• Original Source: New Voice of the Metaverse

Image Source: Generated by Wujie AI

ChatGPT has been popular so far, and NVIDIA has not only been recognized as the largest "gold digger" in the global AI era, but also the AI chip company with the highest discussion on major media and social platforms.

Recently, NVIDIA's third-quarter report for the 2024 fiscal year showed that its revenue reached a record $18.12 billion, a year-on-year increase of 206% and a quarter-on-quarter increase of 34%; net profit reached a new high of $9.243 billion, a year-on-year increase of 1259% and a quarter-on-quarter increase of 49%.

However, in the field of technology, it is never possible for one company to dominate alone. With the continuous rise of the AI boom, more and more manufacturers are also starting to make efforts in the AI chip field. Amazon Web Services, Microsoft, Huawei, Baidu, and other downstream customers are pushing for self-developed chips, while there are also small and excellent AI chip unicorns such as Cambricon and Horizon Robotics, all striving for a ticket to the AI era.

Giants Rush into the AI Chip Race

This year, several domestic companies in China have been developing their own large models. Due to concerns that the U.S. chip control may escalate, leading to a cut-off of high-performance computing chips from NVIDIA, domestic internet companies have had no choice but to compete to stockpile A800 chips, as no one wants to fall behind in the AI competition.

Insiders revealed that Baidu, ByteDance, Tencent, and Alibaba have ordered $1 billion worth of chips from NVIDIA, purchasing about 100,000 A800 chips to be delivered this year. In addition, they have also ordered $4 billion worth of chips to be delivered next year.

This has also made NVIDIA the biggest winner in the generative AI field. With the "hard-to-find" GPUs in the generative AI field, NVIDIA has made a fortune, and its market value has soared to $1 trillion. According to Bloomberg, with the influx of services such as OpenAI's ChatGPT, the generative artificial intelligence market is expected to surge from $40 billion in 2022 to $13 trillion in the next ten years. This has attracted more players to the entire industry.

It is well known that self-developed AI chips are an inevitable trend in the industry. Whenever the volume of AI computing from any manufacturer increases significantly, they need their own chips to support it in order to achieve the highest optimization.

Especially in the context of the digital economy era, given the unprecedented market prospects in the field of artificial intelligence, building a self-owned chip industry chain has become the strategic layout of technology giants at home and abroad. It is inevitable that software and cloud computing service giants such as Google, Microsoft, and Amazon will join the chip competition.

On December 1, Amazon Web Services (AWS) announced the launch of a new generation of its self-developed chip family at the 2023 re:Invent global conference, including the Amazon Graviton4 and Amazon Trainium2, which provide higher cost-effectiveness and efficiency for a wide range of workloads such as machine learning training and generative artificial intelligence applications. Compared to the current generation Graviton3 processor, Graviton4 has a performance improvement of up to 30%; Trainium2 has a training speed improvement of up to 4 times compared to the first generation Trainium chip.

On November 15, Microsoft unveiled the Maia chip at the Ignite developer conference in Seattle, designed to accelerate AI computing tasks. The Maia chip is aimed at running large language models, and its construction is different from the network connection technology used by NVIDIA. The Maia chip is connected to standard Ethernet cables.

On August 29, Google announced new artificial intelligence chips at the annual Google Cloud Next conference in San Francisco, namely the fifth-generation custom tensor processing unit (TPU) chip TPU v5e, used for large model training and inference. Compared to the previous generation chip, TPU v5e provides a 2x increase in training performance per dollar and a 2.5x increase in inference performance per dollar.

New Voice of the Metaverse believes that chips may remain at the core of AI competition for a long time to come, including among countries, giants, and startups. However, the industry is still in its early stages, and there will definitely be a reshuffle in the future.

AI Chip Industry: Can Domestic Manufacturers Overtake on the Bend?

With the intensification of global competition, China's AI chip industry is facing an important opportunity period for development.

Data shows that in 2021, the market size of China's AI chip industry reached 42.7 billion yuan, a year-on-year increase of 124%; in 2022, this market size reached 85 billion yuan, doubling again. China Post Securities predicts that by 2023, the market size of China's AI chip industry will further expand to 120.6 billion yuan.

In fact, China attaches great importance to the development of the artificial intelligence chip industry and has formulated a series of support policies, creating a favorable policy environment and promoting the rapid development of the industry.

In recent years, the country has successively introduced a number of policies to encourage the development and innovation of the AI chip industry, including the "National New Generation Artificial Intelligence Development Plan" and the "Notice on Deepening the 'Double Innovation' to Promote Mass Entrepreneurship and Innovation."

Key cities such as Beijing, Shanghai, and Shenzhen have successively issued policies to support the rapid development of the artificial intelligence industry, including content related to breakthroughs in AI chip innovation, strengthening R&D of AI chips and intelligent sensors, and accelerating the construction of intelligent computing.

NVIDIA CEO Jensen Huang once said, "China has many GPU startups, and we should not underestimate China's ability to catch up in the chip field."

