According to reports from Taiwanese media, Nvidia’s GB200 mass production process has faced multiple setbacks, and the company is now focusing its development opportunities on the open architecture GB300. At the same time, Nvidia has plans for self-developed ASICs (customized chips), with Taiwan selected as the base for its R&D center to attract skilled talent from Taiwanese companies in design services. This may result in a talent drain from local Taiwanese companies.
Relevant IC design industry insiders have disclosed that a wave of talent poaching has already occurred in mid-2024, and the talent war in 2025 is expected to intensify. Major IC design companies such as MediaTek, Wistech-KY, and Creative are on high alert.
Recently, market confidence in Nvidia’s GB200 servers has continued to be challenged. According to supply chain sources, only a few sample units will be shipped in 2024, with shipments possibly increasing slightly to around 2,000 units in the first quarter of 2025. This highlights the significant challenges Nvidia faces in advancing system integration from the chip side. Industry analysts note that while these issues are expected to be resolved gradually over time, Nvidia is actively certifying more third-party supply sources. However, major CSP clients may shift their orders to the GB300, expected to be released in the third quarter, or significantly increase their own development of ASIC servers.
Tech giants are seeking alternatives beyond Nvidia’s GPUs. Apple, for instance, plans to preview the first AI version of the iPhone in July 2024, followed by a paper stating that AI models are trained on Google’s TPU (Tensor Processing Unit). Recently, Apple announced at Amazon’s AWS Reinvent conference that it will use Amazon’s customized AI chips for model training and is evaluating Amazon’s latest Trainium2 chip. These signs suggest that training solutions outside of Nvidia’s ecosystem are also effective.
Nvidia is actively responding to this by reportedly establishing its own ASIC department and expanding customized services. The company is also planning to recruit over 1,000 engineers in Taiwan for fields such as chip design, software development, and AI R&D. The semiconductor industry reveals that Nvidia continues to hire talent in Taiwan, including recruiting senior engineers in ASIC fields, with expertise in early-stage design verification, IP integration, and physical design. Many engineers are being poached by Nvidia.
Regardless of whether it’s Microsoft’s Cobalt, Maia, Google’s TPU, or AWS’s self-developed chips, Taiwan’s ASIC companies are involved in their creation. With high-quality semiconductor talent and existing training experience, they are easily attracted by Nvidia’s high salaries and readily available engineers. IC design industry insiders reveal that U.S. companies often offer RSUs (Restricted Stock Units), and Nvidia’s stock growth of 239% in 2023 and 180% in 2024 has made employees earn more from stock than from their base salaries.
1. Nvidia Confirms Entry into the ASIC Market
In June of last year, Nvidia CEO Jensen Huang revealed to global media that Nvidia’s plans to establish an AI R&D center and a second AI supercomputing center were progressing. He also directly confirmed that Nvidia would begin designing ASIC chips.
Nvidia, with its AI GPU advantage, dominates the global AI industry and has evolved from being a gaming promoter to the “AI Godfather.” This near-monopolistic position has driven many clients, including Google, Microsoft, Meta, Amazon, and other cloud service providers (CSPs), to invest in self-developed ASIC chipsets to mitigate the risks associated with over-reliance on Nvidia’s control of GPU prices and shipments. Companies such as Broadcom, Marvell, and several Chinese IC design firms are also vying for the market to seize the opportunity.
There have been rumors for some time that Nvidia would establish a new department to enter the ASIC market. In this regard, Jensen Huang confirmed this decision, saying “yes!” He believes this move will further expand Nvidia’s customer base. Although CSP clients will become competitors, all CSPs will remain Nvidia’s customers through its ASIC services.
Huang explained that Nvidia has three key advantages that cloud customers cannot do without. First, although Nvidia’s AI chips are expensive, their value is comparable to how smartphones have replaced cameras, music players, and other devices by integrating all functionalities in one device, making it very cost-effective overall. Second, Nvidia’s CUDA has a broad and rich ecosystem. Third, when clients invest in self-developed chips, this significantly increases the cost of the chips.
Taiwanese ASIC design company Alchip’s general manager, Shen Xianglin, said that Broadcom is the leader in the ASIC market, and the competition in this field is becoming increasingly difficult, with high requirements for design, capital, and overall operations.
Shen believes that Nvidia’s entry into the ASIC market is a bad idea, as it could conflict with existing products. He used MediaTek as an example, where the company’s ASIC business has performed poorly despite high customer demand, and the gross margin for ASIC business is not high. Shen stated that Alchip is not worried about competition from rivals.
Regarding Nvidia’s investment plan in Taiwan over the next five years, Jensen Huang stated that they will hire at least 1,000 engineers. They are continuing to recruit talent for a large AI R&D center and plan to establish a second AI supercomputing center, although the location has not yet been determined. Nvidia has already received a 6.7 billion NT$ AI R&D center subsidy from the Taiwanese government and has committed to jointly developing the Omniverse platform with Taiwanese companies to create industrial solutions for the global market. Nvidia also plans to build computing platforms for the Taiwan AI R&D center and provide some computing power to Taiwanese research institutions, partners, or startups for their research purposes.
2. Nvidia Fends off ASIC Challengers
Industry insiders say that as Broadcom, Google, and AWS ramp up their development in the AI chip space, Nvidia is facing increasingly fierce competition from ASIC chip developers.
