According to a report from wccftech, AMD CEO Lisa Su has indicated that she believes the AI Supercycle has just started, and the company has accelerated its product development plans to meet the substantial market demand.

In addition to NVIDIA, AMD is a significant player in the AI market as well. This is not only due to its market impact but also because the company has significantly expanded its AI product portfolio over the past few quarters, attracting attention from major clients like Microsoft and Amazon.

While AMD has not yet replicated NVIDIA’s success in the market, the company remains optimistic about the future, which is why it believes the AI boom has only just begun.

A few months ago, AMD outlined its AI chip development roadmap for the next year. The “Advancing AI” event in this October will showcase the next-generation Instinct MI325X AI chip.

The flagship Instinct MI350 AI chip is scheduled for release in 2025, followed by the Instinct MI400 AI chip in 2026. Despite AMD’s advancements, there remains a generational gap, as competitor NVIDIA is poised to launch its Blackwell architecture AI chips in the coming months.

This move is particularly noteworthy as AMD is focusing on competing with NVIDIA’s CUDA on the software front.

Read more

• [News] AMD Unifies RDNA and CDNA into UDNA Architecture, Aiming to Compete with NVIDIA’s CUDA
• [News] AMD Could Emerge as a Prospective Buyer for Intel’s Potential Sale of Altera

(Photo credit: AMD)

With AI giants in the Western world, such as OpenAI, Google and Meta, stealing the spotlight by their development of generative AI, some big names in China have introduced their AI models over the past 18 months, according to a latest report by CNBC.

Though trying to keep a low profile, five tech conglomerates in China, including Alibaba, Baidu, ByteDance, Huawei and Tencent, have launched their AI models lately, adding a new dimension to the competitive landscape of the ongoing tech rivalry between China and the U.S. Here is a brief introduction of the major Chinese AI models developed by the country’s leading tech firms, based on the information compiled by CNBC and other media reports.

Alibaba: Tongyi Qianwen

In November, 2023, a report by pandaily notes that Alibaba Cloud released its AI model, Tongyi Qianwen 1.0 a few months ago, while the 2.0 version was introduced later in the same year. Another report by South China Morning Post states that as of May, Alibaba reports that its Tongyi Qianwen AI models, often referred to as Qwen, are utilized by more than 90,000 corporate clients across China.

CNBC notes that in terms of Qwen, the company has developed various versions tailored to different functions. For instance, one model specializes in generating content or solving math problems, while another handles audio inputs and provides text-based responses.

It is worth noting that as some Qwen models are open-sourced, developers are allowed to download and use them under certain restrictions, according to CNBC.

Baidu: ERNIE

As one of China’s leading internet companies, Baidu was among the first to introduce generative AI applications in the country. A report by The Verge notes that the Ernie chatbot was available for download in August, 2023, after the approval by the Chinese government.

CNBC reports that Baidu intends to compete with OpenAI’s ChatGPT with Ernie Bot, as the company claims the bot to have 300 million users.

According to CNBC, ahead of the launch of its “Turbo” version, which took place in late June, Baidu stated that its Ernie 4.0 offers capabilities comparable to OpenAI’s GPT-4. According to Baidu, this foundational model has advanced understanding and reasoning abilities.

Similar to other companies, Baidu is offering access to its AI model through its cloud computing services, CNBC says.

ByteDance: Doubao

TikTok parent company ByteDance, though entered the AI race later than competitors like Baidu and Alibaba, has surprised the market with its low-cost Doubao model, which was launched in May, 2024.

According to a report by technode, the model can process 2 million Chinese characters, equivalent to 1.25 million tokens, for just RMB 1 (USD 0.14). In comparison, OpenAI’s latest multimodal model, GPT-4o, costs USD 5 per million input tokens.

CNBC notes that Doubao has various capabilities, including voice generation and coding support for developers.

Huawei: Pangu

Introduced by Huawei in 2021 as the world’s largest pre-trained Chinese large language models (LLMs) with over 100 billion parameters, the Pangu models are now entering their fourth iteration, according to Counterpoint. In May, 2024, the latest Pangu models are said to boast 230 billion parameters.

Interesting enough, Huawei has adopted a different strategy from its competitors with its Pangu AI models, CNBC remarks. The tech giant focuses on developing industry-specific models tailored to sectors like government, finance, manufacturing, mining, and meteorology.

For instance, Huawei claims that its Pangu Meteorology Model can predict a typhoon’s trajectory 10 days in advance in just 10 seconds, a task that previously took four to five hours, according to CNBC.

Tencent: Hunyuan

Last year, Tencent introduced its foundational model, Hunyuan, which is accessible through Tencent’s cloud computing services.

According to CNBC, Tencent has highlighted Hunyuan’s strong Chinese language processing abilities and advanced logical reasoning, supporting features like image generation and text recognition. The model is designed for use across industries such as gaming, social media, and e-commerce.

As the operator of China’s largest messaging app, WeChat, Tencent launched an AI chatbot this year based on the Hunyuan model. The AI assistant, named Yuanbao, can access information and content from WeChat, setting it apart from competitors, CNBC notes.

