The recent International Consumer Electronics Show (CES) has emerged as a focal point of attention in the tech industry, particularly with the towering presence of Nvidia, a company synonymous with cutting-edge technologyOn January 6, during the event, Nvidia's founder and CEO, Jensen Huang, took to the stage, donning a Captain America persona, to unveil the RTX 5090 GPU based on the revolutionary Blackwell architecture.

Nvidia's Graphics Processing Units (GPUs) are multifaceted hardware solutions designed to cater to a wide array of tasks and applications, showcasing remarkable adaptability and expansion capacityPredominantly recognized as a leader in the GPU sector, Nvidia's H100 GPU has, at times, been likened to a benchmark for measuring the computational power of large-scale AI modelsHuang has famously stated, "Nvidia is the engine of the AI world," which has led to an ecosystem of software and hardware that is often considered a standard in the industry.

In industry parlance, computational power is typically viewed as a blend of three performance metrics: computation, storage, and network interconnectivity

It is undeniable that computation power plays a pivotal role in the evolution of AI large modelsSince global demand for computational resources has soared, Nvidia has seized approximately 90% of the market share, leaving competitors like AMD, Intel, and various domestic chip manufacturers scrambling to keep up.

Nvidia's dominance in the market stems from its monopoly on GPU-centric computing chip technologiesIn turn, the competitive landscape has become increasingly distinct, with some companies striving to match Nvidia's GPU parameters in computing while others focus on improving storage performance.

Interestingly, a significant trend has emerged among top-tier global tech firms aimed at reducing their reliance on Nvidia's technologiesCompanies such as Microsoft, Amazon, OpenAI, and Apple are proactively developing their proprietary chips to bolster their large model training capabilities, thus enhancing their influence in the industry

• What is the Endpoint of the Competition Among AI Large Models?
• Launch of the First Free Cash Flow ETFs
• The Dollar Will Soar Even Higher
• The Fed's Liquidity Crisis Resurfaces
• $15 Billion Milestone for ETF in Three Months

This shift bears significant implications as it signifies a challenge to Nvidia's marketing narrative that computational power is paramount in AI discussions.

The previous year has seen remarkable volatility within the AI large model sector, with dramatic shifts occurring almost monthlyThe previous strategy of continuous acquisition of computational resources and data to facilitate large-scale experiments for optimizing AI performance is gradually being sidelinedAs the available data from the real world for training these massive models is nearing depletion, AI companies are starting to recognize a critical issue: the advancement of large language models is increasingly slowing.

To mitigate this problem, major AI companies are exploring new avenues, with a consensus emerging around two pivotal themes: "inference" and "efficiency." As noted by the CEO of DeepAI, the path forward lies in revisiting and refining the architecture of the models themselves, particularly to optimize the inference frameworks

A former co-founder of OpenAI has similarly asserted that future AI advancements will focus on agents, synthetic data, and computational inference timeThus, enhancing inference efficiency has emerged as a central theme for future industry evolution.

At the core of improving the inference capabilities of existing large models lies the realization that many current AI systems lack genuine “intelligence.” Presently, some AI applications generate results by stacking vast amounts of data but fall short of demonstrating true creativity and imaginative problem-solvingWhile leading-edge models are attempting to engage in prolonged thought processes that emulate human reasoning, success remains limited when it comes to practical applicationsIn this regard, Chinese-developed large models are making significant strides ahead of their global counterparts.

As we look toward 2024, the deployment of AI technologies in China is advancing rapidly

Data released by the China Internet Network Information Center indicates that the user base for generative AI products has reached 230 million, with the core AI product market nearing 600 billion yuanThe proliferation of generative AI continues to gain momentum, giving rise to innovative business models and applications.

Examining CES 2025 reveals that numerous Chinese enterprises are capitalizing on specific AI application scenarios, introducing tailored products and solutionsNotably, sectors such as AI headphones, AI glasses, and AI rings are seeing considerable advancementCompanies like Thunderbird Innovation, Shunwei Technology, Yijing Virtual, and Tianjian Cohave showcased new AR glasses and AI eyewear technologies at the eventIn the automotive realm, manufacturers like Zeekr, BMW, and BOE have unveiled intelligent cockpit systems and operating systems that harness AI large models, facilitating the development of future smart transportation solutions.

Whether through self-initiated innovations by tech giants or specialized efforts by emerging enterprises, the realm of artificial intelligence is set to experience continual leaps forward