Apple’s latest technical document reveals that the two main AI models behind Apple Intelligence are trained using Google’s Tensor Processing Units (TPUs) instead of NVIDIA GPUs. According to a report from Commercial Times, this suggests that the demand for NVIDIA chips has outstripped supply, prompting some tech giants to seek alternatives.

Apple first introduced an AI technical document in June, briefly stating that its AI models were trained using TPUs. The latest technical document, which spans 47 pages, provides a detailed explanation of how Apple’s foundational models (AFM) and AFM servers are trained in Cloud TPU Clusters. This indicates that Apple rents cloud servers from cloud service providers to train its AI models.

In the document, Apple stated: “This system allows us to train the AFM models efficiently and scalably, including AFM-on-device, AFM-server, and larger models.”

Apple further mentioned that the on-device AFM models for iPhones and iPads are trained using a total of 2,048 TPUv5p chips, which are currently the most advanced TPU chips on the market. The AFM servers are trained using a total of 8,192 TPUv4 chips.

Google initially launched TPUs in 2015 for internal training use only and started offering TPU rental services to external clients in 2017. These TPUs are currently the most mature custom chips used for AI training. According to Google’s official website, the rental cost of their most advanced TPUs is approximately USD 2 per hour based on a three-year contract.

Though NVIDIA’s GPUs are currently dominating the high-end AI chip market, the enormous number of chips required for AI model training has led to a severe shortage. This is because major tech companies like OpenAI, Microsoft, Google, Meta, Oracle, and Tesla all use NVIDIA chips to develop their AI technologies.

Since the rise of ChatGPT at the end of 2022, which spurred the generative AI market, Silicon Valley tech giants have been racing to invest in AI research and development. In contrast, Apple has lagged behind its competitors and now has to intensify its efforts to bolster Apple Intelligence. On July 29th, Apple released a preview version of Apple Intelligence for certain devices.

Read more

• [News] NVIDIA’s H200 Order Delivered from Q3, Boosting Server Supply Chain with Strong Demand
• [News] US Government Reportedly Investigating Microsoft, NVIDIA, and OpenAI for AI Industry Monopolization

(Photo credit: NVIDIA)

Earlier on July 10^th, in the Galaxy Unpack 2024 event in Paris, Samsung introduced its AI smartphone Galaxy Z Fold6 and Galaxy Z Flip6, along with the “Google Gemini” app installed in the models. However, according to the reports by Reuters and Business Korea, these two tech giants may be facing the investigation of the European Union (EU) antitrust regulators, on whether the collaboration might impede market access for other AI developers or limit competition.

It has been a while since the two tech heavyweight started to team up on AI. Samsung’s first AI phone, Galaxy S24, released earlier this year, has featured its self-developed AI, Gauss, as well as Google’s Gemini Nano.

The EU has ramped up its market monitoring efforts following the implementation of the Digital Markets Act (DMA) in March, focusing on major global tech companies. The DMA identifies seven companies—Google, Amazon, Apple, Meta, Microsoft, Booking and ByteDance —as ‘gatekeepers’ and imposes specific regulations on them to ensure fair competition in the digital market, Business Korea noted.

Now the spotlight has been shifted to the AI sector. According to Reuters, EU antitrust regulators are inquiring whether Google’s multi-year generative AI deal with Samsung hampers rival chatbots on Samsung smartphones. The report noted that last month, the European Commission announced it would send requests for information to understand the impact of the deal, which involves Samsung embedding Google’s Gemini Nano in its Galaxy S24 series smartphones.

According to the Reuters report, regulators are investigating if the pre-installation of Gemini Nano limits the functionality of other chatbots and applications on Samsung smartphones. The EU also asked companies whether they had attempted to enter into pre-installation agreements with Samsung for their chatbots and, if so, requested explanations for any failures. Respondents are required to complete the urvey by this week.

If anti-competitive practices are confirmed, an antitrust investigation against Google and Samsung could be initiated, Reuters stated.

According to Business Korea, after the Galaxy Unpacked 2024 event, TM Roh, President of Samsung Electronics (Head of MX Business Division), addressed regulatory risks during a press conference, saying the company is having various discussions internally and externally regarding EU regulations.

Roh also emphasized Samsung’s commitment to data security, stating that sensitive information is processed on-device (without connecting to external servers), making it inaccessible even to the company,” and highlighted the company’s approach to giving consumers the choice of using AI functions on-device or via the cloud.

Read more

• [News] NVIDIA Faces Potential Antitrust Charges in France, U.S. and EU Close Behind
• [News] Samsung Rumored to Mass-Produce HBM4 with 4nm Process

(Photo credit: Samsung)

Google’s Tensor G4 could mark Samsung’s last mass-produced SoC, as earlier in May, Tensor G5 is reportedly adopting TSMC’s advanced 3nm process. Now here’s the latest development. According to a report by Wccftech, the chip, to be used in Google’s upcoming Pixel 10 lineup, has already reached tape-out, with mass production expected in 2025.

Google’s Tensor G5 would be its first fully self-designed smartphone SoC. Previous Tensor chips, somehow, were modifications from Samsung’s Exynos series, with Samsung being its foundry partner.

The report stated that Google’s decision to collaborate with TSMC is influenced by the Taiwanese semiconductor company’s established reliability in mass-producing wafers using its next-generation nodes.