Notably, China’s large language models, just like its rivals in the West, rely on the strong computing power of AI chips. A previous report by Reuters in November, 2023, states that Tencent is said to have stockpiled a substantial reserve of AI chips from NVIDIA, as the company prepares in advance to train its Hunyuan AI models for the following generations.

How far will the tech giants in China be able to push the boundaries of AI models? The answer may lie in the development of the country’s domestic chips, as the U.S. authority already banned the export to China of AI chips.

Read more

• [News] China’s Core AI Industry Achieved a Market Size of Nearly CNY 600 Billion

(Photo credit: Baidu)

In recent years, data center has been one of the key areas for GaN (Gallium Nitride) manufacturers to tap power electronics market, and GaN applications in the data center power supply market have taken a big step forward. Notably, the rise of AI technology has further fueled this market at present.

In the AI ecosystem, data centers have enormous demands for high-speed computing and power. According to a TrendForce report, NVIDIA’s Blackwell platform will be officially launched in 2025, replacing the existing Hopper platform, and will become NVIDIA’s primary solution for high-end GPU, accounting for nearly 83% of all high-end products.

For high-performance AI servers like the B200 and GB200, a single GPU can consume more than 1,000W of power.

Facing soaring power demands, the power specifications for each data center rack will increase from 30-40kW to 100kW, posing significant challenges for data center power systems. The combination of GaN and liquid cooling technologies will be critical to improving energy efficiency in AI data centers.

The hike in chip power consumption requires servers to achieve higher power density and efficiency.

GaN, which reduces energy losses and increases power density, is now seen as one of the key technologies for optimizing energy efficiency in AI data centers, which has attracted many players, including Infineon, Texas Instruments (TI), Navitas, Innoscience, Transphorm, CorEnergy, Danxi Tech, and GaNext, to join the race.

Among them, both Navitas and Infineon have unveiled their AI data center power roadmaps.

Infineon’s AI Data Center Power Roadmap (Source: Infineon)

Combining the unique advantages of Si (Silicon), SiC (Silicon Carbide), and GaN (Gallium Nitride), Infineon has launched a 3 kW PSU and a 3.3 kW PSU, with an 8 kW PSU expected to be available in the first quarter of 2025.

The new 8 kW PSU will support AI racks with outputs of up to 300 kW or more. Compared to the 32 W/in³ density of the 3 kW PSU, its efficiency and power density will increase to 100 W/in³, further reducing system size and lowering operator costs.

In terms of GaN technology, Infineon’s CoolGaN™ solution can provide over 99% system efficiency in PFC topologies. Moreover, GaN Systems, acquired by Infineon, already released a 3.2kW AI server power supply as early as 2022 and unveiled its fourth-generation GaN platform in 2023.

The new platform achieves efficiency exceeding the Titanium level, with power density increased from 100W/in³ to 120W/in³. Thereby, the industry highly expects the synergistic effect created by the combination of these two companies.

Navitas introduced its GaNSafe™ and Gen-3 Fast SiC technology last year, along with a 4.5kW CRPS design, achieving more than double the power density of traditional silicon solutions. In July this year, Navitas unveiled its CRPS185 4.5kW AI data center server power solution, with a power density of 137W/in³ and over 97% efficiency.

Navitas AI Data Center Power Roadmap (Source: Navitas)

Navitas revealed that over 60 customer projects involving 3.2kW and 4.5kW power solutions for data centers are currently under development.

These projects are expected to bring millions of dollars in revenue growth for Navitas’ GaN and SiC business between 2024 and 2025. It aims to begin small-scale production of AI data center power solutions in 2024.

Aside from Infineon and Navitas, other manufacturers like TI, EPC, CorEnergy, GaNext, and Innoscience also set sights on this market.

TI reached an agreement with Delta, the world’s largest server power supply provider (with nearly 50% market share), as early as 2021. Based on GaN technology and TI’s C2000™ MCU real-time control solution, they are developing high-efficiency, high-power server PSU for data centers.

Thus, their joint efforts and future fruit in the AI server power market are highly anticipated.

Beyond AI data center server, humanoid robot industry that enjoys burgeoning growth this year, also injects new vitality into the GaN market.

Humanoid robot is assembled by sensing, control, motor, and battery systems, in which GaN can has its place in LiDAR system, motor drive, DC-DC converter, and battery BMS, among which motor drive plays a critical role.

According to TrendForce, the demand for motor driver in humanoid robot has skyrocketed due to the mounting demands for degrees of freedom.

To achieve higher power output, high-power-density, high-efficiency, and fast-response motor driver are in demand, and GaN is a perfect fit for it, which also has the ability to strengthen overall robot performance in terms of heat management, compact design, and overall system design.

It is reported that Siemens, Yaskawa Electric, and Elmo have already integrated GaN technology into their robotic motors, and the GaN industry chain is gearing up for seizing more opportunities.

Currently, companies such as TI, EPC, Transphorm, Innoscience, Navitas, and CorEnergy are actively promoting GaN adoption in motor drive market. Among them, Transphorm has supplied GaN FET products for Yaskawa Electric’s new servo motor.

