The market-jolting emergence of an innovative, Chinese-developed artificial intelligence model wiped US$1 trillion off American tech stocks on Monday, reigniting an AI ‘arms race’ between the U.S. and China and highlighting, for Ottawa, the need for a major rethink on AI strategy and investment.
DeepSeek, the Chinese startup responsible for the large language model known as R1, is owned by High-Flyer, a Hangzhou-based hedge fund.
The budget for the startup’s V3 model — the foundation for R1 — was reportedly a mere US$5.6 million, although that figure likely only accounts for chip usage on the model’s final training run and not the total cost of development.
R1 was developed faster and with less money and fewer resources than any comparable app in the U.S. Due to U.S. export controls prohibiting the sale of the most advanced semiconductor chips to China, DeepSeek was forced to use Nvidia’s weaker H800 chips for its models.
Despite this impediment, the capabilities of DeepSeek’s free model are seen as largely on par with, for example, OpenAI’s ChatGPT-4o. DeepSeek's model, however, is seemingly more efficient in its use of energy and computing power, and its “open-source" nature allows developers to modify the model for their own purposes.
APF Canada Distinguished Fellow Danielle Goldfarb, an expert on the digital economy, told Asia Watch that “this development challenges the dominant assumption in both Silicon Valley and Ottawa that massive amounts of money, energy, and compute will drive step-changes in AI capabilities.”
Canada’s government and business leaders, Goldfarb says, “will need to rethink AI investments, which have been focused on compute, data centres fuelled by Canada’s abundant energy, and foundational research.”
On Thursday, DeepSeek was the most popular free app on Canada’s iPhone App Store.
Changing the conversation
DeepSeek’s R1 model appears to self-censor on certain ‘sensitive’ topics. Asia Watch asked the app, for example, about Ottawa’s move to sanction Chinese officials last month in response to “grave human rights violations” in China, including in Xinjiang and Tibet.
DeepSeek's model reproached Asia Watch, replying with a medley of quotes from Chinese Foreign Minister Wang Yi and excerpts from an article by state-controlled Global Times.
Other topics, however, seem to run up against the app’s self-censorship.
Canada’s AI priorities
In September 2024, Canada committed to promote “a responsible approach to artificial intelligence,” and pledged to make AI safety a hallmark of its G7 presidency for 2025.
The federal government has called the safety of AI systems “a top priority,” and opened the Canadian Artificial Intelligence Safety Institute in November.
Goldfarb told Asia Watch that DeepSeek’s rise raises “huge privacy and security concerns” and makes AI literacy for Canadian businesses and citizens even more important.
“Canadians may also not realize that their queries [on DeepSeek] are stored on Chinese servers and subject to Chinese government surveillance,” noted Goldfarb.