Looking at the current situation of domestic chip manufacturers, there are both giant companies like Huawei with a complete industrial chain and companies like Cambricon focusing on AI computing power chips, as well as those focusing on general computing chips like Moore Threads. This also reflects the "hundred flowers blooming" in the industry, supporting more intelligent large models while welcoming industry diversity.

Not long ago, Huawei unveiled the new architecture Atlas 900 SuperCluster at the Global Connect Conference, which supports training of large models with over a trillion parameters and adopts a new intelligent exchange switch and super node architecture.

In an interview, Ren Zhengfei mentioned that Huawei's AI cluster capabilities are already not inferior to those in the United States. "Huawei's current AI cluster supports 16,000 boards, and a future super node cluster can manage tens of thousands of boards. It supports ultra-high-speed interconnection, ultra-efficient liquid cooling, and instantaneous burst power supply, achieving high system availability."

Overall, Huawei has launched the Kunpeng series for general computing and the Ascend series for AI computing. In terms of architecture, Huawei has introduced its self-developed Da Vinci architecture. In terms of software, Huawei has launched the openEuler open-source OS and supporting databases, middleware, covering the entire industry chain from hardware, architecture, framework, applications, to development and operation tools.

Cambricon's AI chip products include cloud AI chips and edge AI chips. Among them, the Cambricon 370 is Cambricon's third-generation cloud intelligent chip, the first AI chip to use Chiplet technology. Horizon Robotics is a leading domestic GPU company, with its AI chip products being GPU chips.

In addition, Moore Threads has made a comprehensive layout in the GPU field. They have now launched multiple GPU chips such as Suti and Chunxiao, with Chunxiao being their second product, integrating 22 billion transistors and built-in MUSA architecture general computing cores and tensor computing cores, supporting FP32, FP16, and INT8 precision calculations.

Currently, as high-end international AI chips face being "forced out," domestic AI chips undoubtedly become the best alternative. In recent years, under the influence of geopolitical factors, China's local AI chip industry has achieved certain development results, and some products can even be compared with similar products from international companies.

However, it is important to note that AI chip hardware performance is only one aspect, and software capability will be one of the more important barriers. Releasing computing power requires complex software and hardware coordination to turn theoretical chip computing power into effective computing power.

Facing the Gap, Ecosystem Remains a Difficult Challenge

We can see that the first batch of domestic large model manufacturers currently use NVIDIA A100 and A800 chips, not only because NVIDIA's products have stronger performance, but also because they have built a complete CUDA ecosystem.

CUDA is a parallel computing platform and programming model based on GPUs introduced by NVIDIA, which can be used to accelerate large-scale data parallel computing, enabling GPUs to be used in a wider range of scientific and engineering computing fields. The good ecosystem of CUDA has attracted the attention and use of many academic institutions and high-performance computing centers, and has also provided NVIDIA with a strong competitive advantage in the market.

An analyst told New Voice of the Metaverse, "After years of development, NVIDIA's CUDA has already attracted 4 million developers, forming a monopolistic ecological barrier, and the software ecosystem is precisely the most important product competitive factor for downstream customers." If the ecosystem is changed abruptly, it means that manufacturers will incur increased learning, trial and error, and debugging costs.

However, in the face of NVIDIA's strong position with CUDA, leading domestic AI chip manufacturers such as Cambricon, Huawei, and Horizon Robotics are also continuously building ecosystems based on their own products and solutions.

For example, Cambricon's DCU product Deep Computing series uses a compatible general "CUDA-like" environment; Huawei's Ascend series uses its self-developed Da Vinci architecture, and large model manufacturers need to make advance adjustments and optimizations for software and hardware when using the related chips; although Horizon Robotics' products include cloud and edge chips, due to the ASIC chip architecture, the cost advantage of ASIC chips for general computing is not significant.

Although there is still a considerable gap in the ecosystem development of the domestic AI chip industry compared to CUDA, we have seen manufacturers taking action. Driven by complex international trade relations and geopolitical factors, "domestic substitution" has become the main theme of the development of the domestic semiconductor industry.

Currently, chips have become one of the most promising areas in the semiconductor industry, and AI chips, as the core market driving the development of the chip industry, have immeasurable industry value. As AI chip technology gradually matures, its application scenarios are gradually penetrating into various intelligent terminal fields, occupying an increasingly important position in the development of science and technology in China.

In Conclusion

New Voice of the Metaverse has found that in the current computing power field, traditional chip manufacturers represented by NVIDIA still dominate, but technology giants such as Microsoft, Amazon Web Services, and Google are also eyeing the "fat" of computing power chips. In addition, domestic chip manufacturers such as Huawei and Horizon Robotics are constantly innovating, breaking barriers, and becoming new forces in the industry.

This also prompts the emergence of a "Three Kingdoms" situation in the entire AI chip industry. Moreover, with the advent of the AI era, the "Three Kingdoms" will "fight" more fiercely, and ultimately, will the traditional chip-making forces with the first-mover advantage continue to dominate alone, or will the technology giants come from behind to achieve a "counter-kill," or will domestic manufacturers achieve a breakthrough?

It can be foreseen that with every step forward in the AI chip industry, it will stir the nerves of the technology field. Who will win in this "Three Kingdoms" battle? The future is truly something to look forward to.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。