TrendForce points out that future robots will exceed our imagination, with precise, fast, and powerful movement capabilities as the key parts, which will inevitably push the motors required to drive these movements advance forward, and this is regarded as a boon for GaN technology.

Read more

• [News] Global GaN Power Device Market Size Expected to Reach USD 4.376 Billion in 2030, CAGR of 49%
• [News] Foundry Giant GlobalFoundries Acquired GaN-Related IP

(Photo credit: Infineon)

Amid concerns on the delay of NVIDIA’s Blackwell, CEO Jensen Huang spoke at the Goldman Sachs Communacopia and Technology Conference a few hours ago, trying to ease the doubts of the market by expressing his optimism on the demand of its products and the company’s future prospects.

“We have a lot of people on our shoulders, and everybody is counting on us,” said Huang, according to a report by Yahoo Finance. He even joked that as the world relies so much on the company, the engineers may have to spend more time on work. “Less sleep is fine. Three solid hours is all we need.”

Huang also elaborated on the demand for the delivery of NVIDIA’s components, technology, infrastructure and software, stating that it is so overwhelming that people may get “emotional,” as it has a direct impact on their revenue and competitiveness.

It is worth noting that Huang also mentioned that NVIDIA heavily relies on TSMC for producing its most important chips, as in many ways, according to a report by Economic Daily News. He said TSMC’s agility and ability to respond to demand are incredible.

Huang stated that most of NVIDIA’s technology is self-developed, and if necessary, orders could be shifted to other suppliers. However, such adjustments could lead to a decline in chip quality, according to the report.

According to a previous report from Commercial Times, NVIDIA has reportedly executed changes to the Blackwell series’ GPU mask. Therefore, the process can now proceed without re-taping out, with NVIDIA’s updated version of B200 expected to be completed by late October, allowing the GB200 to enter mass production in December.

Moreover, in his latest meeting with Goldman Sachs, Huang noted that the first trillion dollars of data centers is going to get accelerated, creating a new type of software, generative AI.

Citing Huang’s remarks, the report by Yahoo Finance stated that it matters a lot because generative AI is not just a tool but a “skill,” so for the first time, the AI chip giant is developing skills that will enhance human capabilities.

According to Yahoo Finance, Huang said that NVIDIA, along with cloud service providers (CSPs), build the infrastructure in the cloud so developers can access these machines to train, fine-tune, and safeguard models.

It is worth noting that Huang tried to materialize the benefit, saying that for every dollar a CSP spends with NVIDIA, it results in USD 5 worth of rentals. He also said while training AI models is resource-intensive, it pays off in the long run.

Citing Huang, the report stated that NVIDIA’s servers may seem expensive at first glance, as it potentially costs a couple of million dollars per rack. However, they replace thousands of nodes. What is remarkable is that the cost of cables for old, general-purpose computing systems is higher than consolidating everything into a single, dense rack, Huang said.

According to Yahoo Finance, Huang also noted that the days of software engineers writing every line of code are completely behind. In his vision, every software engineer will have digital companions working alongside them 24/7.

In addition, NVIDIA, with its 32,000 employees, hopes to be supported by “100 times more digital engineers” in the near future, the report noted.

Notably, there seems to be another piece of good news for the U.S. chip giant. According to a report by Reuters, the U.S. government is said to be mulling, allowing NVIDIA to export advanced chips to Saudi Arabia, which would enable the country to train and operate the most powerful AI models.

According to the report, Saudi Arabia expects to receive shipments of NVIDIA’s most advanced chips, the H200s, which were first used in OpenAI’s GPT-4o.

Read more

• [News] NVIDIA’s Blackwell Overcomes Delays, as GB200 Reportedly Sets for December Mass Production
• [News] Foxconn Remains Unfazed Despite NVIDIA’s GB200 Delay, with Customers Aim to Secure Their Orders

(Photo credit: NVIDIA)

According to a report from Commercial Times citing sources, it’s revealed that NVIDIA has executed changes to the Blackwell series’ 6-layer GPU mask. Therefore, the process can now proceed without re-taping out, as production delays being minimized.

The report noted that NVIDIA’s updated version of B200 is expected to be completed by late October, allowing the GB200 to enter mass production in December, with large-scale deliveries to ODMs expected in the first quarter of next year.

Previously, as per a report from The Information, NVIDIA’s GB200 was said to be experiencing a one-quarter delay in mass shipments. Another report from the Economic Daily News further suggested that the problem likely lies in the yield rates of advanced packaging, which mainly affected the non-reference-designed GB200 chips.

Industry sources cited by Commercial Times addressed that NVIDIA’s Blackwell chip used to be facing instability in metal layers during the HV process, which was then resolved by July.

In addition, since the issue reportedly occurred in the back-end-of-line process, a new tape-out was deemed unnecessary. Still, as CoWoS-L capacity remains a bottleneck, the advanced packaging for GB200 this year is expected to adopt  CoWoS-S.

Read more

• [News] Foxconn Confirms Shipment of NVIDIA’s GB200 to Start in Q4 as Planned
• [News] Foxconn Remains Unfazed Despite NVIDIA’s GB200 Delay, with Customers Aim to Secure Their Orders

(Photo credit: NVIDIA)