Had you even heard about DeepSeek?
Be honest now!
Lost in all the tariff talk is a potential seismic shift in AI power dynamics. China’s DeepSeek produced an OpenAI competitor, DeepSeek-R1, at a fraction of the price. They reportedly spent less than $6MM “training” the model – excluding hardware and prior training costs, but still way cheaper than competitors – for a free application that is on par (or in some cases outperforms) OpenAI, Meta and Google’s top models1 on many metrics. Stocks related to the AI theme fell sharply on the news with highfliers such
as Nvidia and Broadcom down 17% in a single trading session. Not to be outdone, Chinese e-commerce giant Alibaba released its own model the very next day, claiming it outperformed all competitors. But not all is lost. On the contrary, while investors need to be very selective, we believe the broader implications are quite positive from a macro perspective. In fact, we believe that mass adoption of AI will not only increase global economic growth, but will also boost labour productivity which has positive implications for corporate profit margins and acts to restrain inflationary pressures. This is one of the reasons we continue to be bullish on equity markets despite the impressive secular bull market we have enjoyed for years. The key investment takeaway for us is that the best equity opportunities will transition from the compute arms dealers to industries who can leverage the great efficiency provided by AI (e.g., software app developers, power producers to feed data centers, Pharma, Industrials). More on this below.
Does DeepSeek spell the end of the “AI Trade”? Likely not. But it may very well signal a shift in power dynamics and investor interest away from “picks and shovels” – essentially the hardware that provides the computational capabilities (Nvidia and Broadcom chips, supercomputers, etc.) – toward applications and services as mass adoption accelerates due to cost declines and technological innovation. Notably, BMO Internet analyst Brian Pitz believes the potential for lower input costs2 and greater visibility into AI-related revenue (vis-à-vis enhanced personalization across every device) will be a multi-year tailwind for scaled internet names with cloud capabilities such as Amazon, Google, and Microsoft (covered by BMO Software analyst Keith Bachman). Our view is that Chinese internet giants will also most likely benefit from this trend, and they just happen to be dirt cheap at the present time with very low embedded growth expectations. Brian notes the following concrete vertical impact examples from improving gen AI capabilities:
1. Digital Advertising: Effective monetization of video and complex queries becomes more visible – improving monetization of short and long-form video, given improving LLM (large language model) capabilities. AISearch monetization will gain traction given the rise of more complex queries, which increases the number of monetizable queries.
2. E-commerce: Faster delivery speeds and personalization should accelerate greater adoption across verticals (e.g., gen AI underpins faster pick, pack, and ship for Amazon.com and off-platform commerce). This could certainly benefit Canadian e-commerce champion Shopify as well.
3. Online Travel: Personalization and discovery improvement offer gross bookings tailwinds. A simple example of this are AI “agents” that can plan and book entire trips including plane, car, hotel and restaurant reservations while finding the lower prices and most efficient routes/locations.
4. Delivery: Expanding state-of-the-art models to handle the complexities of real-world driving (lidar, radar, cameras, and external audio receivers) could accelerate car manufacturer build and user adoption, which is a net positive for Uber, Waymo and Lyft. Improved indexing, routing, and batching will also improve package and food delivery efficiency.
From an investment perspective, the $600 billion in market cap Nvidia lost in a single day was a stark reminder of the amount of hot money invested in – or rather, chasing – AI momentum stocks. Also, it shows how fickle hedge funds and other fast money traders can be when a negative catalyst appears, particularly given the parabolic move some of these equities have had and the extremely high embedded expectations they imply. In numerical terms, should Tech giants such as Microsoft and Meta decide they want to spend less on Capex (because of poor returns on investment), then growth will slow relative to current expectations and downside could be significant. As a sanity check, we updated the increase in market cap for 12 flagship companies (Tech, Communications, and Utilities) with significant exposure to the theme from 2022 to 2025. The total value of these companies tripled from the beginning of 2023 (right after ChatGPT was released to the public) to the beginning of 2025. That is an increase of US$10 trillion dollars. This means that the present value of incremental future profit opportunities from AI should at least equal this number to justify the hundreds of billions invested in data centers, supercomputers, etc. Given many models are free (or very low cost) and Microsoft has been underwhelmed by the revenue impact of its Copilot AI assistant to date, that’s a high bar indeed.
