A “DeepSeek” Dive into AI Disruption

Click here to read the video transcript

Major disruption is happening in the Artificial Intelligence (AI) landscape this week, with the widespread acknowledgment of DeepSeek as both a big rival to competitors like OpenAI, Meta, and Alphabet and an improved AI technology structure through its DeepSeek-R1 model that has brought into focus the commoditization of Large Language Models (LLMs) and future of AI Capital Expenditure (CapEx). There is now a significant stock price correction of semiconductor manufacturers and other companies on the upstream side of the AI technology stack, such as cloud computing and software companies.

While this disruption is being viewed negatively by investors in the immediate term, we believe it may be beneficial long-term for the global development and implementation of AI across industries.

First, it makes sense to understand DeepSeek-R1 as it relates to other well-known Western LLMs¹, including ChatGPT, LLaMA and Gemini.

Essentially it is a free and open-source model, that, unlike ChatGPT, uses, primarily, what is known as re-enforcement learning² and a mixture of experts’ architecture (MoE)³ approach instead of the method generally used by the most well-known Western LLMs of using massive training data on a massive model scale. The MoE approach is not a new idea, in and of itself, but it involves many smaller specialized sub-models called “experts” that are trained on specialized tasks. A major benefit is that MoE models are designed to improve efficiency and scalability by dynamically selecting specialized sub-models to handle different inputs, thus the dataset sizes can be dramatically scaled up with the same computing budget as a dense model (Feed Forward Neural Network⁴), so the cost efficiencies are quite notable. The key is to know which input to route to each “expert.”

In layman’s terms, using this process can significantly reduce the cost of AI development and reduce the amount of expenditure on expensive chipsets. DeepSeek-R1 can offer similar AI services as industry leader OpenAI’s ChatGPT at a fraction of the cost.

It is important to note that DeepSeek-R1 has a dependency on the current AI infrastructure and CapEx. DeepSeek is essentially a distilled model, meaning it started with Meta’s open-source LLaMA model and infrastructure, and then was refined from there.

So, this new cost-effective large “reasoning” model is driving the big-picture focus on future AI spend across and up the technology stack. Last week, the formal announcement by the US government of the Stargate Project, involving OpenAI, Oracle, Softbank, and MGX Investment, drove further CapEx acceleration (which could potentially help both the Stargate Project and DeepSeek). This joint venture is planning to spend US$100 billion on tech infrastructure projects in the near term, with the figure rising to as much as US$500 billion over the next four years. Consequently, concerns over the valuations of semiconductor companies are warranted, considering that a few of these companies have been trading at multiples as high as 50 times their forward earnings over the last couple of years.

AI enablers and AI adopters

It is important to distinguish between AI enablers and AI adopters.

If DeepSeek is proof that leading-edge AI development can happen with less capital expenditure on expensive processing chips and semiconductors (“semis”), then this clearly may create a large restatement of the earnings potential of the world’s largest AI technology manufacturers. The large correction in many AI semis manufacturers and companies involved in the AI supply chain (NVIDIA primarily, as well as Broadcom, ARM, Schneider Electric, Amphenol, etc.) is a systemic correction on blue-chip headline stocks, also known as “enablers” of AI. However, in our opinion, the CapEx will still be spent, as the broad progress and adoption of AI are not abating anytime soon. It is possible that the learnings from DeepSeek could change spending patterns, with the amount allotted to AI training shifting to inference (as training requires a much larger processing power than inference), which could result in total spend continuing to grow. This is what matters most to semis manufacturers, rather than the new day-to-day capabilities and advancements of the various models.

AI adopters are a broader group of companies diversified beyond semis manufacturers, energy and other companies. These AI adopters are the companies finding new innovative ways to utilize and integrate AI across their processes. Monetization of new technology cycles is moving up from semis to infrastructure/power (datacenters) to software/services. It is important to note that with technology investment cycles, there comes a pivot point where capital shifts from the technology infrastructure (enablers) to software and services (adopters). We have seen this play out numerous times in the last half-century in technology ranging from the early internet to mobile.

What is the impact on i³ Investments^TM?

AI monetization is moving up the stack. We are monitoring and assessing the dynamics between enablers and adopters. We believe further massive adoption of AI may be in the future as, with these new developments, barriers to entry come down significantly.

