AI chip giants NVIDIA and AMD have been under heated competition for a couple of years. NVIDIA, though controls the lion’s share of the market for AI computing solutions, had been challenged by AMD while the latter launched Instinct MI300X GPU in late 2023, claiming the product to be the fastest AI chip in the world, which beats NVIDIA’s H200 GPUs.

However, months after the launch of MI300X, an analysis by Richard’s Research Blog indicates that AMD’s MI300X’s cost is significantly higher than NVIDIA’s H200’s, while H200 outperforms MI300X by over 40% regarding inference production applications, which makes NVIDIA’s high margin justifiable.

AMD’s MI300X: More Transistors, More Memory Capacity, More Advanced Packaging…with a Higher Cost

The analysis further compares the chip specifications between the two best-selling products and explores their margins. NVIDIA’s H200 is implemented using TSMC’s N4 node with 80 billion transistors. On the other hand, AMD’s MI300X is built with 153 billion transistors, featuring TSMC’s 5nm process.

Furthermore, NVIDIA’s H200 features 141GB of HBM3e, while AMD’s MI300X is equipped with 192GB of HBM3. Regarding packaging techniques, while NVIDIA is using TSMC’s CoWoS 2.5D in the H200, AMD’s MI300X has been moved to CoWoS/SoIC 3D with a total of 20 dies/stacks, which significantly increases its complexity.

According to the analysis, under the same process, the number of transistors in the logic compute die and the total die size/total cost are roughly proportional. AMD’s MI300X, equipped with nearly twice the number of transistors compared to NVIDIA’s H200, therefore, is said to cost twice as much of the latter in this respect.

With 36% more memory capacity and much higher packaging complexity, AMD’s MI300X is said to suffer a significantly higher manufacturing cost than NVIDIA’s H200. It is also worth noting that as NVIDIA is currently the dominant HBM user in the market, the company must enjoy the advantage of lower procurement costs, the analysis suggests.

This is the price AMD has to pay for the high specifications of the MI300X, the analysis observes.

NVIDIA’s 80% Margin: High at First Glance, but Actually Justifiable

On the other hand, citing the results of MLPerf tests, the analysis notes that in practical deployment for inference production applications, the H200 outperforms the MI300X by over 40%. This means that if AMD wants to maintain a similar cost/performance ratio (which CSP customers will demand), the MI300X price must be about 30% lower than the H200. The scenario does not take other factors into consideration, including NVIDIA’s familiarity with secondary vendors, the Compute Unified Device Architecture (CUDA), as well as related software.

Therefore, the analysis further suggests that NVIDIA’s 80% gross margin, though might seem to be high at first glance, actually allows room for its competitors to survive. If NVIDIA were to price its products below a 70% margin, its rivals might struggle with negative operating profits.

In addition to achieving better product performance at a lower cost through superior hardware and software technology, NVIDIA excels at non-technical economic factors, including R&D and the scaling of expensive photomasks, which impact operational expenditures (OPEX) and cost distribution as well, while its long-term commitments to its clients, confidence, and time-to-market also play a role, the analysis notes.

Regarding the key takeaways from their latest earnings reports, NVIDIA claims the demand for Hopper remains strong, while Blackwell chips will potentially generate billions of dollars in revenue in the fourth quarter. AMD’s Instinct MI300 series, on the other hand, has emerged as a primary growth driver, as it is expected to generate more than USD 4.5 billion in sales this year.

Read more

• [News] NVIDIA’s Q3 Forecast Fails to Impress, while It Claims Blackwell to Bring Several Billion Dollars Revenue in Q4
• [News] Foxconn Confirms Shipment of NVIDIA’s GB200 to Start in Q4 as Planned

(Photo credit: NVIDIA)

Rumors have been circulating that Intel has been working with investment bankers on options to navigate the company through difficulties, which may include selling off its Field Programmable Gate Array (FPGA) unit Altera and halt its investment project in Germany, according to the report by Reuters.