The rise of ChatGPT has made Nvidia the major beneficiary of generative AI applications, generating nearly $63 billion in revenue over the past four quarters. Driven by increased capital expenditures from major cloud service providers (CSPs), Nvidia’s gross margin for Q3 2024 is approaching 75%.
Although market observers initially believed that Nvidia’s leadership in the AI sector was unassailable, with sales expected to reach nearly $100 billion next year, this optimistic outlook has recently changed. Nvidia faces challenges from the China-U.S. trade dispute and an impending antitrust investigation by China and the EU, while its growth momentum is threatened by an ASIC alliance led by Broadcom, Google, and AWS.
Semiconductor supply chain sources say that AI GPUs and ASIC processors are complementary, not substitutes. This dynamic creates both opportunities and challenges for Nvidia, which has implemented three strategic responses: securing more than half of TSMC’s production capacity, launching an ASIC R&D program, and controlling advanced packaging capabilities outside of TSMC.
Nvidia holds over 90% of the GPU market share and has dominated AI industry pricing and shipments for the past two years, making it the main beneficiary of the generative AI boom. Its market position has prompted global hardware and software companies to seek alternatives to reduce reliance on Nvidia.
Nvidia’s AI chip competitors include traditional rivals Intel and AMD, as well as major clients such as Meta Platforms, Google, AWS, and Microsoft that are developing proprietary ASIC processors.
Moreover, Apple, Broadcom, Marvell, and OpenAI are heavily investing in research and client acquisition, aiming to capture market share and possibly weaken Nvidia’s control over supply, pricing, and technological advancements.
Broadcom’s recent strong financial performance and outlook have surprised the market. Unlike Nvidia’s AI GPUs, Broadcom is confident in the shipment volume of its customized AI accelerator (XPU), expecting it to double compared to the previous quarter, thanks to orders from three major customers. The market speculates that these customers could include Meta, Google, ByteDance, and possibly OpenAI.
This development highlights new potential beneficiaries of the rising demand for AI chips, reinvigorating the ASIC sector as a formidable competitor to Nvidia.
Industry experts predict that as technology progresses and applications expand, ASIC shipments will see significant growth. Although Nvidia’s market share may decline from over 90% to between 70% and 80%, prices and margins are expected to remain favorable.
GPUs have numerous processing cores and excel at large-scale parallel computing, making them especially suitable for deep learning. Their production costs are relatively low, and development time is short, making them ideal for diverse market demands.
In contrast, ASICs are custom-designed for specific tasks, such as AI inference. They require extensive upfront research, design, validation, and production, resulting in higher initial costs and longer development cycles. Google and Amazon began developing ASICs in 2013 and 2015, respectively, with Microsoft and Meta joining in 2019 and 2020, only seeing results now.
Industry observers point out that Nvidia’s AI GPU performance continues to improve and remains unparalleled. Sources say Nvidia plans to establish a new subsidiary dedicated to ASIC development. As mentioned earlier, CEO Jensen Huang confirmed the company’s commitment to ASIC development in June during an interview.
Huang emphasized that this move would expand Nvidia’s customer base. Although CSP operators will become competitors, they will still be Nvidia’s customers because cloud operations require Nvidia’s solutions.
The company retains several competitive advantages: although its AI chips are expensive, the cost-effectiveness is similar to how smartphones replaced standalone cameras and music players. Furthermore, Nvidia’s CUDA provides a strong ecosystem, and customers face significant costs when developing their own chips.
Nvidia’s main client, Alchip, has invested nearly $500 million in the ASIC field. Alchip’s president, Johnny Shen, said that the entry barriers to the ASIC field are increasing, with strict requirements for design, capital, and operational capabilities. Nvidia’s entry into the ASIC field could conflict with its existing product lineup.
Meanwhile, MediaTek’s ASIC performance remains mediocre, with thin profit margins and enormous consumer demand, making Wistech not concerned about the competitive threat.
Related:
- Nintendo Switch 2 Teardown Reveals Key Suppliers by iFixit
- The 2025 Chip Showdown: Top Four Markets at Stake
- Key Highlights and Announcements from NVIDIA GTC 2025
- NVIDIA Unveils Silicon Photonics Switch with TSMC Tech
- AI PC Truth Revealed: What Really Makes an AI PC Smart?
- Nvidia Mobile Chip Comeback: What to Expect in 2025
- ASIC HBM to See 100% Growth in 3 Years Amid Storage Crisis
- Jensen Huang Talks RTX 5090 Price at CES 2025 Interview
- CUDA Cores Explained: How They Power GPUs
- AI Chips Evolve: Can ASICs Challenge GPU Dominance?
Disclaimer:
- This channel does not make any representations or warranties regarding the availability, accuracy, timeliness, effectiveness, or completeness of any information posted. It hereby disclaims any liability or consequences arising from the use of the information.
- This channel is non-commercial and non-profit. The re-posted content does not signify endorsement of its views or responsibility for its authenticity. It does not intend to constitute any other guidance. This channel is not liable for any inaccuracies or errors in the re-posted or published information, directly or indirectly.
- Some data, materials, text, images, etc., used in this channel are sourced from the internet, and all reposts are duly credited to their sources. If you discover any work that infringes on your intellectual property rights or personal legal interests, please contact us, and we will promptly modify or remove it.