To be clear, we expect Artificial Intelligence to be an enduring investment theme. Unsurprisingly, there are many differing estimates of AI’s total impact on global growth over the next decade, driven primarily by increased labour productivity. Recent studies see a potential total positive GDP impact between 8.5% J.P. Morgan and 20% PwC. Accenture, McKinsey, and Goldman Sachs come in at 11.3%, 13.7%, and 15%, respectively. We hasten to add that all these estimates will most likely be well wide of the mark. However, it stands to reason that this technological innovation will be a net positive for the global economy. Additionally, while many jobs are at risk, history has shown that technological innovations have created more jobs than they have destroyed since the second world war.
J.P. Morgan suggested that AI has the potential to be one of the greatest productivity enhancers since the dawn of capitalism in the 18th century. The difference is that employees should be able to harness AI’s power in their work far faster than their ancestors did after the rise of the steam engine and portable power for factory machines. Even the PC/internet revolution took over 20 years to work its way into tangible productivity improvements.
Other sector beneficiaries outside of Tech:Utilities and Pharma
We expect industries outside of “compute arms dealers” (i.e., Nvidia, Broadcom, Dell, etc.) to continue gaining prominence. An example we have discussed previously centers on the enormous incremental electricity generation that will be required to power data centers over the next several years. By some credible estimates, U.S. power demand could grow at a compound annual growth rate of approximately 2.5-3% through the end of the decade after being essentially flat the last 10 years. Given the enormous size of this market already, it will require tens of billions of dollars in new power generation, transmission and distribution across U.S. states and Canadian provinces. Traditional utilities, along with power equipment manufacturers, nuclear/uranium companies, renewable power/natural gas producers and pipeline operators (to transport the gas) will all benefit from this for a long time. Canada just happens to have several investable companies in these fields.
The Pharmaceutical/Biotech industry is already leveraging the power of AI, but still has enormous opportunities ahead. The Economist predicted that 2025 would bring the first Federal Drug Agency (“FDA”) approval of compounds “discovered” by AI. Since it now takes ten years and more than US$2 billion to develop a drug from start to finish with a paltry 10% probability of success, any improvement in cost and speed to market could have a material positive impact on the industry’s profitability. Reducing failures in phase-two trials by just 20% could save nearly US$450 million on a single drug’s development.”
Specialized industry website drugtargetreview.com adds that, “recent breakthroughs in AI, such as predictive modelling, clinical trial optimization, and personalized medicine have demonstrated its potential. An example of this comes from Unlearn, a company leading the charge in applying AI to optimize clinical trial efficiency. At the heart of Unlearn’s approach is the use of AI to improve clinical trials, especially in phases two and three. Their work focuses on creating ‘digital twin generators’ – AI-driven models that predict how a patient’s disease may progress over time. These digital twins allow pharmaceutical companies to design clinical trials with fewer participants, while still providing reliable evidence to assess a drug’s effectiveness. This innovation could greatly reduce both the cost and duration of clinical trials, addressing two major challenges in drug development.
Best of all, this defensive sector has been out of favour for a long time and now trades at one of its lowest valuation multiples relative to the market in the last decade. This means that expectations are very low, and the downside protection should be high, not to mention healthy dividend yields for most large-cap stocks (e.g., Merck, Bristol Myers, Amgen). It does not hurt the industry one bit either that Republicans control the White House, Congress, and Supreme Court (even if drug skeptic Robert F. Kennedy Junior is confirmed as Health Secretary). They have historically been far friendlier to the sector.
Technical Analysis
We start the new month with the major averages very near all-time highs. At face value that’s great, but underneath the surface things aren’t so rosy. For example, there have been multi-week negative divergences building in most broad-based measures of stock performance such as the various advancedecline lines we follow or even the Equal Weight S&P 500 Index. At the same time, the number of individual stocks on the NYSE making 52-week new highs was a fraction of what we were seeing late last year. (98 new highs as of January month end versus 400-450 in Q3 and Q4 last year.) Of course, these indicators can “skate back onside” if the rally progresses but as of right now the breakout looks quite precarious and doesn’t change our call for a more pronounced/prolonged mediumterm correction later in the first quarter.
