With Microsoft, Meta, Google, Amazon, and other tech companies all...ASIC As NVIDIA begins using application-specific integrated circuits (ASICs) to create its own chips, concerns are growing that NVIDIA's future market share will be eroded. In response, NVIDIA CEO Jensen Huang stated in Taipei...InterviewAt the time, he directly pointed out that the market argument that ASICs can replace GPUs is "illogical".
To demonstrate that NVIDIA's competitive advantage remains strong, Jensen Huang further revealed that NVIDIA's R&D spending is approaching $450 billion (approximately NT$1.44 trillion).
Is ASIC a false issue? Jensen Huang: Universality is the key.
When asked by the media in Taipei about his views on the emerging trend of customized chips in the form of ASICs, Huang Jen-hsun took a very tough stance.
Currently, cloud service providers (CSPs) are actively developing ASICs optimized for specific AI models in order to reduce costs (such as Google's TPU, Microsoft's Maia, and Meta's MTIA). However, Jensen Huang believes that AI model algorithms are changing rapidly, and it takes several years to develop an ASIC. By the time the chip is made, the model architecture may have already changed.
In contrast, NVIDIA's GPUs offer high programmability and versatility, adapting to any of the latest AI algorithms. He emphasized that attempting to keep up with the rapidly changing AI wave using rigid ASICs is logically illogical.
$450 billion in R&D infrastructure
To maintain this "versatility" advantage, NVIDIA is investing heavily. Jensen Huang revealed that the company's R&D spending is about to reach $450 billion.
What does this figure mean? It's several times the annual revenue of Intel or AMD, and even exceeds the annual research and development budget of many countries. Through this saturated R&D investment, NVIDIA is able to simultaneously advance chip architecture, NVLink interconnect technology, the CUDA software ecosystem, and AI system integration, making it difficult for competitors (even those customers who invest in their own chips) to match its overall performance and development efficiency.
Analysis of viewpoints
Huang's words were clearly directed at Wall Street analysts and those "half-hearted" clients.
Recent earnings reports from Microsoft and Meta show that they are simultaneously increasing their purchases of NVIDIA GPUs and developing their own custom chips. Analysts are worried that once these ASIC-based custom chips become mainstream, NVIDIA's good days may be over.
But Huang Renxun's logic is simple: speed is the only way to win.
ASICs are advantageous for their specialization and power efficiency, but in the era of AI, where new architectures emerge every month (such as the recent DeepSeek or the new models continuously released by OpenAI), the specialization of ASICs has become a risk of stagnation. NVIDIA has invested $450 billion in R&D to ensure that the iteration speed of its GPUs is so fast that its competitors' ASICs are always chasing the taillights of the previous generation.
