Tag: Jen-Hsun Huang

During a media Q&A session at CES 2026, NVIDIA CEO Jensen Huang reiterated his grand vision for the future of artificial intelligence. Addressing media questions about Moore's Law, energy efficiency, supply in the Chinese market, and next-generation architecture, Huang stated that we are currently at the beginning of a new industrial revolution, and data centers are transforming into "AI factories." A One-Year Update Rhythm: From Blackwell to Vera Rubin In response to media inquiries about how NVIDIA maintains its phenomenal growth rate, Huang provided a clear technology roadmap. He pointed out that NVIDIA is currently moving at full speed, releasing a new architecture every year. From Hopper and Blackwell to the latest Vera Rubin architecture, the performance improvement of each generation is not linear but exponential. "We are putting Moore's Law on steroids," Huang jokingly remarked. He explained that through co-design across the entire technology stack—from CPUs, GPUs, network chips to switches—NVIDIA can achieve performance leaps in a year that would otherwise take years. For example, Blackwell compared to Hopper, and the future Vera Rubin compared to Blackwell, both achieved a 10x improvement in inference cost and energy efficiency. Energy efficiency equals revenue: the new economics of AI factories. Addressing concerns about the energy consumption of AI, Jensen Huang proposed a counterintuitive yet commercially logical view: extreme energy efficiency is key to customer profitability. He analyzed that modern data centers are limited by power supply. With a fixed total power supply, if the chip's performance per watt is improved tenfold, customers can produce ten times more tokens with the same power consumption. For "AI factories" that treat AI tokens as products, this directly translates to a tenfold increase in revenue. Therefore, NVIDIA's pursuit of ultimate performance is not simply for speed, but also to reduce the cost per token, which is the economic driving force behind the widespread adoption of generative AI. Supply Chain and Geopolitics: H200's Deployment in China and Memory Demand. When asked about the Chinese market and export licenses for the H200 chip, Jensen Huang confirmed that they are working closely with the US government to ensure compliance with export control regulations while meeting market demand for computing power. He stated that despite geopolitical challenges, customer demand remains strong, and H200 chips will be shipped in accordance with regulations. ...

NVIDIA founder and CEO Jensen Huang recently warned again against the US government's policy of restricting AI chip exports to China during the GTC conference in Washington, D.C., stating that this move would be "more damaging to us" in the long run and could lead to the US losing its technological dominance in the global AI field. In a pre-conference executive discussion at the GTC conference, many topics focused on US-China trade and the heated competition in AI technology. Before his keynote address, Huang also mentioned the importance of the Chinese market, emphasizing that without a foothold in this market, US technology would lag behind in deployment and application, potentially causing even greater harm to the US in the long run. Huang made similar remarks in an interview during Comptex 2025 this year, directly criticizing the Trump administration's AI product export ban as a mistake. Core Argument: Consolidating the US "Tech Stack" and Winning Over Chinese Developers. Huang told the media that the US goal should be to have "the world built on the US Tech Stack." He believes that to achieve this goal, it is necessary to allow the sale of US-made AI chips to China in order to encourage China's vast developer community to continue using and innovating based on US technology. He emphasized, "A policy that would cause the US to lose half of the world's AI developers is not good in the long run. It would hurt us more." These remarks came ahead of an upcoming meeting between US President Trump and Chinese President Xi Jinping, with AI technology expected to be one of the topics discussed. The ban has ironically spurred China to develop its own AI chips, and Huang hopes the Trump administration will adopt a "more nuanced strategy." Currently, due to the US government's export ban based on national security concerns (fears that advanced chips could be used to enhance China's military capabilities), NVIDIA's high-end AI chips cannot be sold in the Chinese market. Huang points out that this has accelerated China's determination to develop its own chip industry, attempting to break through Washington's restrictions. The report indicates that the Trump administration tends to adopt a "more nuanced strategy" on the issue of selling AI chips to China, but also faces pressure from China hawks across the US political spectrum demanding stricter bans. Jensen Huang revealed that he had previously proposed similar views to the Biden administration team, but felt they were not adopted because "to some extent, some people thought that excluding them (China) and hurting them would be more beneficial to us, but that's not the case." He expressed his hope that the Trump administration could help guide policy direction and warned that if no action is taken, the US technology industry's dominant position in the global market may face the risk of shrinking. The GTC DC conference focused on "Made in America." The GTC conference held in Washington, D.C., not only announced that it would use NVIDIA technology to drive the AI ​​industrial revolution in the United States and accelerate the development of advanced technologies in the United States, but also put forward many statements related to "Made in America." In fact, this represents NVIDIA's strong support for the Trump administration's policies, but on the other hand, it also hopes that the Trump administration will change its attitude towards the China market so that it can regain its original market share of up to 95%.

