CIBC Tech & Innovation Market Update – February 2025

Markets experienced turmoil at the end of January, after a Chinese AI startup revealed it was able to build a large language model using just a fraction of the chips American companies use to train their respective models, and therefore at a fraction of the cost. DeepSeek released its AI chatbot earlier this month, and then published a white paper detailing the training process and cost behind the model. According to the paper, DeepSeek spent only ~$6 million to train its V3 model, a drop in the bucket to the hundreds of millions OpenAI, Google, Anthropic and others have reportedly spent training their respective models. While Monday’s reaction stemmed from a concern that DeepSeek’s existence will pop the AI bubble, it remains to be seen whether DeepSeek’s emergence will be truly disruptive. While DeepSeek’s model could result in lower training costs, an important factor still to be considered is AI inference (usage) costs, especially in a scenario of widespread adoption. We break down how DeepSeek shook up the market and why big tech tumbled while software rallied.

The evening of January 18^th was a somber one for approximately 170 million users of TikTok. The app stopped working in the U.S. around 10 p.m. ET, less than two hours before a nationwide ban was set to go into effect. Instead of their usual home feed, users saw a pop-up message on screen that stated a law banning TikTok has been enacted in the U.S., making the app inaccessible. The message further stated that “President Trump has indicated that he will work with [TikTok] on a solution to reinstate TikTok once he takes office.” Users didn’t have to wait until Monday though. The app began working again on Sunday afternoon after Trump indicated he would sign an executive order delaying the ban. This proposal was upheld on Monday when Trump postponed the ban for 75 days. While legal experts have stated an executive order cannot override an existing law, the order effectively tells the U.S. attorney general to not enforce the law for now. It remains unclear what happens to TikTok after 75 days, but it is becoming increasingly clear a full or partial divestiture of the U.S. business will be required for the app to continue to operate.

When ChatGPT was publicly released in late 2022, some of the first headlines that followed discussed the risks of AI hallucinations and biases in AI responses. These risks, along with the necessity of data protection and security, resulted in initial slow adoption of generative AI across industries that deal with high volumes of confidential data. Over the last two years, startups have flooded the market with solutions that target the niche needs of these sectors, mainly legal and healthcare, and we’ve since seen a strong uptrend in genAI adoption. Financial services has lagged, in part due to stringent compliance regulations that apply to subsectors such as banking and capital markets, and in part due to a lack of available solutions that target the niche needs of financial services providers while still remaining compliant and secure.

Meta announced that it will be ending its third-party fact-checking program, starting in the U.S. In his blog post, CEO Mark Zuckerberg stated that too much harmless content gets censored as a result of Meta’s policy, and pointed out that experts have their own biases and perspectives that show up in choices made about what to fact check and how. Meta’s move to embrace looser guidelines has sparked both praise and criticism online. President-elect Donald Trump, who previously criticized Meta for its fact checking policy, referring to the policy as a way to censor right-wing voices, praised Meta for its decision. Others have expressed concerns over potential increases in false and misleading content.