The Mechanics of AI: An Investor’s Guide to the Technology Landscape

Remaining focused on the longer-term investment opportunities, we return to our discussion on Artificial Intelligence. For investors looking to make sense of current market valuations and tech-driven headlines, having a clear grasp of the technology itself is essential to your broader wealth planning strategy. To provide that clarity, we are breaking down the core elements of AI, including:

• A general summary of how AI works as a framework for discussing investment perspectives and broader economic trends.
• The leading AI models and innovators currently shaping market momentum.
• AI training and industry leaders in data processing and infrastructure.
• Overflow to other investment sectors.

Having a basic understanding of the mechanics of AI helps to extend the discussion into the subsets of the innovation, ensuring you are better equipped to navigate today’s technology-focused market environment.

The Mechanics of AI: The Foundation of the AI Market

The basis of AI is the mathematical model itself. Everything else in the AI industry (training, processing, energy) is there to support the model, creating a vast, interconnected investment ecosystem. An AI model has no consciousness, goals, experience, or ability to understand meaning the way humans do. While the model can answer a question like a human may answer, it gets to that answer in a different way.

Layer after layer of mathematical functions working together give AI the ability to identify relationships in data. From a market perspective, this ability to process, predict, and monetize massive datasets is exactly what drives the commercial value of these technologies. During a training process, the model will guess an answer. For example, “The dog makes a lot of noise when it______.”  At first it guesses randomly. Then it gets better and better. It does this for increasingly complex content and begins to understand how words interact.

Through this training, words and word pieces are assigned a lookup token and vector. The vector is a series of numbers used in formulas and functions to understand how words (which are converted to numbers) interact. Understanding this mechanical complexity helps investors grasp why the barrier to entry for building these models is so high, requiring immense capital, infrastructure, and computational power.

AI does not look up the definition of dog. Instead, AI understands how that word, when converted to numbers (its vector), interacts mathematically with other words. When you ask AI a question, it predicts an answer, one word at a time. While it works quite well, the model only knows how humans discuss dogs from the data it has been trained on, without lived experience.

Language Models vs. Foundation Models

These models can be designed for different purposes. When people use applications like ChatGPT to ask questions, they are using mostly a “Large Language Model.” Let’s define this because it shows up in the news and heavily influences technology sector valuations. First, “language” means it is designed to handle text. Large refers to the number of parameters in the model. A large model would have tens to hundreds of billions of parameters, while a small model may have a million to a few billion parameters.

More parameters would be like a TV with higher resolution. A small model would have blurry resolution, and a large model would be much clearer. However, a small model may be better than a large model if it is designed for a specific task. It could run faster and potentially get to the answer just as well or better than the more complicated model, making smaller, specialized models highly attractive for businesses looking to manage computing costs and scale their AI adoption efficiently.

Although I suggested that we are usually using a large language model when we use ChatGPT, the actual model would be termed a “Foundation Model.” This is because it has more than language built into it. It includes architecture that recognizes patterns in images and sounds, via a conversion to numbers of course. From an investment management standpoint, these foundation models are significant because they serve as the underlying infrastructure that countless other software companies build upon, creating a massive economic multiplier effect.

From the model, to training, to processing, to implementation, there are numerous industries and companies leading the way and vying for market share. Our investment portfolios prudently capture the various areas and opportunities across this expanding technology ecosystem.

Top AI Companies & Leading Innovators

There are currently three leading companies for general purpose AI that are commanding significant capital and market attention: OpenAI (ChatGPT), Anthropic (Claude), and Google DeepMind (Gemini).

• ChatGPT: Scores highly for reasoning, coding, math, and different data types (“multimodal”). Models are already embedded in software such as Microsoft products.
• Claude: Very close to ChatGPT but leads in longer and more involved requests (“longchain reasoning”). Models tilt toward the enterprise market via partnerships with Amazon and Google Cloud.
• Gemini: Very strong on math and research, but not as strong on reasoning. While Gemini may not dominate, its advantage is being embedded throughout Google’s huge technology infrastructure.

From an investment standpoint, OpenAI is privately owned, but portfolios already have indirect exposure to Microsoft, which will benefit from its massive stake in OpenAI. Anthropic is also private, but it is backed by Amazon and Google, both of which are well-represented in portfolios. And Gemini is Google, which many of your portfolio managers emphasize.

Capturing the Enterprise Market: Focused and Open-Weight AI

Other companies are developing more “focused” AI to capture different segments of the corporate ecosystem.

• Meta (Facebook/Instagram/WhatsApp): Developing AI models that companies and developers can use in their own social media and advertising platforms. Meta remains a top holding in the S&P 500 Index, Clipper, and TCW Concentrated Growth.
• Mistral AI: Focuses on enterprise models that companies can embed. A private French company backed by major tech corporations and venture capital.