Before Google, the foundry behemoth has already secured several major clients for its 3nm node. Both Qualcomm and Taiwanese smartphone fabless company MediaTek have reportedly adopted TSMC’s N3E node for their first 3nm chipsets. Apple’s upcoming A18 chips for iPhone 16 models, are said to be manufactured with TSMC’s N3E node as well, according to a report by Commercial Times.

On the other hand, regarding the progress of 3nm, Samsung is still struggling with the low yield rate for its latest Exynos 2500 processors. The company targets to increase the yield rate to over 60% before the product enters mass production, according to a previous report by Korean media outlet ZDNet Korea.

Read more

• [News] Google Reportedly Shifts Orders from Samsung to TSMC for Its Upcoming Tensor G5 in the Pixel 10 Series

(Photo credit: Google)

With the United States expected to further restrict China from acquiring advanced GAA (Gate-All-Around) chip architecture capabilities, coupled with reports of poor yield rates in Samsung’s 3nm GAA generation, the semiconductor industry sources cited in a report from Commercial Times state that TSMC’s 3nm FinFET process is enjoying dominance. Reportedly, due to the high demand and limited supply capacity, upstream IC design companies are beginning to report price hikes.

Seven global tech giants, including NVIDIA, AMD, Intel, Qualcomm, MediaTek, Apple, and Google, are set to gradually adopt TSMC’s 3nm process. As per the sources cited in the report from Commercial Times, Qualcomm’s Snapdragon 8 Gen 4, built using TSMC’s N3E process, has seen a price increase of 25% compared to the previous generation, potentially triggering a subsequent trend of price hikes.

Samsung was the first to commence mass production of 3nm chips using the GAA process in June 2022. However, the first-generation N3 node, SF3E, did not achieve significant success and was initially limited to cryptocurrency applications. Subsequently, the yield rate for its own Exynos 2500 chip also fell short of expectations.

Additionally, Google’s Tensor processors, which are manufactured by Samsung, still use Samsung’s 4nm process in their fourth generation. However, it is said in the report that the fifth generation will switch to TSMC’s 3nm process.

In the second half of the year, numerous AI products will be launched in the consumer market. Among the three major players in the mobile chip market, Qualcomm’s Snapdragon 8 Gen 4, MediaTek’s Dimensity 9400, and Apple’s A18 and M4 series will all be built using TSMC’s N3 family. Moreover, Google’s Tensor G5 will also compete in the market.

It is rumored that Qualcomm’s Snapdragon 8 Gen 4 has already initiated the first wave of price increases. The industry sources cited in the report claim that the procurement cost of mobile chips was already high, with last year’s flagship 8 Gen 3 costing around USD 200. This year’s flagship chip might exceed USD 250. Whether competitors will follow suit remains to be seen.

However, industry sources cited by the report also point out that the price increase is within a reasonable range. Compared to the 5nm process, the cost per wafer for the 3nm process is about 25% higher. This increase does not yet take into account overall wafer quantities and design architecture factors.

TSMC President C.C. Wei has also revealed that TSMC products are highly power-efficient and have better yield rates. When considering the cost per chip, TSMC is the most cost-effective.

Read more

• [News] Major Clients Reportedly Fully Allocate TSMC’s Production Capacity Until 2026, 3nm Process in High Demand
• [News] NVIDIA CEO Jensen Huang Supports C.C. Wei’s Theory and Backs for TSMC to Increase Price

(Photo credit: TSMC)

According to a report from Wall Street Journal, Alphabet CFO Ruth Porat announced in a statement on May 30 that Google has committed to investing USD 2 billion in Malaysia. The investment includes building its first data center, expanding Google Cloud, and further developing artificial intelligence (AI).

Porat highlighted that this will be Google’s largest investment project in Malaysia. Google estimates that this investment will contribute over USD 3.2 billion to Malaysia’s GDP and create 26,500 jobs by 2030.

As per a report from Bloomberg, Google stated that in addition to developing cloud computing services, it will also support AI literacy programs for students and educators.

In its earnings call in April, Porat mentioned that the significant year-over-year increase in capital expenditures over recent quarters reflects Alphabet’s confidence in the potential of AI. She projected that the quarterly capital expenditures for the second to fourth quarters of this year would be comparable to or slightly higher than those in the first quarter.

On May 2, Microsoft Corp. announced that it will invest USD 2.2 billion in Malaysia over the next four years to support the country’s digital transformation. The investment projects include developing digital infrastructure, creating AI skill opportunities, establishing a National AI Excellence Center, and enhancing Malaysia’s cybersecurity capabilities.

Earlier this week, Malaysian Prime Minister Anwar Ibrahim announced the National Semiconductor Strategy, which includes providing at least USD 5.3 billion in financial support and training 60,000 semiconductor engineers, aiming to make Malaysia a global chip hub.

Amidst the U.S.-China rivalry and other geopolitical tensions, global companies are seeking to diversify their supply chains. Facing competition between the U.S. and China, Malaysia is reportedly keen to maintain a neutral position in the semiconductor supply chain landscape. According to the Malaysian Investment Development Authority (MIDA), the country currently provides 13% of global testing and packaging.

Read more

• [News] Malaysia’s Major Investment Aims to Establish Global Chip Hub
• [News] Overseas Expansion of Testing and Packaging Facilities – Japan, Malaysia, and Singapore Emerge as Top Choices

(Photo credit: Google)