DeepSeek Challenges AI Giants
Advertisements
DeepSeek has taken the tech world by storm with its groundbreaking V3 big model launched in December 2024, followed by the R1 model and the multimodal Janus-Pro modelThis series of releases has been labeled the "DeepSeek phenomenon," making ripples not only in the AI community but across the entire tech landscape.
Lex Fridman, a prominent podcast host known for interviewing tech entrepreneurs like Elon Musk, coined the term "DeepSeek moment" to emphasize its significance, proclaiming, "I believe five years from now, it will still be remembered as a pivotal event in the history of technology."
So, what brought DeepSeek into the spotlight? One of the key factors is its use of "smarter" algorithms that have slashed AI training costs by approximately 60% while maintaining or even surpassing the performance of existing modelsTo put it simply, where others spend $100 to train an AI model, DeepSeek manages it for just $40. This cost-effective approach directly addresses a major pain point in the industry—previously, the competition was focused on who could purchase the most expensive chips, but now it has shifted to who can make the best use of available hardware.
DeepSeek is not merely introducing a new technological route; it has also managed to penetrate the narrative stronghold of Silicon Valley and Wall Street regarding global AI discourse.
While pride and excitement proliferate, it is essential to remain grounded
Advertisements
New AI startups, including DeepSeek, still lack the arsenal of funding, technology, and talent that giants like OpenAI and Anthropic wieldNotably, despite the advances embodied by the V3 and R1 models, their algorithmic optimizations do come with limitations: their performance in complex scenarios lags behind what’s achievable by bigger models which can afford extensive resource expenditureThis scenario is akin to using a streamlined version of Photoshop for professional-grade editing—sufficient for everyday use, but revealing limitations in specialized environmentsFurthermore, it will take time before we are equipped to disrupt established giants such as Nvidia in the hardware domain.
Given the rapidly evolving nature of the "DeepSeek phenomenon," the future landscape of computational power is unlikely to follow a linear trajectory.
On one hand, products like those from DeepSeek—offering increased throughput and reduced development and consumption costs—could ignite a sudden surge in AI applications, a scenario that has been the dream of all industry players
Advertisements
However, on the flip side, as training costs plummet, more companies are likely to enter the field, leading to an exponential rise in consumer applications and a flourishing AI ecosystem, which will in turn spike the demand for chips beyond expectations.
This duality creates a paradox regarding computational power.
Industry experts have referred to a white paper released by Tencent, indicating that for AI Agent applications to achieve exponential growth, three hurdles must be overcome: a penetration rate greater than 15%, a task completion rate exceeding 80%, and a user trust level above 60%.
Taking trust as an example, prior surveys by Gartner indicated that 64% of respondents expressed reluctance to use AI in customer service scenarios.
Currently, the technological capabilities of AI Agent applications suffice only for simple tasks, such as customer service and schedule management
Advertisements
In contrast, more complex decision-making, like medical consultations or legal advice, still has considerable gaps to bridgeAlthough the largest potential for AI applications lies in education, healthcare, and finance, an AI "doctor" with a 5% misdiagnosis rate still encounters skepticismIt’s similar to the perception of autonomous vehicles; while statistically safer than human drivers, a single accident involving self-driving technology draws intense scrutinyTrust in AI remains at a nascent level and is further complicated by privacy regulations worldwide, user habits, energy constraints, technical route disagreements, multi-agent collaboration issues, and ethical dilemmas.
Many in the industry predict that a watershed moment for AI trust—where confidence surpasses 60%—is unlikely to occur before around 2026. Whether the "DeepSeek phenomenon" could accelerate this timeline is still uncertain.
Some argue that 2025 will mark the dawn of AI Agent applications
- India Cuts Rates by 25 Basis Points
- Green Transformation: A Necessity for African Industry
- How to Navigate the AI Investment Boom?
- Amazon Doubles Revenue and Profit in Q4
- Hong Kong's High-Yield Stocks Unresolved
DeepSeek's innovative approach, utilizing heterogeneous computing architectures, mixed CPU+FPGA+ASIC deployments, and dynamic load-balancing algorithms, has demonstrated a twofold increase in computational outputThis raises the question: could this exploration of technology disrupt the current computational monopoly and potentially lead to computational overcapacity?
To better understand this, we need to look at the state of computational resources, which are characterized by significant imbalancesFor one, geographical disparities existNorth America, particularly the US, dominates the global computational power landscape, with China trailing but primarily receiving its high-performance computational resources from North American sources.
There is also supplier imbalanceNvidia is the primary GPU manufacturer, with sales projected to reach around 7 million by 2025; Intel and AMD lead in CPU production; major suppliers of ASIC chips include Broadcom and Marvell, accounting for over 60% of the market; while cloud computing infrastructure is predominantly held by Google, Microsoft, and Amazon, controlling approximately 65% of the global market share
This scenario illustrates the computational monopoly that exists today.