Regarding the status quo of the FPGA market, a report by TechNews states that its applications have been concentrated in small-scale sectors such as communications, defense, and chip prototyping, with Xilinx and Altera dominating the field. As a result, rumors have emerged that Intel might sell its entire Altera division to another chip company looking to expand its product portfolio.

Notably, per industry sources cited in the report from TechNews, it’s further suggested that AMD could be a potential buyer, as it would help the US chip giant expand its FPGA product lineup, which would be more effectively ingrated with its current porfolio.

Altera generated USD 342 million in revenue in the first quarter of 2024, a significant decrease of 58% compared to USD 816 million in the same period last year.

On the other hand, AMD’s Embedded Solutions Division, which includes products acquired from Xilinx in 2022, reported a 46% year-over-year decline in sales to USD 846 million for the first quarter, falling short of Wall Street expectations. Both companies’ recent financial reports have been underwhelming.

In addition to AMD, Marvell, a company specialized in network IC design, has also been reported as a potential buyer for Altera.

Previously revealed in a report by Bloomberg on August 29 citing sources, Intel is said to be considering several potential strategies, including spinning off its product design and foundry businesses, canceling some of its regional facility construction plans, or pursuing mergers. These options are expected to be discussed at the board meeting scheduled for September.

Read more

• [News] Intel Reportedly Mulls to Sell FPGA Unit Altera and Freeze USD 32 Billion German Project amid Crisis
• [News] Lip-Bu Tan Resigns from Intel Board Amid Reported Disagreements

(Photo credit: AMD)

Japanese NAND Flash giant Kioxia is striving for a V-shaped recovery in its performance. According to a report from Japanese news outlet 47news, benefiting from the quick rebound in the semiconductor market, Kioxia’s revenue this fiscal year is reportedly set to reach an all-time high, with operating profit nearing a historic second-highest level.

It is reported that Kioxia’s revenue for the fiscal year 2024 (April 2024 – March 2025) is estimated to reach JPY 1.6 trillion, setting a new historical high.

This is expected to be accompanied by an operating profit of around JPY 300 billion. Kioxia’s strong performance this fiscal year is in contrast with the previous fiscal year (April 2023 – March 2024), which recorded a loss of JPY 252.7 billion, the largest in its history.

The report also suggests that Kioxia is forecasted to maintain similar strong performance in the next fiscal year 2025 (April 2025 – March 2026).

The company’s highest annual revenue record stands at JPY 1.5265 trillion for the fiscal year 2021, with a peak operating profit of JPY 456.8 billion in the fiscal year 2017.

Kioxia was formerly known as Toshiba Memory. It became an independent entity spun off from Toshiba in June 2018 and rebranded as Kioxia in October 2019.

Recently, Japanese news outlet Nikkei reported that Kioxia submitted its listing application to the Tokyo Stock Exchange on August 23, with the goal of going public as soon as October.

Reportedly, Kioxia’s valuation is expected to exceed JPY 1.5 trillion (roughly USD 10.3 billion). The deal is anticipated to surpass the JPY 420 billion raised by chip equipment maker Kokusai Electric during its 2023 IPO, which was the largest of that year.

It is also expected to exceed the projected listing of Tokyo Metro in October, estimated at JPY 640 billion to 700 billion.

Read more

• [News] Kioxia’s IPO Paves the Way for SK Hynix to Secure Funds for HBM Investment
• [News] Kioxia Filed for Listing, Reportedly Marking Japan’s Largest IPO of the Year

(Photo credit: Kioxia)

TSMC’s angstrom-level A16 process is creating a buzz even before mass production. According to a report from the Economic Daily News, not only has major client Apple already booked the capacity for TSMC’s A16, OpenAI has also joined in to secure TSMC’s A16 capacity due to its long-term need for self-developed AI chips.

Regarding this matter, TSMC stated on August 30 that the company does not comment on market rumors or on business dealings with individual customers.

Despite that TSMC’s A16 process is scheduled to enter mass production in 2026, the report has hinted that the first batch of customers has already surfaced.