Following NVIDIA's announcement of a $50 billion investment in Intel and the confirmation of their collaboration on developing multiple generations of data centers and PC products, speculation arose that Intel might gradually abandon its Arc graphics cards and integrated graphics design. However, Intel CEO Chen Liwu stated in a subsequent interview that the collaboration with NVIDIA would be complementary, not replacemental, and therefore the Arc brand and related GPU products would continue to exist. NVIDIA CEO Huang Renxun stated that NVIDIA will provide Intel with GPU chip modules in the future, allowing Intel to directly integrate them into its own x86 architecture CPUs, forming a new type of integrated solution. This means that some future Intel processors may use NVIDIA's RTX graphics architecture as the basis for integrated graphics computing, rather than entirely using Intel's self-developed Arc architecture graphics design. This led to speculation that Intel might stop investing in GPU R&D to reduce cost pressures, given that Intel has invested heavily in the graphics card market over the past few years but has been unable to shake the leading position of NVIDIA and AMD. In response, Chen Liwu stated that while he could not disclose specific product roadmaps at this time, this collaboration is complementary, and Intel will maintain its own GPU product line and continue to invest in related R&D resources. This marks Intel's second public statement, following a commitment from former Consumer Computing Group President Michelle Johnston Holthaus, reiterating that Arc GPUs will not disappear. However, market observers remain cautious, questioning whether Intel will have the incentive to invest heavily in catching up with NVIDIA, which is no longer a "competitor" but a partner. If Intel focuses on CPUs, AI, and manufacturing processes, the priority of GPU investment may gradually decrease. Currently, it seems Intel will at least maintain Arc GPU updates and support within the existing product cycle, continuously monitoring market response before deciding on the scale of future investment. For gamers and developers, Arc graphics cards will not disappear in the short term, but whether we will see a more ambitious next-generation architecture remains to be seen and depends on Intel's future developments.

Update: Following NVIDIA's confirmation that the Trump administration has eased restrictions on AI chip exports to China, AMD has also announced the resumption of sales of its MI308 GPUs in the Chinese market. AMD further stated that its export license application is under review and is expected to be approved soon, allowing for immediate resumption of shipments once the license is issued. Facing the restrictions of the AI ​​chip export ban, NVIDIA has finally achieved a breakthrough after continuous lobbying and pressure. CEO Jensen Huang confirmed in an interview with CCTV News in Beijing that NVIDIA has received approval from the US government to resume sales of its H20 AI accelerator chips to China, ending the export blockade that began in April. Huang stated, "I hope the H20 can ship soon. I am very happy about this very, very good news." The H20 chip was originally a version specifically downgraded for the Chinese market by NVIDIA to comply with the US Department of Commerce's export regulations. Despite this, the chip was still completely banned from export earlier this year, further hindering NVIDIA's business in the Chinese market. According to previous reports, to avoid further losses, NVIDIA's Jensen Huang not only lobbied the US government through industry organizations but also repeatedly downplayed the security risks from China, urging the US to rationally assess the impact on AI chip exports. In an earlier interview with CNN, he stated that the US did not need to be overly concerned about the Chinese military developing AI capabilities through NVIDIA chips because "any dependence on US technology would put China at risk of being restricted." This statement was interpreted as paving the way for loosening export policies and securing space for the restart of business in the Chinese market. In addition to the news of NVIDIA's return to the Chinese market during the H2020 holiday, Huang also revealed that NVIDIA is preparing to launch a newly positioned RTX Pro GPU. He emphasized that this new product is "very important" and will be designed for professional fields such as computer graphics processing, digital twins, and artificial intelligence. While more technical details and release dates for the RTX Pro GPU have not yet been revealed, judging from the name, the series is very likely to target the workstation, design, and engineering professional markets, and may even inherit the positioning of the previous Quadro family, combined with a new development path based on the RTX architecture. However, Jensen Huang did not explicitly state whether the RTX Pro GPUs would be sold in the Chinese market. Given the US government's continued cautious approach to high-end chip exports, even easing restrictions on H2O does not equate to a complete opening. Currently, the Chinese AI chip market is growing rapidly. With local competitors like Huawei and Cambricon continuously pushing for self-developed chips, NVIDIA's return to the Chinese market would undoubtedly be a significant boost to its future revenue. According to foreign media estimates, the return to H2O is expected to restore billions of dollars in potential orders for NVIDIA, further solidifying its leading position in the global AI chip field. With the RTX Pro GPUs and H2O proceeding simultaneously, whether NVIDIA can achieve success in both the professional and general AI markets will be a key focus for industry observers.