First, with semis manufacturers, the Managers’ outlook is still generally positive. For NVIDIA, specifically, there is concern about a shift from the use of their Graphic Processing Units (GPUs) to Application Specific Integrated Circuits (ASIC), which would favour Broadcom and Marvell Technology, as these ASIC chips are custom designed with specific functions and, therefore, are more efficient than a general-purpose graphics processor, like a GPU. We can also look beyond AI semis/infrastructure to diversified AI adopters, including industries such as software, internet, pharmaceutical, and financial services, which could potentially benefit from cost and processing efficiencies from these new developments. Within our Quality Growth mandates this could impact companies such as ServiceNow, Amazon, Meta, Schneider Electric, and Wolters Kluwer (the latter three are also held in our Global Dividend Growth portfolios).

The i³ Investments Team uses an AI-driven model* to analyse stocks, however, our portfolios are, overall, relatively underexposed to the AI theme and focus on cash-flow growth through sustained productivity, some of which is derived from companies adopting AI.  We believe that sustained productivity and resulting profit margin and cash flow growth potential for such companies is likely to result in a dividend renaissance in the USA.

There are many reasons to stay bullish, Meta is expected to spend US$60-65 billion in CapEx in 2025 with plans to build a new 2GW+ data center, and Mukesh Ambani’s Reliance is reportedly buying NVIDIA’s GPUs and also has plans to build a 3GW data center in Jamnagar India. We mentioned the Stargate Project earlier, and while investors may be skeptical, given the size of the initial investment, we believe it is likely to pay off.

Also, it is important to keep in mind the “Jevons Paradox”, which is an economic principle suggesting that increased efficiency leads to increased consumption of a resource because the lower cost makes it more attractive to consumers and industries. By applying this principle in this case, greater proliferation of AI would drive more usage/inference, which will continue to be good for the entire AI complex, including NVIDIA, Broadcom, Taiwan Semiconductor. So, even with their stock price being impacted in the near term, we believe there is still plenty of potential ahead.

The i³ Investments™ Teams’ AI model* is built to help identify key features that can signal the benefits of AI adoption by companies, such as rapid increases in top-line revenue and consistent dividend growth. Our AI model architecture has been built to detect these signals, and in doing so, the i³ Investments™ Team can identify and select stocks with these desired attributes from anywhere in the AI supply chain. We are constantly developing and enhancing our AI model capabilities to help detect companies that can provide Growth, Payout, and Sustainability of cash flow, our “GPS”, across all of the i³ Investments strategies.

Performance as of December 31

The indicated rates of return are the historical annual compounded total returns including changes in unit value and reinvestment of all distributions and does not take into account sales, redemption, distribution or optional charges or income taxes payable by any securityholder that would have reduced returns. The rates of return for periods of less than one year are simple rates of return. Performance is calculated net of fees. Mutual funds and ETFs are not guaranteed, their values change frequently, and past performance may not be repeated. For series of the Funds that have less than one year of investment performance, in accordance with regulatory requirements, their performance cannot be shown. Index performance is shown for comparison purposes. Index returns do not reflect the impact of management fees, transaction costs or expenses, and you cannot invest directly in an index.

Access the i³ Investments™ capabilities

The i³ Investments Team at Guardian Capital LP utilizes GEMX*, an AI model system that can be directed to focus on various investment outcomes and geographies.

Series W and WF are exclusively available through the Funds’ Principal Distributors. SMA wrap models are also available through select sponsoring firms.

All investments are subject to risk, including loss. There is no assurance that any investment strategy will be successful. Asset allocation and diversification do not ensure a profit or protect against loss.

Please speak with your dedicated Guardian Capital LP sales representative to find out more or visit guardiancapital.com/investmentsolutions.

* GEMX is the i³ Investments^TM team’s proprietary analytics model, incorporating AI into a multi-factor algorithm programmed and trained by the i³ Investments™ team.

The i³ Investments^TM team combines quantitative and fundamental analysis in managing investment portfolios. The quantitative component of the team’s investment process has evolved as new tools and datasets have become available and, over time, new quantitative models which incorporate aspects of artificial intelligence have been incorporated.  The i³ Investments^TM team provides a modern approach to portfolio construction, combining the advantages of quantitative analysis, big data, and artificial intelligence with the experience, perspective, and decision-making of our investment team.  The application of quantitative investment analysis that incorporates artificial intelligence and machine learning in a forecast model is forward-looking and the simulated results are subject to inherent limitations. Investment strategies which rely on predictive artificial intelligence and quantitative models may perform differently than expected, as a result of, among other things, the factors used in the models, the weight placed on each factor, changes from the factors’ historical trends and the limitations of technology in the construction and implementation of the models. There is no guarantee that the use of the quantitative model and artificial intelligence will result in effective investment decisions. There are no guarantees that dividend paying stocks will continue to pay dividends. All investments are subject to risk, including loss. There is no assurance that any investment strategy will be successful.

