DeepSeek's R1 Shakes Markets: How Far Could This Go?

-

On January 27th, the financial world was taken by surprise as Chinese tech startup DeepSeek unveiled its R1 model, an AI system touted as low-cost yet highly capable, causing a dramatic ripple across markets.

The Rise of DeepSeek

DeepSeek, founded by Liang Wenfeng—a hedge fund manager who pivoted in 2021 to invest his financial gains into Nvidia chips and recruit China’s top AI talent—has positioned itself as a significant contender in the global AI arena. The company's mission: to rival American tech giants’ AI systems with more affordable and efficient alternatives. 

(Photo: Chinatalk Media | Sohu)

After the news, the S&P 500 plummeted over 2%, with tech stocks bearing the brunt of the selloff (-6%), fueled by fears of an impending price war in the AI sector. Analysts have already dubbed DeepSeek "the Temu of ChatGPT," suggesting its products could rival Silicon Valley giants' offerings at a fraction of the cost.

Indeed, DeepSeek’s R1 surpasses ChatGPT in mathematical computations and reasoning abilities. It has quickly become the most downloaded app in both China and the U.S. But there’s a catch: while OpenAI’s ChatGPT cost an estimated $5 billion annually to develop over years of research, rumors suggest DeepSeek achieved its results with a mere $6 million and within just two months, leveraging pre-existing models.

Nvidia’s Fallout

This new narrative disrupts the belief that Big Tech needs to spend tens of billions of dollars on AI development, casting a shadow on Nvidia—the primary beneficiary of such investments. Nvidia’s shares tumbled 15% yesterday, with broader ramifications across the AI supply chain:

  • ASML (-7%)

  • Siemens Energy (-20%)

  • Schneider Electric (-9%)

  • Data Center REITs: Digital Realty (-11%) and Equinix (-7%)

  • Utilities: Vistra (-27%)

Misinterpretation or Market Overreaction?

Despite yesterday’s market turmoil, a closer look reveals a more nuanced story. DeepSeek has two AI models: the older V3 and the headline-grabbing R1.

  • V3 Model: Known costs of $6 million and a two-month development timeline using Nvidia’s H800 chips—a lower-capacity version designed to comply with U.S. export restrictions.

  • R1 Model: Performance claims comparable to ChatGPT-4, but development costs and methods remain undisclosed. Speculation suggests billions of dollars and cutting-edge Nvidia chips, possibly smuggled into China via Singapore to evade restrictions, played a role. Nvidia’s advanced Hopper and Blackwell chips might have been key to R1’s development, which aligns with Nvidia’s explosive revenue growth in Singapore—a fourfold increase year-over-year. 

The "Mix-of-Experts" Advantage

DeepSeek’s V3 leverages a “Mix-of-Experts” (MoE) architecture, significantly reducing computational needs compared to traditional dense Large Language Models (LLMs) used by Big Tech. This approach selectively activates only parts of the model for specific tasks, slashing computational and training costs.

However, MoE models face reliability challenges. Routing—deciding which parts of the model to activate—is complex, limiting MoE’s effectiveness in generalist applications. This explains why tech giants have yet to fully embrace MoE for scalable, versatile AI systems. 

Market Implications

Yesterday’s reaction may have been overblown. While DeepSeek’s competitive entry is undeniable, the broader impact on Big Tech remains uncertain. To quantify potential losses, consider that Nvidia, Microsoft, Alphabet, Amazon, and Meta have collectively contributed 700 points to the S&P 500’s 2,168-point rise over the past two years. Assuming a credible competitive threat, a 4-5% correction (200-300 points) seems more plausible than the panic-driven selloff we witnessed.

The Bigger Picture

What happened underscore that we are still in AI’s infancy. Victory is far from guaranteed for Big Tech. A historical parallel can be drawn to the internet era: telecommunications companies that built the infrastructure became commoditized, while e-commerce, cloud, and SaaS companies reaped the rewards.

The key question now is: who will truly benefit from AI infrastructure in the long run? Will it still be Big Tech, or will new players like DeepSeek emerge as dominant forces?

Time will tell. For now, today’s upheaval reminds us that disruption often comes from unexpected corners, and markets must recalibrate to the evolving landscape of AI.

Comments

Post a Comment

Popular posts from this blog

RUMORS OF LESS SEVERE TARIFFS SEND MARKETS ON A ROLLERCOASTER

-
Image
Yesterday, markets experienced heightened volatility following rumors suggesting a softer stance from the Trump Administration on its tariff policies. The speculation was quickly dismissed by President Trump himself, who labeled the reports as "fake news" on his social platform, Truth . However, the initial rumors created waves across financial markets and asset classes.  According to The Washington Post , the Administration might be considering scaling back its campaign promises, applying tariffs only on imports deemed critical to U.S. national and economic security.  This would mark a departure from pre-election threats of imposing a 60% tariff on Chinese imports and 10%-20% tariffs on goods from other trading partners. Key sectors likely to be impacted, as per the Post , include: Defense: Tariffs on steel, iron, aluminum, and copper Medical Devices: Items like syringes, vials, and needles Renewable Energy: Batteries, solar panels, and rare earth elements The rum...
Read more

Quantum Computing: A Reality Check or the Beginning of a Bust?

-
Image
Annabelle Chih / Bloomberg / Contributor / Getty Images Yesterday, quantum computing companies faced a sudden and dramatic crash, losing nearly half their value and wiping out approximately $20 billion in market capitalization. The catalyst for this sell-off was Jensen Huang, CEO of Nvidia, whose remarks at CES in Las Vegas highlighted how far we still are from the practical application of quantum computing technology.  Huang stated, “If you were to say 15 years, that might be a little too soon. If you said 30, that might be a little too late. But if you picked 20, I think a good number of us would agree with that.” He also emphasized that quantum computers are not a one-size-fits-all solution. While they have immense computational potential, their ability to process data is still highly limited given current technological constraints. These comments sent shockwaves through the quantum computing sector. Companies like Rigetti Computing (-45%), Quantum Computing Inc. (-43%), IonQ...
Read more

The Trump 2.0 Era: What Kind of Government Can We Expect, and How Will It Impact Markets?

-
Image
Donald Trump has officially become the 47th President of the United States. The new administration is expected to act swiftly, leveraging executive orders (comparable to legislative decrees in other systems), which take immediate effect and bypass the traditional legislative process, including Congressional approval. "I will act with historic speed and strength and fix every single crisis facing our country," declared the newly elected President.  Evan Vucci/AP Key Priorities of the Trump Administration Immigration Reduction: Expected to lead to immediate wage pressures as companies struggle to find lower-cost labor, potentially driving inflation higher. A similar market dynamic was observed in the UK post-Brexit, where reduced immigration contributed to "stickier" inflation, with average inflation remaining higher compared to the EU since the start of the war. This forced the Bank of England to implement more aggressive measures to control price growth.  Tariff Inc...
Read more