In a significant turn of events in the landscape of artificial intelligence, the Chinese startup DeepSeek has emerged as a disruptive force, challenging traditional models that have long dominated the sectorNot only has DeepSeek's open-source strategy lowered the barrier to entry for AI technology, encouraging a broader participation in AI applications, but it has also amplified its influence globally at an unprecedented rate.

This transformation, however, comes with astonishment. “Wait, waitThis step might be wrong.” This was a pivotal moment that encapsulated AI mimicking human-like reasoning, where the technology paused to reassess its previous conclusions, reminiscent of how humans tackle problems with self-reflection.

DeepSeek’s team made this notable discovery while training their DeepSeek-R1-Zero model, coining it the "Aha Moment." This phenomenon wasn’t merely a singular moment of realization for the model but also for the researchers observing its behavior.

In their research paper, the DeepSeek team highlighted how this "Aha Moment" served as a powerful reminder that reinforcement learning can unlock new levels of intelligence within AI systems, shaping the future of more autonomous and adaptive models.

At the start of January, DeepSeek made waves worldwide by launching their app on January 11. Only nine days later, on January 20, they unveiled a state-of-the-art open-source model named R1, achieving performance levels comparable to top models from OpenAI—but at a fraction of the training cost.

Remarkably, by January 26, the DeepSeek app had surged to the top of the download charts on Apple's App Store, a feat achieved astonishingly fast, and maintained this dominant position across mobile app download rankings in approximately 140 countries

Advertisements

This meteoric rise demonstrated the powerful appeal and utility of DeepSeek's offerings.

On January 31, DeepSeek was officially integrated into the platforms of major American tech giants: Nvidia, Amazon, and Microsoft, marking a validation of its technology on a global scale.

The phenomenon triggered by DeepSeek signifies a transformative “Aha Moment” in the global AI arena, altering the mindset that building large AI models requires exorbitant financial backing or immense computational power.

The concept of low-cost Chinese models is shaking the foundations of American technological supremacy, as highlighted in a recent cover story of The Economist, which argued that DeepSeek challenges the prevailing notion within the AI industry that powerful hardware is crucial for advancement.

The insights extend beyond technology to finance, as Deutsche Bank expressed that DeepSeek’s emergence may lead to reevaluations of the inflated valuations of American tech companies.

As more “Aha Moments” occur, the rules of engagement within the AI industry seem to be undergoing a radical transformation.

“Small Efforts Yield Miraculous Results”

The disruption brought by DeepSeek stems not from a groundbreaking technological advancement but rather from its adept optimization of algorithms and model compression techniques, drastically lowering the costs associated with training and inference.

Prior to DeepSeek's emergence, the strategy favored by major tech players was one of substantial investment: build larger models, use more extensive datasets, and deploy massive computing power.

For instance, the cost of training OpenAI’s ChatGPT-4 reached upwards of $78 million, possibly even crossing the $100 million mark

Advertisements

In stark contrast, DeepSeek’s R1 model was trained for less than $6 million, approximately 5% to 10% of the cost associated with similar performance models.

This radical reduction in costs signifies a breach in the “moats” that American tech giants had meticulously constructed around their operations, indicating that the realm of large model training is no longer their exclusive domain.

Deutsche Bank noted that DeepSeek shattered a previously held belief within the AI industry: if one wished to compete in the AI arena, substantial pre-investment in building data centers was mandatoryUnlike traditional industries where first-mover advantage is absolute, the AI sector operates on different dynamics.

In the past, Europe believed it stood no chance in the AI race due to its inability to raise substantial funds comparable to American tech entities for formidable computational resourcesHowever, the successful implementation of DeepSeek’s model signifies that even with limited financial and computational capacity, remarkable AI models can be developed—this insight opens doors for European researchers.

An Open Strategy with Intent

If DeepSeek's low-cost model has managed to breach the capital-based defensive walls of American tech giants, their open-source strategy appears to be an assault on Silicon Valley’s technical stronghold.

Differing from the proprietary approaches of American firms such as OpenAI and Anthropic, DeepSeek opted to open-source its models and released thorough research papers and technical documentation for global developers and researchers to leverage.

This approach not only democratizes the use of AI technology, facilitating a shift from oligopolistic practices to broader public engagement, but also allows DeepSeek's influence to proliferate across the globe.

Jim Fan, a senior research scientist at Nvidia, remarked, “We are living in a unique era: a non-American company is genuinely embodying the original mission of OpenAI—to conduct truly open frontier research that empowers everyone.”