Additionally, disparities arise in enterprise access to advanced chips; tech giants like Microsoft, Meta, Google, Amazon, and xAI hoard around 3.55 million equivalent H100 chips, without even considering upstarts like OpenAIComparatively, businesses in other economies struggle to match this access to high-end chips.
There’s also a structural mismatch in supply and demand for computational resourcesVertically, as multimodal applications become more widespread, the demand for computational power in inference has surpassed training needs, but the resources remain primarily allocated to the training side, necessitating adjustments over timeHorizontally, there is a considerable waste of computational resources on processes like data cleaning and model debugging, which are not the core focus.
An example of the shift in the landscape can be seen in the changing focus of computational power from training to inference, particularly since the latter half of 2024, where Nvidia remains the dominant player in the inference segment.
In this current market structure, regional players and suppliers seeking to challenge the dominance of established giants are more likely being overly optimistic
The prospect of breaking free from reliance on Nvidia-like monopolies will remain highly ambitious unless quantum chips can achieve large-scale commercial viability—a feat possibly 5 to 10 years away.
This begs the question: why do US AI giants feel increasingly anxious in light of DeepSeek’s achievements?
On the surface, it appears that these companies sense a crisis—but not because their technology is being surpassed; rather, it is the pathways to achieving their goals that are evolving, presenting new, viable options for emerging companiesIt mirrors the shift in competition from traditional fuel vehicles focused on engine performance to electric vehicles where battery management technology becomes essentialThe emergence of DeepSeek has proven that the Silicon Valley route of accumulating hardware and data is not the sole avenue to success; efficient utilization of existing resources is equally valid.
The launch of DeepSeek’s cost-effective R1 model coincided with announcements from major tech players like OpenAI, Oracle, and Nvidia regarding their $500 billion computational infrastructure initiative, known as the Stargate project
This connection amplifies the perceived intensity of the anxiety among American tech giants.
An experienced AI observer disclosed to a reporter that the "efficiency revolution" initiated by DeepSeek marks a significant transition in AI development from technical idealism to engineering pragmatism, validating that the marginal returns acquired through optimizing computational topology under existing hardware constraints far exceed the linear growth from merely increasing chip quantity.
This context elucidates the deeper, strategic anxieties faced by these tech giants—where engineering innovation begins to stand out in the AI competition, traditional technological advantages held by Silicon Valley will see a necessary reassessmentThe resulting shifts will also prompt a reevaluation of the overarching narrative surrounding AI, which influences associated capital and market perceptions.
Macro trend researcher and economist David Woo recently noted in an interview that, for the past two years, discussions have centered around the "exceptionalism" of the US economy, with AI serving as an essential factor in shaping this narrative
The US stock market now comprises 63% of global market capitalization, having gained an additional 10 percentage points since the introduction of ChatGPT, with the “big seven” tech companies collectively accounting for 25% of the US marketThese giants have solidified their positions through the substantial advantages offered by AI technologies, thereby indirectly bolstering the dominance of the US capital market.
Consequently, AI, US tech giants, and capital markets are inherently interconnected, with Silicon Valley and Wall Street jointly orchestrating the global AI narrative.
In the present scenario, the rise of DeepSeek—a mysterious force from the East—has ignited a challenge to this established AI narrativeThe sustained fervor within the tech and capital realms over the past several days reflects the anxieties felt by tech giants and the heightened tension within capital markets
As of February 3rd, prior to market opening, Nvidia’s stock had dropped by 20% since January 24thYet, this decline may also be attributed to short-term fluctuations in investor sentiment.
The previously mentioned seasoned AI analyst asserts that the historic significance of DeepSeek is rooted in two principal aspects: first, it engages in open-source practices, which fundamentally signify a communal sharing of human resources and technological innovations; second, it provides an alternative technological pathway beyond merely stacking computational power and dataInsights garnered from interviews over recent days suggest that this dual value proposition has resonated as a consensus within the industry.
On a practical level, DeepSeek unveils a harsh truth: as innovation delves into more profound realms, engineering capacity becomes more critical than academic breakthroughs, cost control surpasses parameter competitions in terms of importance, and societal acceptance eclipses algorithmic precision in significance.
When we move beyond geopolitical conflicts around who wins and who loses, focusing solely on the industry itself, the essence of business reveals that only profitable companies can endure
When the capital bubble eventually bursts, it may become evident that while powerful technology remains crucial, practical application and sustainability are the realities every enterprise faces today, forming the inevitable challenge for all new AI startups.
After all, not every new venture can raise $6 billion like OpenAI or xAI in the blink of an eyeIn this context, DeepSeek’s approach to technology has been emulated by numerous AI enterprises globallyIts open-source strategy has received acclaim from research institutions worldwide, while its pricing model has sparked a wave of enthusiasm among consumers globally.
The combination of efficient, low-cost technological innovations, along with immense global attention, has afforded DeepSeek a vital window of opportunity—something that even plentiful financial resources may struggle to secureThis scenario has become a focal point of admiration and envy for other AI newcomers, such as OpenAI and Anthropic.
Post Comment