In addition to Apple, which has been in continuous collaboration with TSMC, the most notable new customer of TSMC’s A16 is said to be OpenAI, the developer of ChatGPT, which is actively investing in the design and development of its own ASIC chips.

Industry sources cited by the same report reveal that OpenAI had initially been in active discussions with TSMC about establishing a dedicated fab. However, after assessing the potential benefits, the plan to build a dedicated facility was shelved.

Strategically, OpenAI is now partnering with U.S. companies like Broadcom and Marvell to develop its own ASIC chips, and potentially emerging as one of Broadcom’s top four customers.

Since both IC design giants are long-term clients of TSMC, the ASIC chips they are helping OpenAI develop are expected to be produced using TSMC’s 3nm process family and the subsequent A16 process, according to the chip design roadmap.

It is worth noting that OpenAI not only holds a critical position in the development of AI applications beyond Apple’s ecosystem but also contributes to the advancement of AI applications in Apple devices.

In June of this year, Apple unveiled its personalized intelligent system, Apple Intelligence, which has integrated ChatGPT. This strategic move has led observers to believe that OpenAI plays a key role in Apple’s AI development.

As OpenAI continues to invest in the design and development of its own ASIC chips, it is reportedly expected to maintain its influence in the AI computing field.

TSMC unveiled its angstrom-class A16 advanced process during the company’s 2024 North America Technology Symposium on April 25, set to be mass-produced in 2026.

Per TSMC, Compared to TSMC’s N2P process, the A16 offers an 8% to 10% speed increase at the same Vdd (operating voltage), a 15% to 20% reduction in power consumption at the same speed, and a density increase of up to 1.1 times, supporting data center products.

Read more

• [Insights] OpenAI CEO’s Ambition to Build Semiconductor Plant Grows Stronger
• [News] Battle of the Titans in the Angstrom Era – TSMC’s A16 Competes with Intel’s 14A and Samsung’s SF1.4

Last week, rumors have been circulating that Intel has been working with investment bankers on options to navigate the company through potholes. Latest update: options may include selling off its FPGA unit Altera and putting a halt to its USD 32 billion investment project in Germany, according to the reports by Reuters and Wccftech.

Citing sources familiar with the matter, the reports suggests that Intel CEO Pat Gelsinger and senior executives are expected to present a plan to the board later this month to revitalize the company.

Their proposal, according to the reports, includes divesting non-essential businesses and cutting down capital expenditures, such as selling off Altera and freezing the construction of the fab project in Magdeburg, Germany.

As Intel would be now unable to allocate funds from its once substantial profits to support Altera, a major producer of field-programmable gate arrays (FPGA) it acquired in 2015 for USD 16.7 billion, the sale of the company has reportedly been brought to the table. In 2023, Intel is said to be planning to spin Altera out through an IPO in three years and sell a portion of its stake through the process, but no date has been set.

Reuters indicates that Altera could also be sold entirely to another chip company interested in expanding its business portfolio, and Intel has quietly started exploring the possibility of such a sale.

On the other hand, due to delays of subsidy approvals. Intel has already been said to postpone its construction of Fab 29.1 and 29.2 in Magdeburg, Germany, as the new timeline now pushes the start of construction to May 2025. The proposal to put a halt to the project would also align with the company’s plan to reduce its capital spending by 17 percent to USD 21.5 billion in 2025, Wccftech notes.

Sources familiar with the matter said that Intel’s plan does not currently include splitting up the company or selling its foundry business to buyers like TSMC, Reuters notes.

An Intel spokesperson declined to comment Sunday, Reuters notes.

Intel is currently facing significant challenges. On August 1, the company announced financial results that fell short of Wall Street expectations and revealed plans to cut over 15% of its workforce.

Shortly after, former Intel board member Lip-Bu Tan has stepped down after just two years. Tan served as the CEO and executive chairman of electronic design automation (EDA) software company Cadence Design Systems Inc.

Read more

• [News] Lip-Bu Tan Resigns from Intel Board Amid Reported Disagreements
• [News] Intel Reportedly Considering Foundry Spin-Off or Merger to Mitigate Losses

(Photo credit: Intel)