As the Nintendo Switch 2 is set to launch on June 5th in Japan, the United States, and other regions, Nintendo released a special Creator's Voice video featuring NVIDIA CEO Jensen Huang explaining how the two companies collaborated on custom processors for the Nintendo Switch 2. Huang stated that NVIDIA and Nintendo have been collaborating for over a decade, sharing the same belief that technology should serve creativity, and that "fun" is a goal worth meticulously crafting. Huang also mentioned that former Nintendo president Satoru Iwata shared his dream game console: a console with the performance to run large-scale cinematic games, yet small enough to be portable. This seemingly impossible task ultimately led to the creation of the original Nintendo Switch. While Iwata couldn't actually participate in the Nintendo Switch's launch, the custom processor initially used in the Switch required over 500 engineers and a full year of development from NVIDIA. The entire system, from chip architecture and operating system to APIs and even the game engine, was completely rethought and redesigned to ensure the console could run high-performance games while also being portable. After the Nintendo Switch sold over 150 million units, attracting numerous gamers and families worldwide, assisting many independent game developers in unleashing their creativity, and redefining the game console, Nintendo and NVIDIA have once again collaborated to launch the Nintendo Switch 2, ushering in a new chapter. To achieve even greater goals, Jensen Huang stated that the custom processor used in the Nintendo Switch 2 is an unprecedented innovation for NVIDIA, incorporating three major technological breakthroughs: • The most advanced graphics processing capabilities in mobile device history • Full support for real-time ray tracing and high dynamic range (HDR) for brighter highlights and deeper shadows • A new architecture supporting backward compatibility Furthermore, the processor used in the Nintendo Switch...

Following NVIDIA CEO Jensen Huang's criticism of Trump's AI technology export restrictions during Computex 2025, Reuters has learned that NVIDIA plans to launch a new AI accelerator chip for the Chinese market, with mass production expected as early as June this year. This AI chip will be priced lower than the previously banned H20, ranging from approximately $6500 to $8000, almost half the price of the H20. However, it will use the newer Blackwell architecture, possibly based on modifications to the current NVIDIA RTX Pro 6000D, and will use GDDR7 memory instead of the higher bandwidth HBM memory, and will not even use TSMC's CoWoS packaging design. The new US AI technology export restrictions limit GPU memory bandwidth to 1.7TB to 1.8TB per second, while the H20 was originally designed with a memory bandwidth of 4TB per second, thus failing to comply with the regulations. NVIDIA previously explained that the H20, based on the Hopper architecture, was difficult to modify to comply with US government export restrictions on new technologies. Therefore, it switched to the new Blackwell design and replaced the memory with GDDR7, which seems to better comply with the new export restrictions. However, NVIDIA stated that it is still evaluating limited options. Given NVIDIA's current market strategy, it is clear that it will not easily abandon the Chinese market. However, due to the US government's export ban, it is difficult for its products to reach the $50 billion demand for artificial intelligence applications in China. This has even led many Chinese companies to switch to NVIDIA's consumer-oriented graphics cards for AI acceleration or to use solutions from other companies. In a recent Q&A session with global media and analysts at Computex 2025, Jensen Huang stated that due to the impact of the US government's ban, NVIDIA is currently unable to continue selling H2O accelerator chips in the Chinese market. This has also caused NVIDIA's market share in AI technology applications in China to decline from 95% to around 50%. Furthermore, customers who previously used NVIDIA products have switched to solutions provided by companies such as Huawei. As a result, Chinese companies may continue to strengthen their own AI technology development capabilities, which will have a greater impact on the US government's expectation that all AI technologies originate from the United States.

