Buzzes have been circulating for several days about a new artificial intelligence company behind a powerful ChatGPT competitor in China.

At last week’s World Economic Forum in Davos, DeepSeek, a Hangzhou-based startup, was highlighted as a key reason why countries like the US are ramping up efforts in AI innovation. The startup’s recently launched R1 model sparked a lot of attention, with engineers on tech forums comparing its performance to some of the leading AI models from OpenAI and Microsoft Corp. Its performance on a UC Berkeley-affiliated AI leaderboard also contributed to its rising reputation.

Interest in DeepSeek intensified dramatically within the last 36 hours, drawing the attention of significant figures in the tech world. Notably, investor Marc Andreessen and AI pioneer Yann LeCun from Meta Platforms Inc. joined the conversation, with Andreessen calling DeepSeek’s model “one of the most amazing and impressive breakthroughs" he’s ever witnessed.

By the end of the weekend, DeepSeek’s AI assistant had surged to the top of Apple’s iPhone download charts and was among the most downloaded apps on Google’s Play Store. The surge in interest overwhelmed the startup’s systems, causing the service to go offline for over an hour. As a result, the company limited signups to users with mainland China phone numbers, citing "large-scale malicious attacks" as the reason for the decision.

The sudden spike in attention had significant consequences. DeepSeek’s AI model, which it claims was developed at a fraction of the cost of its competitors without significant performance compromises, caused a dramatic market reaction. Investors began questioning major US tech companies' spending strategies, leading to a nearly $1 trillion loss in US and European tech stocks.

Nvidia Corp., an AI chipmaker, saw a record $589 billion wiped off its market value on Monday alone. Some stocks, including Nvidia’s, partially recovered in after-hours trading.

By Monday, it became apparent that the immense demand for DeepSeek's services was causing significant strain on the company's systems. The startup announced on its status page, "Currently, only registration with a mainland China mobile phone number is supported." DeepSeek did not clarify whether this limitation is temporary or how long it will be in effect.

This marked the company's most extended significant outage since it began reporting its operational status. Unlike competitors, DeepSeek's assistant provides transparency by showing the work and reasoning behind its responses to user queries. This feature has received positive reviews on the Apple App Store and Alphabet Inc.'s Android Play Store.

Founded by Liang Wenfeng, the head of a quant fund, DeepSeek's open-sourced AI model is prompting a reevaluation of the vast amounts of money companies have invested to stay competitive in the AI sector.

"While it remains to be seen if DeepSeek will prove to be a viable, cheaper alternative in the long term, initial worries are centred on whether US tech giants' pricing power is being threatened and if their massive AI spending needs re-evaluation," stated Jun Rong Yeap of IG Asia.

Like many AI models developed in China, DeepSeek is designed to self-censor on politically sensitive topics. Unlike ChatGPT, which offers more open responses, DeepSeek avoids discussions about issues such as Tiananmen Square, President Xi Jinping, and the potential for China to invade Taiwan. This could surprise international users who may not have encountered Chinese-made chatbots before.

DeepSeek’s early success contrasts the notion that the most sophisticated AI systems need significant computing power and energy. This idea has fuelled stock surges for companies like Nvidia and its suppliers. While the startup has not fully disclosed the costs of development and energy use for DeepSeek, it has shared estimates suggesting that the expenses are much lower than those of OpenAI’s latest models. This is notable, especially given that DeepSeek emerged from China, a country facing increasing US trade sanctions on advanced Nvidia chips, challenging the notion that these restrictions are fully effective.

As computer scientist Kai-Fu Lee pointed out earlier this month at the Asian Financial Forum in Hong Kong, "The US is great at research and innovation and especially breakthrough, but China is better at engineering. In this day and age, when you have limited computing power and money, you learn how to build things very efficiently."

Nvidia, the leading chip supplier for training AI software, has praised DeepSeek's latest model as an "excellent AI advancement" that fully adheres to US government restrictions on technology exports. The company highlighted that DeepSeek's work demonstrates how new models can be created through test time scaling.

This statement from Nvidia addresses concerns from some analysts and experts who questioned whether the Chinese startup could have achieved the breakthrough it claims. Nvidia further emphasised that while inference, running AI models and using them for data processing and predictions still rely heavily on its products.

"Inference requires significant numbers of Nvidia GPUs and high-performance networking," said the company.