The AI competition opens up a new battleground, and Nvidia ushers in a rival?

With the artificial intelligence boom sparked by ChatGPT, “larger AI models are better” became an industry consensus, which boosted Microsoft (MSFT.O) A competition for tech giants such as Google (GOOGL.O), Amazon (AMZN.O), and Meta Platforms (META.O) in chip procurement, Nvidia ( NVDA.O) is the biggest beneficiary with its GPUs that excel in AI training. However, this race may be about to change, and the industry faces multiple hurdles in the pursuit of larger AI models.

Nvidia's leadership and bottleneck challenges

Nvidia's GPUs dominate AI model training due to their ability to perform efficient parallel computations. Currently, the main measure of AI capability is the number of model parameters, and more parameters mean more GPUs are needed. However, questions have arisen within the industry about the benefits of scaling up the model. Waseem Alshikh, co-founder of Writer, stated, “After exceeding the one-trillion parameter, the benefits tend to be minimal.” Microsoft CEO Satya Nadella also said at the recent Ignite conference that doubts surrounding the expansion of AI models could spur more innovation.

Despite this, AI like OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei Industry leaders remain strongly opposed to these questions, arguing that AI's potential for expansion has not reached its limit.

Data bottlenecks and the future path of AI

Thomas Wolf , chief scientific officer at Hugging Face, points out that the lack of quality training data is probably the biggest challenge facing AI development. “We exhausted the internet as a resource for training data a few months ago.” This limitation may drive the future to a smaller model based on company or personal data rather than the current large-scale model dominated by large cloud companies.

Meta's chief AI officer Yann LeCun emphasized that developing models with memory, planning, and reasoning capabilities is the only way to achieve true general artificial intelligence (AGI) The key, rather than simply relying on larger chips.

The rise of reasoning and opportunities for new contenders

AI's focus is gradually shifting from training to reasoning (the process of generating answers or results), which is bringing new dynamics to the chip market. Inference computing may not rely on Nvidia's GPUs as much as training, and AMD (AMD.O), Intel (INTC.O), Amazon custom chips, and startups may all have a share in this field. Microsoft's Eric Boyd believes that in addition to model size, technical improvements in the inference process are also critical.

Nvidia has recognized the potential of inference and mentioned in its recent earnings report that the inference business accounts for 40% of data center revenue and is growing rapidly. Its newly launched NVL72 server system showed a 30-fold increase in inference performance, demonstrating strong competitiveness in this field.

Multiple winners in a new phase

The shift in AI competition from training to reasoning means that industry opportunities will become more scattered. Although Nvidia remains a winner in the short term, AMD, Intel, and other competitors may gradually erode Nvidia's market share as the importance of reasoning increases. For investors, the focus at this stage is no longer just supporting larger model training, but preparing for a new set of winners that may emerge when using the model.