit (GPU) that dramatically improves AI computing performance. Blackwell is equipped with the HBM3E 12-stack product exclusively supplied by SK Hynix. Data processing speed is 2.5 times faster than its predecessor Hopper, and inference performance has improved up to 30 times. Power consumption is reduced to about one-quarter. Big tech companies such as Amazon, Microsoft (MS), and Google placed orders even before its release to build AI data centers. When unveiled in March this year, CEO Huang introduced Blackwell as "the most powerful chip in the world" and said it would "open a new era of computing."


NVIDIA stated in this earnings announcement that full-scale production and shipment of Blackwell will begin in the fourth quarter of this year and gradually expand from next year. However, supply is struggling to keep up with the rapidly increasing demand. CEO Huang acknowledged, "Currently, market demand exceeds supply," and said, "We are maximizing production capacity to resolve this."


Colette Kress, NVIDIA’s Chief Financial Officer (CFO), said, "As chip supply increases, Blackwell sales in the fourth quarter may exceed the previously expected 'billions of dollars'." Such remarks from NVIDIA’s management are interpreted as emphasizing the establishment of a stable supply chain while dispelling concerns about production disruptions due to recently raised design flaws and server overheating issues with Blackwell.


NVIDIA Blackwell Mass Production Begins... AI Triangular Alliance Grows Stronger Chairman Chey Tae-won of SK Group (left) and Chairman Wei Zhejia of TSMC are posing for a commemorative photo at TSMC headquarters in Taipei, Taiwan, on June 6. Provided by SK Group

The 'triangular alliance' among NVIDIA, SK Hynix, and foundry TSMC is expected to be further strengthened with the launch of Blackwell. The success of Blackwell is impossible without the support of SK Hynix and TSMC. SK Hynix supplies high-bandwidth memory (HBM) to TSMC, and TSMC manufactures the AI accelerators designed and ordered by NVIDIA based on this. Through the cooperation of SK Hynix and TSMC, NVIDIA’s 'Blackwell' is completed.


The three companies in the AI triangular alliance are also continuing their high-flying performance in earnings. SK Hynix recorded its highest-ever quarterly revenue, operating profit, and net profit in the third quarter of this year. TSMC also exceeded market expectations during the same period, recording approximately 32 trillion KRW in revenue and about 14 trillion KRW in net profit, achieving an operating margin of 47.5%.


As Blackwell shipments ramp up, the proportion of HBM sales in SK Hynix’s DRAM is expected to expand from the 30% range in the third quarter to over 40% in the fourth quarter. SK Hynix began mass production of the HBM3E 12-stack used in NVIDIA Blackwell at the end of September and plans to supply HBM4 12-stack in the second half of next year. The HBM4 12-stack product will be applied to NVIDIA’s 'Rubin' GPU, scheduled for release in 2026.


Earlier, Kwak No-jeong, CEO of SK Hynix, expressed a strong commitment to actively respond to the AI market at the 'SK AI Summit 2024' earlier this month, stating, "We are currently developing 16-stack HBM3E and plan to provide samples to customers early next year."


© The Asia Business Daily(www.asiae.co.kr). All rights reserved.


Join us on social!

Top