From a business model perspective, it is important to note that Meta and Mistral AI are “open-weight AI,” allowing companies to run the models on their own servers locally. Conversely, ChatGPT, Claude, and Gemini are “closed AI,” meaning the models run via processing centers, not locally. This dynamic drives massive, recurring cloud computing revenue for the major tech platforms hosting them. Plus, many other companies like Adobe, Bloomberg, Thomson Reuters, and countless others are building internal AI solutions within their respective industries to drive operational efficiency and create new, proprietary revenue streams.

The AI Hardware Supply Chain: Chips, Training, and Data Centers

Out of the box, AI models have no knowledge and must be trained by running them through huge datasets. This is an ongoing process and is massive in terms of computing power and cost, creating a distinct competitive advantage for mega-cap technology companies with deep pockets.

Modern models are trained on tens of thousands of advanced chips, primarily produced by NVIDIA. Each chip can be $25,000 to $40,000, and the process requires about 20,000 to 50,000 chips. Considering all costs (including electricity, which is significant), one frontier AI model can cost hundreds of millions of dollars to train. This staggering capital expenditure creates a formidable barrier to entry, effectively limiting the top-tier AI race to a handful of exceptionally well-capitalized players.

User demand will put continuous pressure on data centers as companies and individuals adopt the new technology. In our 2025 third quarter commentary, we discussed the scale of this buildout in terms of many trillions of dollars, representing a generational infrastructure supercycle.

The big players in this area serve as the foundational hardware supporting this global expansion. To effectively capture this growth, you’ll need to look at the entire infrastructure supply chain:

Beyond Tech: AI’s Impact on the Broader Economy

The economic impact of AI extends well beyond the tech sector, creating secondary investment opportunities and risks across the broader market:

• Industries that support AI (The Secondary Beneficiaries): Non-tech industries that support the physical buildout of AI infrastructure include electric utilities and power generation, construction and engineering, cooling/HVAC, and real estate.
• Industries that benefit from AI adoption (Driving Margin Expansion): Healthcare, financial, manufacturing, professional services, advertising, and media are leveraging these tools to drive operational efficiencies and reduce long-term costs.
• Industries that may be pressured or disrupted by AI (Obsolescence Risks): Back-office service providers, entry-level work involving basic research, drafting, and analysis, and traditional software vendors may face pricing pressure or structural disruption.

Conclusion: Navigating Valuations and Asset Allocation

While we feel that some share prices in the U.S. are too high, many are not overpriced when considering the transformative economic potential of AI. Looking at the prices in a vacuum without factoring in this growth would lead us to go overly light on investment in large U.S. companies. We are maintaining a healthy weighting at about 40% of equity exposure to U.S. larger-cap to ensure we capture this momentum.

We continue to appreciate the benefits of the international equity holdings which provide more traditional and more diversified exposure (and as noted above, some key companies in the AI technology food chain). In other words, we do not have to lighten one category or another; the usual mix works well in this environment.

If you have questions about how these technological shifts impact your specific asset allocation, please contact our team for a portfolio review. As we gradually approach the warm season, we hope that you can make time for lots of fun activities and travel. An occasional sunny summer day is sneaking in every so often, and we will eventually see this in the markets, too.

Straight From Aufman Associates: Answering Your Top AI Investment Questions

Is OpenAI publicly traded, and how can I invest in it? 

While OpenAI (the creator of ChatGPT) is a privately held company and not directly publicly traded, it is still possible to gain exposure to its technological advancements through other channels. For example, many investment portfolios already have exposure to Microsoft, which stands to benefit from its relationship with OpenAI. Looking at the established public partners of these private innovators is a common way investors participate in their growth.

Beyond the household names, what other types of AI companies are out there to invest in? 

When looking at how to invest in AI, it is helpful to look beyond consumer applications and focus on the underlying infrastructure ecosystem. As we noted, the cost to train these models is massive, requiring tens of thousands of advanced chips and massive data centers. This creates opportunities across the entire supply chain, including chip designers (like NVIDIA and AMD), semiconductor manufacturers (like TSMC), and the companies providing data center facilities and networking components.

With so much focus on AI investing, are U.S. tech stocks currently overpriced? 

It is true that some U.S. share prices appear high at first glance. However, when evaluating these opportunities, one must factor in the transformative economic potential driving the sector. Looking at prices in a vacuum might lead an investor to go overly light on large U.S. companies. To balance this, we maintain a healthy weighting of about 40% to U.S. larger-cap equities, paired with diversified international holdings, ensuring a mix that works well in this environment without having to drastically lighten one category over another