[Apple Shockwave] Nvidia Targeting Possible Because It's Apple

Apple Fully Discloses AI Training Using Google Chips
Jensen Huang Fires Back as Zuckerberg Praises Apple Critically
'Far-off Story' for Companies Struggling Without GPUs

NVIDIA CEO Jensen Huang praised Meta CEO Mark Zuckerberg, who is standing up against Apple’s attempt to introduce a closed artificial intelligence system. In response, Apple revealed that it used Google’s chips instead of NVIDIA’s for its own AI, Apple Intelligence, escalating the rivalry. Whether these two companies, struggling to find common ground in the AI era, will form a collaborative relationship or continue their conflict is drawing attention as a key point to watch in both the semiconductor market and the AI sector.

Jensen Huang, CEO of NVIDIA, and Mark Zuckerberg, CEO of Meta, are laughing after swapping jackets following a discussion at the SIGGRAPH event held on the 28th.
[Photo by AP News]

On the 29th (local time), at SIGGRAPH, a computer graphics event held in Denver, Colorado, Jensen Huang mentioned during a conversation with Zuckerberg that Meta likely owns about 600,000 NVIDIA GPUs.

Huang said, "The computers Zuckerberg built are amazing. You joined the GPU era later than others, but you have scaled up more than anyone else."

Zuckerberg responded to Huang’s remarks, saying, "We are good customers. That’s why we can have a Q&A with Jensen here today," eliciting laughter from the audience. He was hinting that he had been heavily purchasing NVIDIA GPUs.

Zuckerberg changed Facebook’s name to pursue the metaverse business but had a bitter experience. What saved Zuckerberg was the AI revolution. He quickly adapted to the AI era by purchasing NVIDIA GPUs in large quantities.

Zuckerberg also praised Huang. When Zuckerberg said, "It’s fun to watch you. Everyone’s attention has shifted to this (GPU), and you guys are still working hard," Jensen replied, "We went to unexpected, unpopular places, but now it’s cool."

Currently, their common adversary is Apple. Zuckerberg and Huang have unpleasant memories of their relationship with Apple. Zuckerberg experienced a hit to Facebook’s advertising performance after Apple blocked app tracking transparency on the iPhone. This incident confirmed Apple’s power as the platform dominating the iPhone. Apple is also known to have pre-installed OpenAI’s ChatGPT for free during the rollout of Apple Intelligence.

Zuckerberg has publicly expressed hostility toward Apple by open-sourcing his self-developed AI, Llama. During the conversation with Huang, Zuckerberg stated, "I get angry when talking about closed platforms. While Apple seemed to have won the mobile era, I am confident that the open ecosystem will win in the next generation." Previously, Zuckerberg had harshly criticized Apple’s closed AI policy through a blog post when releasing Llama 3.1.

Huang also has bad memories with Apple. Many companies lined up to request chips from NVIDIA, but Apple was different. Apple used NVIDIA GPUs in its computers in the past, but after experiencing overheating issues, it had conflicts with NVIDIA and has long since cut ties with them.

While the two were having a friendly conversation and Huang was changing into a new leather jacket with Zuckerberg, Apple launched an unexpected counterattack. Apple revealed through a paper that it used Google’s chip, the 'TPU,' instead of NVIDIA’s GPU for training Apple Intelligence, which will be unveiled this fall. Although it was somewhat known that Apple was not using NVIDIA’s GPUs, this was a public declaration.

The paper titled 'Apple Intelligence Foundation Language Models' (AFM) disclosed that the AFM on-device and AFM server models, which form the basis of Apple Intelligence, were trained on a 'Cloud TPU cluster.'

It is presumed that the purpose of the paper Apple released was to clarify controversies about using YouTube subtitles for AI training, but public attention focused solely on the use of Google’s chips for AI training.

Apple did not purchase Google’s chips. Google does not sell TPUs externally. This means Apple rented Google’s cloud to conduct training. Apple had partially revealed in June that it used Google TPU for training, but this time it disclosed more detailed information. What is certain is that Apple is pushing forward Apple Intelligence without NVIDIA’s support.

Sam Altman, CEO of OpenAI, is seen sitting in the audience rather than on the podium at Apple's WWDC 2024 event on June 10. Apple announced that OpenAI's ChatGPT will be integrated with Apple Intelligence and pre-installed on the iPhone, but Altman was not allowed to appear on stage.
[Photo by Getty Images Yonhap News]

CNBC assessed Apple’s move as a sign that efforts to move away from NVIDIA GPUs, which are essential for AI training, are becoming serious. It interpreted this as an attempt to seek alternatives to NVIDIA, which requires paying tens of thousands of dollars per chip and is difficult to purchase on time.

Apple specialist media AppleInsider has a slightly different view. The outlet argued that the initial training performed with Google’s chips has little long-term significance. AppleInsider predicted that Apple is more likely to focus on developing its own AI chip, called the 'ACDC project.' Apple plans to invest over $5 billion to expand servers supporting Apple Intelligence. Additionally, AppleInsider expects Apple to pursue more mergers and acquisitions to support software development.

It is not easy to talk about AI development excluding NVIDIA now. Many challengers declare they will surpass NVIDIA by developing new chips, but the dominant position NVIDIA has held for over a decade in GPUs and related hardware and software ecosystems is difficult to break. Some argue that Apple’s unique closed AI system allows it to declare independence from NVIDIA.

There is also an argument that this phenomenon is just their own league. A researcher recently told a reporter, "We can conduct sufficient AI-related research if we just have GPUs. It’s unfortunate that we can’t because we don’t have GPUs." The conflict between Apple and NVIDIA is a story from another country, while our companies and researchers are still desperate for GPUs.