Market Ethos
27 July 2026
Taking AI Inventory
Sign up here to receive the Market Ethos by email.
In 2022, as yields rocketed higher, a recurring question investors asked their advisors was: “How much do I have invested in bonds, and what exactly is duration?” Gold’s rise in 2025 prompted a different question: “. How much gold do I own?” In 2026, the question has shifted to: “How much AI exposure do I have in my portfolio?“
How much gold was a straightforward calculation; how much AI is not. This “bubble”, driven by revolutionary technology and the race to build the infrastructure to deliver, is no longer just a hyperscaler, semiconductor or memory story. The demand has spread to other technology suppliers — from networking gear to servers and integrators.
Beyond the technology sector, this infrastructure buildout is impacting utilities, power suppliers from gas to nuclear to renewables, storage, backup power, construction, engineering, industrials, HVAC, and copper demand to rare earths.
As the market impact of AI becomes pervasive, every portfolio is exposed. This elevates the question from ‘How much AI?’ to ‘What kind of exposure?’ to ‘Where is my AI exposure coming from?’ Those are more difficult questions to answer, and it isn’t from lack of diligence. The market has quietly made AI exposure a default setting, and there’s no line item on a statement that says so.
This week’s Ethos is not another entry in the AI bubble-or-not-bubble debate. There is a compelling rationale for owning a lot of AI exposure or a little. But not knowing how exposed your portfolio is, or how it is being exposed, is not good. This is a framework to help answer those questions.
The AI everywhere challenge
If you asked anyone a year ago how much AI exposure is in emerging markets, most would likely have said very little. Whoops, there is a ton, from China to Korea to Taiwan. Starting closer to home, the S&P 500’s AI score based on our model is roughly ~45% in AI-related or impacted companies. AI-related is the chip supply chain, the cloud and data centres, the model builders and software. Add in a number of other companies in non-tech sectors that are impacted, and you get up to nearly half the index. If you own the market, you have made a large, concentrated AI allocation just by being allocated to the largest equity market in the world.
But this is not just an American story, although the flavour changes as you travel to different markets — not just in terms of overall exposure, but also the type of exposure.
A third of global equities — about a quarter among emerging markets or Japan — are influenced by AI-related technologies. Europe and Canada are less exposed, at or below 10%. Altogether, that means a bit under $50 trillion of the global market capitalization appears to be influenced by this changing technology, to varying degrees.
But the headline number is only half the story, because the type of AI is completely different in each bar. American- domiciled AI is mainly chips and model platforms — NVIDIA, Microsoft, Alphabet. Japan’s AI is chips plus something the S&P barely has – robotics and automation. Canada’s AI is almost entirely application software: Shopify, Constellation — with essentially zero chip or model exposure. Emerging markets are the purest picks and shovels trade on earth: foundries, memory, packaging.
While most companies span multiple AI subsectors, we categorize each company based on the most dominant. NVIDIA is an Enabler, Infrastructure includes both software or utilities providers like Palantir or Emera, Alphabet is in Models/Platforms, Salesforce in Applications and Physical AI are robotics and other companies.
Same label, but four very different situations. A client holding Japan and a client holding the TSX both “own AI”, but their exposures have almost nothing in common.
Hotdog, not hotdog
Here’s the challenge: There is no single AI sector, no GICS classification, no Morningstar category, no index with a settled definition of AI. “Technology” doesn’t work — utilities building data centre power are in the trade, and plenty of technology companies are not. Is Amazon an AI company? Morningstar files it under internet retail. Tesla? That’s currently an automaker. Both share prices are clearly sensitive to AI.
Want proof the definition problem is real? Take emerging markets. A standard cap-weighted EM fund scores about 24% AI on our framework. An EM fund that excludes China — resulting in higher concentration in Taiwan and Korea — scores about 43%. Same asset class, similar line on your statement, and the AI exposure nearly doubles based on a single twist on exposure.
Some would say, this is a hotdog, not a hotdog challenge. Or for those readers versed in machine learning vocabulary, it is a labelling issue. Is a company, or more specifically its share price, being impacted by this AI revolution?
Complicating matters even more, a hotdog today may not have been a hotdog a year ago and may not be a hotdog in 2027. Cisco was not an AI-sensitive company a year ago, today it is. There is no perfect approach, and while some will disagree with our framework, it is clear, and evidence based.
Hotdog, not hotdog — this is our approach that uses two signals:
- The industry signal – A company counts if it operates in an AI-relevant industry — semiconductors, chip equipment, infrastructure software, and so on, in tiers. This catches the structural exposure. But it misses Amazon (internet retail) and Tesla (autos).
- The consensus signal – A company also counts if the market’s established AI-thematic ETFs already hold it. Call it the “the market has already voted” test. This catches Amazon and Tesla — but alone, it would miss smaller industry names no thematic fund bothers with.
A name that fires either signal gets an AI weight. That weight is a score between 0 and 1: when both signals fire it is a 1. A market-consensus-only name lands between 0.6 – 1.0, scaled by how many of the eight thematic funds hold it, while an industry-only-name receives a score of 0.9, 0.8, or 0.4 depending on how directly its business ties to AI. Funds are scored by looking through to what they actually hold, and every name lands in one of five buckets — chips & hardware, cloud & data centers, model developers, AI software, and robotics — because as the regional chart shows, what kind matters as much as how much it has.
One thing this deliberately is not is a measure of AI revenue. It measures how much equity sits in AI-adjacent companies in your portfolio. A different question — but the one that determines how your portfolio behaves when the trade lurches.
Final thoughts
The irony is not lost on us that AI was used to build the model and score AI. At the end of the day this is a framework. Is it perfect? Definitely not. But, it gives another lens into what is under the hood in a portfolio.
Nearly half the S&P is one theme. The same theme shows up as a chip designer in California, a foundry in Taiwan, a robot maker in Japan — and the risk isn’t the same in any two of them. There’s no sector tag that will surface it for you, and the fund or index you happen to pick can double it without asking.
This isn’t buy or sell. Know your number and know what kind. A 12% AI weight is perfectly defensible. So is 45%. What is hard to defend is a shrug — because in 2026, this shrug is a position, and a big one.
Ask the question about your portfolio. Not knowing if it’s full of hotdogs or not, is not an acceptable answer.
