Title: Is DeepSeek Pioneering a New Era in AI?
Unbelievable turmoil struck Wall Street recently, prompting a staggering $589 billion drop in Nvidia's market value. The catalyst for this seismic shift was a Chinese startup named Deepseek, who claimed to have devised an AI model that demands significantly less computational power - and consequently, fewer Nvidia chips - than previously believed. This audacious boast sent shockwaves through the tech industry, leaving the mighty Nvidia reeling and causing panic among investors.
Deepseek managed to train its new AI model for merely $6 million, utilizing a handful of scaled-down Nvidia chip systems. The veracity of these claims remains a subject of speculation, fueled by whisperings within the industry that Deepseek may be secretly hoarding a surplus of Nvidia chips due to US export restrictions.
The troubling news sent the stock of Nvidia nose-diving by almost 17%, unsettling investors who had recently driven up the price in anticipation of lucrative AI business ventures. US President Donald Trump weighed in, labeling it a wake-up call for American companies while expressing a positive sentiment concerning the potential to make AI more affordable.
However, some experts take a skeptical approach. X. Eyeé of the AI consulting firm Malo Santo states that rather than pondering whether Deepseek will surpass existing market leaders, the focus should be on how swiftly research methods employed by the Chinese startup can be effectively implemented in the US.
The broader tech industry remains hopeful that they will still require a substantial number of chips moving forward. Skepticism lingers, however, with Stacy Rasgon, an industry analyst, voicing doubt that the Chinese startup actually managed to train its model for $5 million, commenting "That didn't happen."
The CEO of AI firm Scale AI, Alexander Wang, revealed that according to his information, Deepseek has access to 50,000 H100 chip systems from Nvidia, but he can't discuss the matter due to US export restrictions. The US government appears to only allow Nvidia to sell less powerful chips to China.
Deepseek's AI model has been making major strides with its cutting-edge training methods, which have proven highly effective in reducing computing power and associated costs compared to traditional AI models. These innovative tactics include pure Reinforcement Learning (RL), which lowers the need for enormous labeled datasets, structuring the training process via Curriculum Learning, harnessing Sparse Activation through Mixture-of-Experts (MoE), and implementing a Multi-stage Training process.
The fallout from Deepseek's shockwaves extended beyond Nvidia, impacting other tech stocks as well. Broadcom's chipmaker saw a similar drop of around 17%. Energy firms were hit even harder, with companies such as Constellation Energy and Vistra suffering significant losses. However, the prospect of more affordable AI gave a much-needed boost to the stocks of Apple and Meta, the Facebook company.
Nvidia expressed concern over Deepseek's claims, fearing a potential decrease in demand for their high-end chips due to the Chinese startup's alleged advancements in reducing computational power requirements. Investors speculated about the legitimacy of Deepseek's $6 million training budget, questioning whether US export restrictions might have played a role.