In a global media and analyst interview at Computex 2025, NVIDIA CEO Jensen Huang explained the reasons for building a new office in Taiwan, citing significant connections with Taiwan's supply chain ecosystem and the need for NVIDIA to invest more R&D resources. Huang also repeatedly criticized the Trump administration's export ban on advanced AI-related technologies, deeming it a flawed decision. ▲Huang explained the reasons for building a new office in Taiwan, citing significant connections with Taiwan's supply chain ecosystem and the need for NVIDIA to invest more R&D resources. Regarding the announcement during Computex 2025 that the new Taiwan office, named "NVIDIA Constellation," would be located in the Shilin-Beitou area of ​​Taipei, Huang stated that NVIDIA's partnership with Taiwan's supply chain ecosystem is becoming increasingly close, leading to a continuous increase in R&D personnel in Taiwan. The current office space is insufficient, necessitating a move to a more spacious environment. Some believe that NVIDIA's increased investment in the Taiwan market stems from the fact that its GPU products are primarily manufactured and packaged by TSMC, and related components also rely on the Taiwanese supply chain. This has led Jensen Huang to repeatedly emphasize the deep connection between NVIDIA's success and Taiwan. The upcoming establishment of NVIDIA's second-largest office outside the US in Taiwan, coupled with the announcement of collaborations with Foxconn, TSMC, and Taiwan's National Science and Technology Council (NSTC) to build local AI infrastructure, signifies that for NVIDIA, Taiwan will no longer be just a manufacturing location, but a crucial development hub. ▲NVIDIA CEO Jensen Huang confirmed that the new Taiwan office will be located in Shilin and Beitou, criticizing the Trump administration's current decisions as flawed. Regarding the Trump administration's export restrictions on AI-related technologies, Huang has repeatedly criticized this in interviews, considering it a mistake. ▲ Jensen Huang believes the Trump administration's export ban on artificial intelligence products was a mistake. Huang stated that the Chinese market remains extremely important, especially since many artificial intelligence technologies are rapidly advancing in this market. Examples include DeepSeek, which quickly gained attention this year, and Qwen, developed by Alibaba. Even Chinese companies like Huawei, unable to easily obtain AI computing products built with advanced processes, are striving to gain more development opportunities in this wave of AI technology. Meanwhile, US companies have been severely restricted by the Trump administration's policies, preventing their AI application products from profiting in the $50 billion Chinese market. Furthermore, the ban on the sale of the H500O accelerator, originally designed specifically for the Chinese market, caused NVIDIA's market share in China, which was previously as high as 95%, to drop to around 20% in a short period. This resulted in a $55 billion loss in its financial statements and billions of dollars in other losses. Given the continued growth in demand for artificial intelligence, almost all companies are actively investing in this market. However, the Trump administration's flawed policies have restricted the development of companies within the United States. Huang Renxun urged the Trump administration to abandon its insistence that "artificial intelligence technology must be provided by the United States" and allow companies within the United States to become the providers and technology leaders of the vast majority of artificial intelligence computing infrastructure, thereby controlling market development. Otherwise, it will only allow companies within China to find ways to circumvent the ban and accelerate their artificial intelligence computing growth, while the United States' artificial intelligence technology will fall behind due to the restrictions.

At Computex 2025, NVIDIA's NVLink Fusion semi-custom AI infrastructure solution and its announced collaborations with companies including MediaTek and Qualcomm were highlights. However, Broadcom, a previous partner of NVIDIA, was notably absent from the announcements. This led to questions about Broadcom during NVIDIA CEO Jensen Huang's Q&A session with global media and analysts. ▲NVIDIA CEO Jensen Huang welcomes Broadcom to join the collaboration at any time. However, Huang did not address whether NVIDIA currently competes with Broadcom, but emphasized that NVIDIA welcomes Broadcom to join the collaboration at any time. Huang stated that NVIDIA's NVLink technology is already in its fifth generation and has accumulated considerable experience in high-speed chip-to-chip interconnects, thus possessing a deep understanding of related application technologies. Especially with the development trend of AI technology, NVLink technology can be used to connect numerous accelerated computing resources, thereby achieving a larger-scale increase in computing power. The announcement of the NVLink Fusion solution, in collaboration with companies like MediaTek and Qualcomm, connects their ASICs, CPUs, and