On March 7th, the U.Sstock market saw a significant upswing, particularly highlighted by Nvidia's stock price soaring over 4%, which marked a historic milestone as it crossed the $900 mark for the first timeThis surge brought Nvidia's market capitalization to an astounding $2.32 trillion, narrowing the gap with Apple to under $300 billionThis remarkable ascent underscores the potent blend of technology and necessity that defines the present-day market, particularly in the realm of artificial intelligence (AI) and semiconductors.

The progress of general models in AI technology is inciting a new wave of competition in the AI chip arenaMany industry experts attribute Nvidia's overwhelming position as a market leader to the robust ecosystem it has built around its CUDA (Compute Unified Device Architecture) platformUnlike standard GPU products, CUDA allows developers to utilize C language to devise and operate general programs on GPUs, facilitating a novel programming framework that enhances data interchange between GPUs and CPUs

This strategic innovation has effectively fortified Nvidia’s competitive advantage and established a significant barrier to entry against potential rivals.

Nevertheless, as the hardware landscape evolves, some users are increasingly opting to run their CUDA programs on competitive platforms, which introduces pressure on Nvidia’s dominanceRecently, foreign engineers encountered an unexpected modification within the licensing terms while installing CUDA version 11.6. This update explicitly prohibits using translation layers to run CUDA-based software on alternate hardware platformsReactions to this development suggest that Nvidia is attempting to thwart initiatives like Intel and AMD’s ZLUDA program or restrict Chinese GPU manufacturers from utilizing CUDA with translation layers—an apparent strategy to maintain its supremacy in the hardware ecosystem.

The evolution driven by advanced model technologies signals a new stage in the AI industry, prompting Chinese enterprises to accelerate their research and development efforts in a bid to seize a foothold in this transformative market

• How to Create High Value Products?
• Cutting Costs Without Cutting Corners
• U.S. Debt Crisis Fuels Bank Failures
• Bitcoin Plummets, Crypto Market in Turmoil
• U.S. Faces Two Major Setbacks

Presently, the Chinese chip landscape reveals two predominant trajectories: NPU (Neural Processing Unit) and GPGPU (General-Purpose Graphics Processing Unit). The NPU falls into the ASIC category, necessitating the construction of proprietary software stacks and lacking compatibility with CUDA, which demands considerable resources for adaptation.

From a broader industry perspective, the share of NPU chips utilized for general AI applications remains relatively lowNPU-equipped AI chips are typically designed in-house by cloud computing companies for their operational needsFor example, Google Cloud employs its own TPU (Tensor Processing Unit) for its requirements, while Amazon opts not to use this in favor of its own framework instead of relying on Nvidia or Microsoft's technologiesIn stark contrast, GPGPU chips are widely adopted across various cloud service providers to satisfy the computing demands of large AI models.

Nvidia's decision to curtail the use of CUDA is likely to exert a noticeable impact on certain AI chip manufacturers

Presently, leading chip companies in China may navigate this change relatively unscathedHowever, businesses that have yet to establish a comprehensive software stack may need to explore alternatives in navigating potential setbacksAn industry expert suggested that assessing whether domestic chips will face repercussions from Nvidia's restrictions hinges on evaluating the robustness of their ecosystems and the existence of proprietary function libraries"Major Chinese manufacturers have largely crossed this hurdle," he noted, highlighting companies like Huawei, Cambrian, Hygon, and Moer Thread, which have begun establishing a solid foundation of software and hardware synergy devoid of translation layer constraints, thereby fostering resilience against external market shocks.

Conversely, nascent GPGPU startups lacking a fully formed function library and computational architecture may find themselves grappling with development challenges if Nvidia enforces tightened CUDA permissions

The landscape appears particularly daunting for these emerging players, who must now accelerate their integration of viable software solutions.

The industry consensus surrounding Nvidia’s motivation for restricting access to CUDA centers on reinforcing its competitive defenses and consolidating its hegemonic status within the semiconductor supply chain to counteract mounting pressures from global hardware manufacturersFollowing this theme, AMD's CEO, Lisa Su, addressed the notable relevance of CUDA's competitive edge during a recent inquiryShe expressed reservations about such barriers, emphasizing the rapid pace of change and evolution in the AI sector"When market transformations occur at such velocity, I find it hard to believe in any lasting competitive moats," she noted, emphasizing the remarkable speed of advancements in generative AI.

Interestingly, this clause prohibiting alternative hardware use is reportedly not novel, having existed in Nvidia's end-user license agreements since 2021. Over the past three years, it has not substantially hindered relevant manufacturers, as the capabilities of domestic GPGPUs have proven their efficacy in the market