lindongw.com

In recent discussions concerning the global artificial intelligence (AI) landscape, China's DeepSeek has emerged as a transformative force, prompting a reassessment of the value of Chinese tech assets and even the entire Chinese economic portfolioThis sentiment was echoed by a buoyant fund manager during an interview with a financial journalist, expressing their excitement over the power shift in the AI sector.

The narrative surrounding China's AI development is evolving rapidly, with a growing consensus that the nation is not merely striving to catch up to global leaders but is now aiming to surpass themThis pivotal shift in perception has sent ripples through the capital markets, causing a pronounced reaction from investorsIn the A-share market, the Wind DeepSeek index surged by an impressive 27.48% in just three trading days after the Spring Festival, capturing the attention of discerning fund managers who view this phenomenon not as a mere blip in stock prices but as a significant recalibration of the valuation framework for China's tech assets.

Previously seen as a formidable competitor with a seemingly insurmountable lead, OpenAI is now viewed through a different lensFund manager Lei Tao from Debon Funds noted that despite significant hardware constraints, their relentless efforts have borne fruit, culminating in remarkable large model productsThis revelation is contributing to a worldwide reevaluation of China's capacity for innovation in the AI domainFund managers across the board resonate with this viewpoint, underscoring that DeepSeek's disruptive power lies in its demonstration that engineering innovation can circumvent resource limitations, thus fundamentally altering the competitive landscape within the AI industry.

Cao Shiyu, manager of the GF SSE STAR Market AI ETF, articulated the revolutionary aspect of DeepSeek by highlighting its substantial reduction in training and inference costs; for instance, the training cost of its R1 model is roughly one-twentieth that of OpenAI's offerings

Advertisements

This challenges the prevailing notion that higher-performing models can only be realized through sheer computational powerThe ability to achieve high-performance models through engineering optimization not only ignites market debates about computational demand but also provides a favorable outlook for applications and edge computing.

Likewise, Wu Haijian from Western Li Asset Management emphasized that the essence of DeepSeek's breakthrough lies in its ability to minimize computational costsWhile not introducing groundbreaking innovations from scratch, DeepSeek has skillfully enhanced the efficiency of model training and inference by implementing systems such as the mixture of experts (MoE) architecture for dynamic expert selection, low-precision mixed computing, multi-token prediction, and parallel communication among experts deployed on serversThis meticulous orchestration of computational resources has significantly optimized training and inference efficiency.

Moreover, Geng Jiazou, a manager at Wanjia AI Fund, noted that the smaller training clusters and the paradigm of inference-as-training developed by DeepSeek weaken the effects of the US's restrictions on computational resources, exemplifying how engineering innovation can sculpt new industry dynamicsHe asserted, “DeepSeek may lack substantial theoretical advancements, yet the empirical enhancements in engineering have optimized the utilization of limited resources to an unprecedented extent.” This is particularly evident in the training techniques employed during the DeepSeek V3 and R1 processes, which, though familiar, showcased a clever reconceptualization of training paradigms, leading to significant reductions in the need for extensive high-precision data annotation and enhancing the model’s aptitude for tasks requiring mathematical reasoning.

The rise of DeepSeek has simultaneously sparked market volatilityObserving the unfolding events of 2024, it's evident that AI is dominating discussions, with the computational sector—home to stocks like Cambricon and new entrants—experiencing unprecedented bullishness

Advertisements

As DeepSeek emerges into the spotlight, it seemingly dismantles the myth of exorbitant computational costs, causing related sectors to react negatively.

As evidence, the international leader in computational power, NVIDIA, saw its stock plummet by 16% on January 27, accumulating a staggering $552 billion loss in its market capitalization within a weekConversely, A-share computational stocks tumbled before the Spring Festival, exemplified by a daily drop of over 10% in Cambricon's stock priceThis stark contrast to the post-holiday uplift of the DeepSeek index, which rose 27.48% in three trading days, depicts a classic seesaw effect in market dynamics.

However, following this tumult, both domestic and foreign computational stocks have begun to revitalizeLei Tao remarked that current narratives surrounding DeepSeek’s cost-cutting capabilities may be overstated. “Currently, there exists a significant gap between public sentiment and industry realityWhile short-term market narratives may pressure assets like NVIDIA linked to training computational capacities, the long-term ramifications are likely to be minimal.” He added that DeepSeek’s meteoric rise would certainly impact market sentiment in the short term, emphasizing the need for investors to prepare for potential shocks and respond adeptly to changing market dynamics.

As market attention shifts toward the applications of DeepSeek, several fund managers anticipate a greater focus on sectors benefiting from the model’s capabilitiesThey predict that DeepSeek will herald new opportunities in AI applications and commercial scenarios, fostering a more equitable landscape in the sector and accelerating the expansion of AI applications.

Looking at investment strategies, Chen Yanpeng from Nuodean Fund highlights that DeepSeek is open-source, making it accessible for a range of Chinese AI applicationsThis broad access predictably benefits various sectors, although stock selection should consider intricate factors like business models and market receptivity

Advertisements

While enthusiasm about the model is currently high, he warns of the tendency for investor interest to wane and advises a careful selection of companies demonstrating substantial operational breakthroughs, especially in fields where DeepSeek excels, such as finance, healthcare, and programming.

As this wave of AI investment rolls out, funds like Morgan Stanley express optimism about the diversity of future advancements in AI, attributing part of their optimism to DeepSeek’s performance, which could accelerate the commercialization of AI applicationsThis sentiment indicates a burgeoning demand in the inference domain, including potential benefits for inference computational chips and edge computing.

In this rapidly evolving landscape, fund managers have entered a “busy mode,” prioritizing research and adjustments in response to the surging AI marketOn the first working day after the holiday, over twenty fund managers, including Fang Yan from Fuguo Fund and Huang Wei from Ping An Fund, conducted research on Shanghai Steel UnionQuestions focused on the introduction of DeepSeek into product iterations and commercial promotion strategies.

While some managers seek opportunities through research, others quickly adjust their holdings, aligning their portfolios with the stronger application and edge sectors in light of the diminished computational plays post-holiday. “We have made adjustments in our holdings, reducing our exposure to NVIDIA chain stocks, as we previously maintained a balanced portfolio with sufficient positions in application and edge sectors,” explained a Shanghai-based fund manager heavily invested in AI.

Amidst this contextual backdrop, Wu Haijian underscores the ongoing trend towards applications and edges over pure computational advancements, advising vigilance towards the NVIDIA ecosystemHe notes that companies with significant data resources, licenses, and solid technologies are poised to gain a competitive edge.

Taking stock of the broader implications, investment firms are integrating DeepSeek into their core operations, enhancing research, sales, and risk control efforts

Advertisements

Advertisements