other computing components via NVLink technology. This not only allows NVIDIA's solutions to penetrate more AI applications but also leverages NVIDIA's accelerated computing resources, which is beneficial for NVIDIA's future development. The NVLink Fusion solution also, to some extent, influences Broadcom's collaboration with companies like Intel, AMD, Google, Meta, Microsoft, Cisco, and HP on the UALink (Ultra Accelerator Link) network interconnect standard. This may explain why Broadcom was not included in the NVLink Fusion solution's partner list. In fact, considering NVIDIA's previous acquisition of Mellanox and its integration into its networking business, while NVIDIA maintains its network technology partnership with Broadcom, they are clearly also competitors in the AI ​​application server market. The addition of NVLink Fusion competing with UALink further highlights the complex relationship between NVIDIA and Broadcom—a relationship of both collaboration and competition. When discussing the NVLink Fusion solution, Jensen Huang also shared his views on the development of application-specific integrated circuits (ASICs) for ASIC applications, arguing that ASIC design development is equally important. However, he noted that up to 90% of ASIC designs ultimately fail because they are typically designed for specific computational needs and struggle to compete in today's rapidly evolving AI landscape. He cited Google's TPU as an example, an ASIC-based acceleration component primarily used for Google's services. Google's current Gemini AI model, however, relies on NVIDIA's GPUs for acceleration, offering different acceleration effects for different computational methods. Therefore, when discussing NVIDIA products, Huang emphasized that designs are built on compatible architectures and are adapted to specific needs using software like CUDA. He stated that while NVIDIA improves product performance annually, customers don't necessarily need to upgrade every year. Software adjustments allow NVIDIA's computing products to meet user requirements, and even after a certain usage cycle, replacements can quickly seamlessly transition to previous applications.

Update: AMD and Qualcomm have also signed cooperation agreements with HUMAIN. The AMD-HUMAIN partnership includes a $10 billion procurement scale and will assist HUMAIN in building a next-generation AI cloud computing platform. The Qualcomm partnership will involve building advanced AI data centers, hybrid AI across edge and cloud, and cloud-to-edge services in Saudi Arabia. Qualcomm will provide HUMAIN with advanced AI and processor solutions. HUMAIN's consideration in collaborating with both AMD and Qualcomm is to ensure that its AI technology supply is not limited to a single vendor. During a state visit with US President Trump and Saudi Crown Prince and Prime Minister Mohammed bin Salman, NVIDIA CEO Jensen Huang stated that NVIDIA will collaborate with Saudi Arabia, with HUMAIN, a subsidiary of the Saudi Public Investment Fund focusing on AI development, purchasing hundreds of thousands of new NVIDIA GPUs over the next five years to build an AI factory with a computing capacity of up to 500 megawatts within Saudi Arabia. The first phase will deploy an AI supercomputer equipped with 18,000 NVIDIA GB300 Grace Blackwell super chips and operating with NVIDIA InfiniBand networking. This collaboration will leverage sovereign AI infrastructure and expertise to propel Saudi Arabia into the global race for large-scale AI. Jensen Huang stated, "AI, like electricity and the internet, is an indispensable and vital infrastructure for every country. We are working with HUMAIN to build AI infrastructure for the people and businesses of Saudi Arabia to realize Saudi Arabia's grand vision." HUMAIN CEO Tareq Amin said, "Our collaboration with NVIDIA is a significant step towards Saudi Arabia's ambitious goal of leading in AI and advanced digital infrastructure. We are jointly building capabilities, strength, and a new global community to shape a future driven by smart technology and empowered talent." On another front, HUMAIN will deploy Saudi Arabia's first NVIDIA Omniverse Cloud, using digital twin technology to simulate and test various physical AI solutions. NVIDIA will also strengthen Saudi Arabia's local AI computing ecosystem and train thousands of developers, equipping them with the skills to use accelerated computing and AI to solve complex problems. NVIDIA and the Saudi Data & AI Authority (SDAIA) will collaborate to deploy up to 5000 Blackwell GPUs to build a sovereign AI factory and enable smart city solutions. Meanwhile, NVIDIA and the Saudi Arabian Data and AI Authority will train scientists and engineers from government agencies and universities, equipping them with the ability to develop and deploy physical and agent-based AI models. Aramco Digital, the digital and technology subsidiary of Saudi Arabian oil company Aramco, will develop AI computing infrastructure and collaborate with NVIDIA's startup ecosystem to establish an AI enterprise platform, as well as a center of engineering and robotics excellence incorporating the NVIDIA platform. NEWS: ...

