The AI chip war is in full swing. At this year's Taipei International Computer Show (Computex), chip giants such as Nvidia, Intel, and AMD took turns releasing the latest chip products and technology trends, showcasing their latest advances in artificial intelligence, product updates, and technology routes. Although this science and technology event is called a computer exhibition, it has become a “battleground” for giants to compete for AI.
Nvidia and AMD volumes “change chips year by year”, Intel wants to fight a “price war”
Nvidia first appeared. As a leader in the AI industry for the past two years, Nvidia CEO Huang Renxun made a big announcement that the company's most powerful AI chip Blackwell series products have been fully put into production and are expected to be shipped to customers in the second half of 2024.
According to reports, Nvidia's first Blackwell chip is named GB200, which claims to be the “most powerful chip in the world” at present. Currently, the supply chain has high expectations for GB200. It is estimated that in 2025, the shipment volume may exceed one million units, which will account for nearly 40% to 50% of Nvidia's high-end GPU shipments.
In addition to showcasing the latest mass-produced Blackwell chip, Hwang In-hoon also announced the technology roadmap for the next three years: By 2025, Blackwell's upgraded version, Blackwell-Ultra-GPU, will be launched, equipped with a 12-layer HBM3e, and the memory capacity will increase by about 50%. The next AI platform in 2026 will be named Rubin, and Rubin Ultra will be launched in 2027. In other words: new versions of AI technology are expected to be launched every year; each one is faster and more energy efficient than the previous one.
Notably, Nvidia is starting a new generation of products each year, breaking “Moore's Law.” Previously, the company had been following the industry's average update cycle of two years to launch chips.
After Nvidia announced the new AI chip information, the “rival” AMD was no slouch either. CEO Su Zifeng said that in the fourth quarter of this year, AMD will launch an improved version of the MI300X, the MI325X, which will be equipped with HBM3E (High Bandwidth Memory) memory, with larger memory and improved computing power. The Mi325x is a strong opponent against Nvidia Hopper. According to Su Zifeng, the computing performance of the MI325X is 1.3 times that of the Nvidia H200 (the H200 is the Hopper product with the strongest performance of the previous generation architecture, the performance is 2 times that of the H100), and is scheduled to be launched in the fourth quarter of 2024.
In terms of chip iteration speed, AMD is on par with Nvidia, and has also proposed maintaining a higher speed year by year. In addition to the MI325X launched later this year, the Mi350 series based on the new cDNA4 architecture will be launched in 2025, and the MI400 series will be launched in 2026.
AMD is the second largest GPU manufacturer in the world and one of the main CPU manufacturers. It is second only to Nvidia in the GPU market and a competitor to Intel in the CPU field. Su Zifeng also highlighted this advantage in her speech: “Many enterprise customers want to perform general computing and artificial intelligence computing without adding GPUs. AMD is the only company that can provide a full range of CPU and GPU networking solutions to data centers.”
As a result, AMD also announced that it will launch the 5th generation EPYC CPU processor for data centers, codenamed Turin. The processor is based on Zen5 architecture and will be launched in the second half of this year. The flagship product has 192 Zen5 cores and 384 threads. Su Zifeng explained that Turin's performance advantages are outstanding when running smaller big language models.
In order to keep up with the fierce competition between Nvidia and AMD, Intel offered a price weapon, revealing that the price of its Gaudi 3 AI chip will be far lower than that of competitors. Additionally, a new generation of Xeon data center processors, as well as Lunar Lake notebook processors for the next generation of AI PCs, have also been introduced.
Intel emphasized the importance of price advantages: “As digital transformation accelerates, businesses are under increasing pressure to update outdated data center systems to achieve cost savings, achieve sustainability goals, and maximize the use of physical space in the rack.”
Guadi 3 is the latest generation AI chip launched by Intel in April this year. It is scheduled to be widely launched in the third quarter of this year. According to the official positioning, the performance of this chip is 1.5 times that of the Nvidia H100, which is roughly equivalent to the H200.
Intel says the Gaudi 3 accelerator kit, which includes eight AI chips, costs around $125,000. According to estimates from supplier Thinkmate, a similar server system equipped with eight Nvidia H100 AI chips could cost more than 300,000 US dollars, about 2.5 times that of the former.
Unlike chip designers Intel and AMD, Intel not only designs chips, but also makes chips. However, Intel's revenue has declined over the past two years due to being surpassed by rivals TSMC and Samsung Electronics. CEO Gail Singh said Intel wants to regain its position as the world's leading chip manufacturer.
The stock price gap is huge! The total market value of Nvidia's US stock is equivalent to 22 Intel and 11 AMDs
As of Tuesday's close, Nvidia's closing stock price was $1164.37, and the total market value of US stocks was $2.86 trillion, which is equivalent to the total market capitalization of 22 Intel ($127.8 billion) and 11 AMD US stocks (market capitalization of $258.5 billion).
Judging from financial data and stock price performance, these three giants also have a “huge gap.” Nvidia's earnings per share increased by 629% in the latest fiscal quarter, and the stock price increased nearly 7 times in the past two years; AMD's earnings per share rose 177% in the first quarter, and the share price rose 147% in two years; while Intel recorded a loss, the increase was only 17% during the same period.
Previously, in the AI craze, Intel had always been on the edge, while Nvidia and AMD were far ahead. Tech giants such as Meta, Microsoft, and Google all snapped up Nvidia chips.
Currently, a fierce “AI war” has begun completely in Silicon Valley. In the most competitive AI hardware layer and model layer, more and more tech giants are “fighting” with each other. For example, Nvidia is also facing a mass siege from peers to customers; Microsoft, Google, and Amazon are all developing their own AI chips.
Although industry leader Nvidia is currently in the limelight, there are also AMD that continues to accelerate product updates and performance improvements to catch up, and Intel is attracting users through cost-effective server CPUs and AI chip products. In the future, the competition between these three chip giants will be more intense.
Dear cowboy friend,
What do you think of the AI competition between the three chip giants?
Will the stock price interpretation change in the future?
Welcome to post your opinions in the comment area~
edit/new