The main reason we say that is because most of the indicators in our medium-term timing model – which measures three- to six-month trends – are rolling over and going negative, which suggests markets are setting up for a more pronounced/ prolonged medium-term corrective process through late Q1 and possibly into early Q2. This includes weekly momentum gauges for both the S&P/TSX and the S&P 500, which are now fully negative for the first time since early 2024, but also breadth oscillators such as the percentage of stocks trading above short- and medium-term moving averages and sentiment gauges, which have been contracting steadily since mid-December. Granted, the unwinding of the overly bullish extremes that had been evident throughout most of 2024 will ultimately be bullish for the markets through the second half of the year, but as of right now, all segments of the market – everyone from U.S. retail investors to professional futures traders – are scaling back risk.
In terms of downside risk, the general rule is that sell signals in our medium-term timing model typically result in the major indexes coming back to their 200-day moving averages, which are currently 23,525 for the S&P/TSX and 5,625 for the S&P 500. Peak-to trough, that currently works out to a decline of 8-9% and will likely be the best buying opportunity we’ve had since the late 2023 medium-term correction (i.e., further new highs are expected following this correction).
Risk to GDP leads interest rates lower
The Bank of Canada (“BoC”) eased the policy rate by 25 basis points (“bps”) in late January, but unlike previous cuts, this was characterized as a risk management-driven decision in the face of tariff threats. If not for tariffs, we do not believe the BoC needed to ease policy in January as the economy was starting to show encouraging signs, including the stabilization of the real estate market and the inflation risks that are more balanced. By cutting rates, the BoC indicated that slower GDP may worry them more than the risk of inflation.
That is the cue also taken by the bond market where the bulk of the decline in yields was driven by declining real yields, not inflation expectations. After peaking at 3.56% in mid- January, the Government of Canada 10-year nominal yield rallied 60 basis points lower, led entirely by real rates. While the worst may have been avoided or at least delayed (tariffs have been delayed until March 1 or beyond), the trade uncertainty and the potential for the United States-Mexico-Canada Agreement (“USMCA”) – a free trade agreement between the three countries – to be renegotiated will weigh on investments, capital expenditures, and hiring, having already had an impact on business and consumer sentiment. Also potentially weighing on GDP will be the current political instability that risks impacting Canada’s attractiveness and slow foreign investments until at least the spring.
As for inflation expectations, they did not rise despite the trade dispute and this can be attributed to the combination of a couple of factors: 1) The BoC assesses that the economy is operating below its output capacity, at least until 2026, should be deflationary; 2) under a trade war, GDP is expected to slow, which would mitigate any price pressures; and 3) CPI is already running below the BoC 2% target, thanks in part to the recent temporary sales tax break.
Objectively, with expectations of potential 2%+ hit on GDP in the worst of the scenarios, the uncertainty will likely continue to support real yields in the near term. While the BoC does not believe monetary policy would be the only solution, the impact to the economy would likely force a more aggressive policy response. Our BMO economists would expect the BoC to lower its overnight rate to 1.50% should tariffs be fully imposed. Assuming no tariffs and considering the uncertainty, they had to revise their 2025 growth forecast down from 1.9% to 1.7%, but kept their expectation for the BoC to further ease its policy rate by 50 bps by the summer unchanged, noting the more meaningful downside risk to the policy rate.
They also noted that this is a fluid situation, meaning forecasts are likely to be altered again. The same can be said about the bond market. At this stage, it is adjusting to the economic uncertainty and potential policy response, with the yield curve moving lower and the steepening trend pausing. With cash and cash equivalent rates moving below 3% and a normalized yield curve (i.e., longer-term yields higher than short-term yields), there is an incentive to extend duration and move into longer maturities. At the same time, caution is warranted when uncertainty drives yields further away from fundamentals. Rates have already declined significantly, leaving limited opportunities for carry and capital gains. We believe some sectors in the recent rundown in yields have become expensive, including the 10-year term sector that remains at the widest yield differential to the U.S. 10-year Treasury yield in over 40 years.
