On February 12th, during the morning trading session, Dongfang Gangwan, a private equity institution led by investor Dan Bin, made a public statement that has stirred excitement and debate in the financial and technological arenasThe backdrop of this statement is the ongoing discussions surrounding the latest quarterly reports of prominent billionaire private equity figures, particularly how Dan Bin has been bullish on leading tech stocks like Nvidia and is now voicing his perspectives on the implications of DeepSeek’s innovative advancements.

As a champion of active investment strategies over the past two years, Dan Bin has been identified with a strong inclination towards certain overseas tech stocks, prominently featuring Nvidia in his heavy portfolioHis substantial investment positions indicate a remarkable confidence in the future trajectory of chip technology and artificial intelligenceWith leverage tools accentuating his investments, it is a clear reflection of his high expectations from Nvidia, a company that has consistently positioned itself at the forefront of the AI and semiconductor industries.

In response to the buzz created by DeepSeek, an AI model that has gained traction online, Dan Bin raised three profound investment-related inquiries that have captivated the interest of analysts and investors alikeThe questions ponder critical implications of the advancements achieved by the DeepSeek team, particularly the team’s ability to develop globally competitive AI models despite facing limitations in computing power, and what that may signal for the future of artificial intelligence—namely, whether the need for computational resources is diminishingFurthermore, Dan Bin interrogated the significance of DeepSeek's strategies in evading traditional barriers established by CUDA—Nvidia's powerful coding architecture—and what the impacts of cost reduction and equitable access in China’s AI landscape might mean for potential future investment risks.

Contrary to what some may believe, Dan Bin asserts that the demand for global computing power is destined to increase rather than decline as a result of DeepSeek’s achievements

Advertisements

He contends that a major misconception is the idea that advancements in algorithms can replace the need for data and computational capabilities, suggesting instead a synergistic relationship between these technological pillarsIf previously, an inefficient algorithm could only support a mere ten users per chip, improved efficiencies could push that capacity to a hundred without inflating costsThus, if the value derived from a product escalates dramatically while its price remains stagnant, economics dictates that demand will inevitably soar, leading to broader adoption and integration of AI technologies.

Further elaborating on these insights, Bin posits that the trend towards making AI models more accessible and affordable has been evident over the last two years and is expected to continue into 2025. He frames DeepSeek's emergence as a critical contributor to this trend, viewing it as an essential phase in the journey towards widespread AI applicationThis, however, does not mean that the exploration of cutting-edge models and cost-effective mature models are the same endeavorThe pursuit of top-tier AI capabilities requires substantial resources and technological horsepower, a challenge not just for DeepSeek but for several tech giants vying for market leadership.

The evolution of technology generally follows a predictable "innovation-following-cost reduction" trajectoryIn practical terms, innovators invest heavily in research and experimentation, ultimately discovering effective technological solutions that pave the way for commercializationFollowing this, a wave of 'followers' emerge, seeking to replicate these solutions while optimizing costs furtherThis reciprocal relationship fosters a cycle of improvement and learning, evident in various industries—from pharmaceuticals (where innovative drugs lead to generics), to electric vehicles like Teslas versus their Chinese counterparts, and further into the realm of AI.

Dan Bin also discussed the so-called Jevons Paradox, which describes the phenomenon where improvements in resource efficiency lead to increased overall consumption rather than decrease

Advertisements

Historically applied to coal consumption after the steam engine was enhanced, the paradox illustrates that as technology makes a resource cheaper to use, it is adopted more widely, often resulting in higher overall consumptionSimilar dynamics are evident in the automotive sector, where enhanced fuel efficiency leads drivers to use their cars more, racing towards higher total fuel consumption.

Translating these insights to AI, the inference demand is poised to grow as technology improves and as AI becomes more ingrained in daily applicationsThe comparative inference costs of various models highlighted a critical challenge: with limitations in earlier economic models projecting high costs for AI applications, the recent dips in inference prices promise to expedite adoptionJust within a month's time frame, significant optimizations have slashed inference costs by an astonishing factor of one hundred, heralding the imminent arrival of a robust ecosystem of AI applicationsThis shift could dramatically alter the landscape of technology.

Regarding the enduring dominance of CUDA, Dan Bin expressed his confidence that its foundational barriers will not diminishCiting AMD’s current endeavors to convert CUDA codes, Bin pointed out the inherent performance losses and adaptation costs involved—akin to running Windows on a MacWhile technically feasible, the outcomes typically lag behind native experiences due to compatibility issuesFurthermore, he emphasized that innovative teams seeking to optimize Nvidia’s chips have not abandoned CUDA but rather improved on its capabilities by directly modifying PTX instruction sets, showcasing a deep understanding of NVIDIA’s architectural nuancesThe misconception that this signified circumvention of CUDA represents a misunderstanding of the symbiotic relationship and dependencies between the technologies.

With continuous advancements in AI technology, Dan Bin projects that a myriad of investment opportunities will emerge as the applications for AI become pervasive

Advertisements

Advertisements

Advertisements