On February 1, OpenAI's CEO, Sam Altman, acknowledged for the first time that the company's proprietary approach may be “historically misguided,” asserting the need for a different open-source strategy amidst complexities and challenges being faced.

The significance of DeepSeek’s open-source movement transcends merely dismantling the technical barriers around AI technology.

Recently, several countries, including the United States, Italy, Australia, and South Korea, have been formulating policies to restrict or even ban DeepSeek's technologies.

However, a column in Bloomberg declared that DeepSeek's open-source model complicates the U.S. government's ability to stifle its influence, as its technologies and codes have already entered the global AI community. “It has become almost impossible for Washington to eliminate DeepSeek's impact,” the article stated.

Silicon Valley's Response

In the wake of DeepSeek's aggressive advances, tech giants in Silicon Valley have wasted no time in responding.

On February 1, OpenAI rolled out o3-mini, its first inference model made available for free to users.

On February 5, Google introduced an array of updates to their Gemini AI model; shortly after, OpenAI announced ChatGPT Search would be available to all without requiring registration.

Chris Lehane, OpenAI’s Chief Global Affairs Officer, acknowledged the seriousness of the competition, stating, “The news surrounding DeepSeek signals that this is a very real contest

Advertisements

The outcomes of this competition will genuinely shape the future of our world.”

During the current earnings season for U.S. stocks, DeepSeek has emerged as a frequently mentioned topic during earnings calls for various tech companies.

Rene Haas, CEO of chip design giant ARM, praised DeepSeek for being “very beneficial to the industry,” stating that it enhances efficiency and reduces costs while expanding overall computational demand.

Lisa Su, CEO of AMD, highlighted that DeepSeek is spurring innovations “favorable to AI applications,” noting the value of developing new methods for training and inference that demand fewer resources.

Sundar Pichai, CEO of Google, explicitly stated that DeepSeek has achieved exceptional results.

Qualcomm CEO Cristiano Amon asserted the launch of DeepSeek-R1 represents a pivotal moment for the AI sector, forecasting that future models will be smaller, more efficient, and tailored to vertical applications, propelling accelerated growth in AI penetration rates.

Mark Zuckerberg, CEO of Meta, acknowledged that DeepSeek has executed “many innovative approaches” for rapidly and inexpensively training its models, which Meta is still in the process of "digesting" for potential application within its own systems.

Shyam Sankar, CTO of Palantir Technologies, noted that DeepSeek has demonstrated the “commoditization” of AI models and reminded everyone of the escalating AI arms race with China, urging collective mobilization in this competitive landscape.

Reevaluating Chinese Assets

The ripple effects unleashed by DeepSeek have reverberated throughout Silicon Valley and into Wall Street

External investments are beginning to reassess the value of Chinese assets, propelling a significant shift in perception.

Deutsche Bank, in its latest research report, likened the uproar from DeepSeek to the shockwaves caused by the Soviet Union's launch of the Sputnik satellite in 1957, predicting that 2025 could mark a watershed year for the global investment community to reevaluate China's capabilities.

Goldman Sachs highlighted that DeepSeek’s rise implies a crucial pivot in the AI industry from hardware fundamentals to software applications, presenting new opportunities for diverse growth within global markets, especially concerning the long-term revaluation of Chinese tech stocks.

Notably, Goldman pointed out that DeepSeek’s R1 model, with its cost-effective capabilities analogous to leading models like GPT-4 and Llama, offers promising growth prospects and productivity improvements, consequently aiding in bridging the 66% discrepancy in valuations between U.S. and Chinese tech stocks.

Goldman Sachs maintained an overweight rating on the MSCI China Index, projecting an increase to 75 points by the end of the year from a base of 66 points, with optimistic scenarios suggesting a rise of up to 28%.

Following Goldman, Deutsche Bank also released a report asserting that the popularity and expanding application of DeepSeek's low-cost R1 model abroad signify a critical acknowledgment of Chinese intellectual property, showcasing a breakthrough in high value-added sectors and underscoring China's rapid expansion within the global supply chain.

Deutsche Bank emphasized that 2025 will be a pivotal year for investors to recognize China's leadership in the global competitive landscape, noting that Chinese companies are delivering high-quality products at competitive prices across various manufacturing and service domains, likely exceeding profitability expectations fueled by supportive policies on consumption and financial liberalization.

The Deutsche Bank report explicitly stated, “The launch of DeepSeek has fundamentally shaken the prevalent perspectives advocating for containing China.”

Advertisements

Advertisements