Bonds are currently playing their role in portfolios by providing some hedging against economic risk, but may also be pricing extreme scenarios under uncertainty, and yields may continue to decline for the moment. We continue to recommend reducing overweight exposure to the cash and short-term sectors, but considering the current volatility we also believe that a gradual approach is warranted as better entry levels may be available in the near future for longerterm investments if – as we believe – tariffs may not be imposed at the end.
Please speak to your BMO Nesbitt Burns Investment Advisor if you have any questions or would like to discuss your portfolio.
1DeepSeek and Llama (from Meta) are both open-sourced compared to OpenAI, Anthropic, Amazon, and Google, which are closed sourced models. DeepSeek’s achievement with its open-source model could benefit Meta given open-source models’ ability to leverage collective intelligence, innovate faster through collaboration, and provide greater accessibility for researchers compared to closed-sourced models.
2Google disclosed the cost to serve AI queries fell ~80% YTD through July, and comments from September 2024 indicate a 97% decline.
General Disclosure
If you are already a client of BMO Nesbitt Burns Inc., please contact your Investment Advisor for more information. The information and opinions in this report were prepared by BMO Nesbitt Burns Inc. Portfolio Advisory Team (“BMO Nesbitt Burns”). This publication is protected by copyright laws. Views or opinions expressed herein may differ from the views and opinions expressed by BMO Capital Markets’ Research Department. No part of this publication or its contents may be copied, downloaded, stored in a retrieval system, further transmitted, or otherwise reproduced, stored, disseminated, transferred or used, in any form or by any means by any third parties, except with the prior written permission of BMO Nesbitt Burns. Any further disclosure or use, distribution, dissemination or copying of this publication, message or any attachment is strictly prohibited. If you have received this report in error, please notify the sender immediately and delete or destroy this report without reading, copying or forwarding. The opinions, estimates and projections contained in this report are those of BMO Nesbitt Burns as of the date of this report and are subject to change without notice. BMO Nesbitt Burns endeavours to ensure that the contents have been compiled or derived from sources that we believe are reliable and contain information and opinions that are accurate and complete. However, BMO Nesbitt Burns makes no representation or warranty, express or implied, in respect thereof, takes no responsibility for any errors and omissions contained herein and accepts no liability whatsoever for any loss arising from any use of, or reliance on, this report or its contents. Information may be available to BMO Nesbitt Burns or its affiliates that is not reflected in this report. This report is not to be construed as an offer to sell or solicitation of an offer to buy or sell any security. BMO Nesbitt Burns or its affiliates will buy from or sell to customers the securities of issuers mentioned in this report on a principal basis. BMO Nesbitt Burns, its affiliates, officers, directors or employees may have a long or short position in the securities discussed herein, related securities or in options, futures or other derivative instruments based thereon. BMO Nesbitt Burns or its affiliates may act as financial advisor and/or underwriter for the issuers mentioned herein and may receive remuneration for same. Bank of Montreal or its affiliates (“BMO”) has lending arrangements with, or provides other remunerated services to, many issuers covered by BMO Nesbitt Burns’ Portfolio Advisory Team. A significant lending relationship may exist between BMO and certain of the issuers mentioned herein. BMO Nesbitt Burns Inc. is a wholly owned subsidiary of Bank of Montreal. Dissemination of Reports: BMO Nesbitt Burns Portfolio Advisory Team’s reports are made widely available at the same time to all BMO Nesbitt Burns investment advisors. Additional Matters TO U.S. RESIDENTS: Any U.S. person wishing to effect transactions in any security discussed herein should do so through BMO Capital Markets Corp. (“BMO CM”) and/or BMO Nesbitt Burns Securities Ltd. (“BMO NBSL”) TO U.K. RESIDENTS: The contents hereof are intended solely for the use of, and may only be issued or passed onto, persons described in part VI of the Financial Services and Markets Act 2000 (Financial Promotion) Order 2001. BMO Wealth Management is the brand name for a business group consisting of Bank of Montreal and certain of its affiliates, including BMO Nesbitt Burns Inc., in providing wealth management products and services.