At GTC 2025, NVIDIA CEO Jensen Huang emphasized that Blackwell architecture-based accelerator products are now in full production and expected to enter the market in the second half of this year. He also reiterated that computing power remains crucial in the development of artificial intelligence technology, and NVIDIA will continue to adhere to its previously proposed One Year Rhythm technology growth target. Therefore, NVIDIA not only plans to launch the next Rubin display architecture product in 2026, but also plans to advance the Feynman display architecture named after the renowned American physicist Richard Feynman in 2028. ▲At GTC 2025, NVIDIA reiterated that computing power remains the "truth" supporting the development of artificial intelligence technology. Blackwell architecture products are now in full production. Although previous accelerator products designed with the Blackwell architecture experienced delays due to design flaws, after resolving the issues in collaboration with TSMC, Jensen Huang stated that Blackwell architecture products are now in full production. NVIDIA will collaborate with numerous companies to launch various server application products, as well as with cloud providers. Blackwell architecture products will also be applied in telecommunications networks, edge computing, and even the autonomous vehicle and robotics markets, thereby accelerating the development of more artificial intelligence technology applications. ▲GPU accelerated designs are already widely used in many fields. ▲Blackwell display architecture products are now in full production and will be available for market deployment as early as the end of this year. In addition to emphasizing the full production of Blackwell architecture products, Huang also explained that even though artificial intelligence technology accelerates computational efficiency, the fundamental nature of computation still requires computing power, meaning that the underlying computing technology stacking remains necessary. Jensen Huang used the example of using a large-scale natural language processing (NLP) model to infer seating arrangements for different guests at a wedding banquet. While most current NLP models can arrive at an answer with a small number of words, the results may not meet requirements or may even be incorrect. To enable a large-scale NLP model to optimally adjust seating arrangements based on different guest relationships and needs, the inference process inevitably requires more words for deeper reasoning, significantly increasing the number of computations. Therefore, if faster processing speed and response time are desired, more computing power must be added, rather than relying solely on artificial intelligence computation. ▲Taking the seating arrangement of guests at a wedding banquet as an example, if artificial intelligence is to perform deliberate reasoning and calculation, the process must generate more than 20 times the amount of vocabulary and require more than 150 times the computing power. This stacking of computing power remains necessary and will continue to expand. In further explanation, Huang Renxun stated that the key aspects of artificial intelligence development include "cognition" and "inference." The former involves knowing and understanding the "acquired" information, while the latter analyzes and thinks to arrive at a reasonable answer. Current artificial intelligence technology, after converting user commands into vocabulary input, continuously generates more vocabulary during execution. These vocabulary words are then used as input in subsequent inference processes, and the most suitable answer is obtained through multiple iterations of inference. This process means that for artificial intelligence to arrive at a reasonable inference answer through "deliberation," its computational process must handle a much larger number of vocabulary words. To handle a large number of vocabulary words, higher computing power resources must be consumed. Furthermore, if artificial intelligence is to arrive at an answer faster, computing power must be further stacked. Even though NVIDIA has proposed NVIDIA Dynamo, an open-source inference software that can accelerate AI computing and reduce the overall cost of AI computing, in the long run, the computing power demand behind AI technology will still grow exponentially, and there may even be a much larger demand for computing power. ▲The open-source inference software NVIDIA Dynamo proposed this time mainly optimizes existing artificial intelligence computing, but the actual improvement is still limited, mainly relying on subsequent computing power stacking. ▲According to NVIDIA's approach, the operation of robots can be accelerated by different artificial intelligence operation methods to improve the smoothness of their work execution and improve the accuracy of their operation judgment. For example, the open-source Issac GR00T N1 allows robots to intuitively react to actions through an intuitive computing system, and performs more complete task inference through another deliberate system. This allows robots to improve the smoothness of their actions and make more accurate work judgments. Therefore, in his keynote speech at GTC 2025, Jensen Huang said that although the accelerator design of the Hopper display architecture has only been around for a few years, there is not much to say about it at present, given the current trend of computing power growth. The next development will be from the current Blackwell to the next Rubin, and will soon enter the next generation of Feynman display architecture. If we use the computing power performance of the Hopper display architecture as a benchmark, the Blackwell display architecture achieves approximately 68 times the computing power, while the subsequent Rubin architecture shows a growth rate of up to 900 times. Furthermore, if we use Hopper as a benchmark for cost per computation, we find that Blackwell requires only 0.13 times the cost of Hopper to achieve the same performance, and Rubin requires only 0.03 times the cost of Hopper. This means that with the same overhead, Blackwell and Rubin can drive significantly higher computing power.