AI chip! NVIDIA PK AMD TSMC takes orders with advanced manufacturing processes and strong momentum
Nvidia and AMD are fiercely competing in the AI chip market this year. The AMD MI300A series products have started mass production and shipment this season, and have been actively adopted by customers. Nvidia will launch an upgraded version of the AI chip to compete, and TSMC will take orders from Nvidia and AMD, becoming a big winner.
Industry estimates indicate that Nvidia and AMD have a total shipment of at least one million AI chips this year, with an upper limit of 1.5 million chips, injecting strong momentum into TSMC's advanced 5-nanometer and 3-nanometer process orders.
TSMC has never commented on customer and order dynamics. However, TSMC President Wei Zhejia mentioned at the supply chain management forum in December last year that due to external factors such as high inflation and continuously rising costs, there is still uncertainty in 2024. However, benefiting from the rapid development of AI applications, 2024 will also be a year full of opportunities.
The industry points out that the global AI craze will start to explode in 2023 and continue to become the industry's focus in 2024. Unlike 2023, Nvidia, which previously dominated the AI high-performance computing (HPC) field, will face the challenge of AMD's MI300 series product line starting to ship and compete for the market this year.
It is reported that AMD MI300A products will begin mass production and shipment this season. Its central processing unit (CPU) and graphics processing unit (GPU) small chips will be produced using TSMC's 5-nanometer process, while the IO small chips will be produced using TSMC's 6-nanometer process. They will be integrated through TSMC's new system integrated chip packaging (SoIC) and advanced packaging such as CoWoS.
In addition, AMD's MI300X products without integrated CPU chips are also being shipped simultaneously. Compared to NVIDIA's GH200 with integrated CPU and GPU, and H200 with pure GPU computing, AMD's new AI performs better than expected in terms of computing power, with a lower price and a high cost performance advantage, attracting system manufacturers to adopt it.
Due to the fact that cloud service giants such as Microsoft and Meta began ordering AMD MI300 series products one or two years ago, and requested ODM factories to design AI servers specifically using the MI300 series product line to diversify risks and reduce costs. The industry estimates that the demand for AMD MI300 series chips in the market this year will reach at least 400000 units. If TSMC provides more production capacity support, there is a chance to look at 600000 units.
In response to the fierce competition from AMD, Nvidia is upgrading its product line and is expected to launch new products such as B100 and GB200 using TSMC's 3-nanometer process by the end of the year, as its H200 and GH200 chips continue to be in short supply.
The legal representative is optimistic that Nvidia's AI chip shipment momentum this year will be at least 1 million, with a multiple increase compared to 2023. With the addition of AMD MI300 series chips, TSMC's total AI high-performance computing chips from NVIDIA and AMD will exceed one million this year, with an increase of 1.5 million chips, which will help improve TSMC's utilization of advanced processes such as 3 nanometers and 5 nanometers. This will drive TSMC's performance to rise quarter by quarter and return to the growth track for the whole year.
TSMC has never commented on customer and order dynamics. However, TSMC President Wei Zhejia mentioned at the supply chain management forum in December last year that due to external factors such as high inflation and continuously rising costs, there is still uncertainty in 2024. However, benefiting from the rapid development of AI applications, 2024 will also be a year full of opportunities.
The industry points out that the global AI craze will start to explode in 2023 and continue to become the industry's focus in 2024. Unlike 2023, Nvidia, which previously dominated the AI high-performance computing (HPC) field, will face the challenge of AMD's MI300 series product line starting to ship and compete for the market this year.
It is reported that AMD MI300A products will begin mass production and shipment this season. Its central processing unit (CPU) and graphics processing unit (GPU) small chips will be produced using TSMC's 5-nanometer process, while the IO small chips will be produced using TSMC's 6-nanometer process. They will be integrated through TSMC's new system integrated chip packaging (SoIC) and advanced packaging such as CoWoS.
In addition, AMD's MI300X products without integrated CPU chips are also being shipped simultaneously. Compared to NVIDIA's GH200 with integrated CPU and GPU, and H200 with pure GPU computing, AMD's new AI performs better than expected in terms of computing power, with a lower price and a high cost performance advantage, attracting system manufacturers to adopt it.
Due to the fact that cloud service giants such as Microsoft and Meta began ordering AMD MI300 series products one or two years ago, and requested ODM factories to design AI servers specifically using the MI300 series product line to diversify risks and reduce costs. The industry estimates that the demand for AMD MI300 series chips in the market this year will reach at least 400000 units. If TSMC provides more production capacity support, there is a chance to look at 600000 units.
In response to the fierce competition from AMD, Nvidia is upgrading its product line and is expected to launch new products such as B100 and GB200 using TSMC's 3-nanometer process by the end of the year, as its H200 and GH200 chips continue to be in short supply.
The legal representative is optimistic that Nvidia's AI chip shipment momentum this year will be at least 1 million, with a multiple increase compared to 2023. With the addition of AMD MI300 series chips, TSMC's total AI high-performance computing chips from NVIDIA and AMD will exceed one million this year, with an increase of 1.5 million chips, which will help improve TSMC's utilization of advanced processes such as 3 nanometers and 5 nanometers. This will drive TSMC's performance to rise quarter by quarter and return to the growth track for the whole year.