¹ Refers to a LLM trained primarily on data from Western cultures, which can lead to a significant bias in its outputs, reflecting predominantly Western perspectives.

² Reinforcement learning (RL) is a machine learning (ML) technique that trains software to make decisions to achieve the most optimal results. It mimics the trial-and-error learning process that humans use to achieve their goals.

³ A Mixture of Experts (MoE) is a type of neural network architecture designed to improve computational efficiency and scalability by dynamically selecting specialized subnetworks (experts) for different inputs. Instead of activating all parameters for every task, MoE selectively activates only a subset, making it highly efficient for large-scale AI models

⁴ Feed Forward Neural Network (FFNN) is one of the simplest types of artificial neural networks. In the context of LLMs, FFNN is a core component within the LLM architecture characterized by a structure in which information moves in one direction only, from the input layer, through one or more hidden layers, to the output layer, without any loops or feedback mechanisms. Essentially, it is a fundamental building block used to process and extract meaning from text data within an LLM.

This commentary is for general informational purposes only and does not constitute investment, financial, legal, accounting, tax advice or a recommendation to buy, sell or hold a security. It shall under no circumstances be considered an offer or solicitation to deal in any product or security mentioned herein. It is only intended for the audience to whom it has been distributed and may not be reproduced or redistributed without the consent of Guardian Capital LP. This information is not intended for distribution into any jurisdiction where such distribution is restricted by law or regulation.

Please read the prospectus and Fund Facts before investing. Important information, including a summary of the risks, about each Fund is contained in its respective offering documents. Commissions, trailing commissions, management fees and expenses all may be associated with investments in mutual funds. Mutual funds are not guaranteed, their values change frequently and past performance may not be repeated.

There can be no assurance that the Fund’s portfolio will continue to hold the same position in companies referenced here, and the portfolio may change any position at any time. The securities discussed may not represent the Fund’s entire portfolio and in the aggregate may represent only a small percentage of portfolio holdings. It should not be assumed that any of the securities discussed were or will prove to be profitable, or that the investment recommendations or decisions we make in the future will be profitable, or will equal the investment performance of the securities discussed.

The opinions expressed are as of the date of publication and are subject to change without notice. Assumptions, opinions and estimates are provided for illustrative purposes only and are subject to significant limitations. Reliance upon this information is at the sole discretion of the reader. The opinions expressed are as of the published date and are subject to change without notice. Assumptions, opinions and estimates are provided for illustrative purposes only and are subject to significant limitations. Reliance upon this information is at the sole discretion of the reader. This document includes information concerning financial markets that were developed at a particular point in time. This information is subject to change at any time, without notice and without update.

This commentary may also include forward-looking statements concerning anticipated results, circumstances, and expectations regarding future events. Forward-looking statements require assumptions to be made and are, therefore, subject to inherent risks and uncertainties. There is a significant risk that predictions and other forward-looking statements will not prove to be accurate. Investing involves risk. Equity markets are volatile and will increase and decrease in response to economic, political, regulatory and other developments. Investments in foreign securities involve certain risks that differ from the risks of investing in domestic securities. Adverse political, economic, social or other conditions in a foreign country may make the stocks of that country difficult or impossible to sell. It is more difficult to obtain reliable information about some foreign securities. The costs of investing in some foreign markets may be higher than investing in domestic markets. Investments in foreign securities also are subject to currency fluctuations. The risks and potential rewards are usually greater for small companies and companies located in emerging markets. Bond markets and fixed-income securities are sensitive to interest rate movements. Inflation, credit and default risks are all associated with fixed income securities. Diversification may not protect against market risk and loss of principal may result. Index returns are for information purposes only and do not represent actual strategy or fund performance. Index performance returns do not reflect the impact of management fees, transaction costs or expenses. Certain information contained in this document has been obtained from external parties, which we believe to be reliable; however, we cannot guarantee its accuracy.

Guardian Capital LP is the Manager of the Sustainable Funds and also manages portfolios for defined benefit and defined contribution pension plans, insurance companies, foundations, endowments and investment funds. Guardian Capital LP is a wholly owned subsidiary of Guardian Capital Group Limited, a publicly traded firm listed on the Toronto Stock Exchange. For further information on Guardian Capital LP, please visit www.guardiancapital.com. All trademarks, registered and unregistered, are owned by Guardian Capital Group Limited and are used under license.

Published: January 30, 2025