Lots of volatility to close out the week as Trump has yet again threatened additional and heightened tariffs on China.
I won’t waste much of the newsletter talking about this, as I want to get into the main topic relatively quickly – that being the current state of the markets regarding artificial intelligence.
The one thing I will mention before we dig in is that Bull List stock Aritzia had an outstanding quarter. I won’t speak much on the earnings in this piece as it would likely run on a bit too long, but if you want to view my quarterly commentary and report on Aritzia, click here.
The stock is now up over ~300% from the inventory fears it was going through in 2023, and has been a prime example of how one should ignore the noise and stick to the fundamentals.
The reason for this newsletter
I have gotten a multitude of requests regarding artificial intelligence over the last month or so. While some of the companies that members have mentioned are high quality, there are also numerous speculative options being asked about.
Retail investing inflows into the market have reached all-time highs, and the amount of stocks owned by retail investors is reaching 2021 peak levels.
I’m not suggesting we will face a 2021-like correction. However, this current era of the market sure feels a lot like 2021.
Many Premium members have asked why I do not have many AI names in my highlighted companies. And to that, I would say I have plenty. Some just require some digging.
The obvious ones would be names like Amazon, Alphabet, and ASML. However, we can even look outside the box to companies like Toromont Industries (TSE:TIH) that will likely supply the equipment, planning, design, and maintenance of data centers.
Or, Brookfield Renewables, which is expected to be an industry-leading renewable power generator in the space for a long time. As we know, data centers require a ton of power.
I have been researching pure-play AI names for quite some time now in search of a new company for the Bull List. However, I just haven’t found anything overly attractive at this point in time.
I could easily fill the Bull List with stocks that I think will go higher over the next 6 months. However, this isn’t really my style. This platform is set up to deliver long-term returns, not short-term gains.
The AI Gold Rush – A Boon, or a Bubble?
I keep hearing people calling artificial intelligence the new gold rush. And without question, the spending numbers back that up. Below is a chart of their capital expenditures. Important to note, these numbers are in the billions.
Through a gold rush, it was often only the ones selling the picks and shovels that made any sort of money.
I have seen some of the most puzzling deals in quite some time over the last year, including AMD giving OpenAI equity (10% of the company) in order to buy AMD chips.
In addition to this, we have OpenAI committing hundreds of billions of dollars to hyperscalers over the coming years. The company generates $12B in revenue per year and is losing money on every ChatGPT prompt that is made.
OpenAI earns revenue mainly from licensing and subscriptions, while hyperscalers like Microsoft and Alphabet monetize through selling computing power. Businesses like NVIDIA, ASML, Taiwan Semiconductors, etc., are the picks and shovels plays.
No matter how profitable AI ends up being, outfits still need to buy the chips to operate the grand ambitions they have for AI. Yes, OpenAI might be losing money on every prompt, but they’re still paying full price for the chips. The risk here? AI doesn’t scale cleanly, demand softens, or cheaper chips from China undercut pricing.
Bubble theory says prices detach from fundamentals when valuation multiples expand faster than earnings.
That’s happening here. Not among all companies, but more so names that specialize in AI. Many AI firms trade at levels priced to perfection.
Jeff Bezos acknowledged we are indeed in an AI bubble. However, he countered that with the fact that he believes it is a “good” bubble. By this, he means that although there will be winners and losers when it comes to stocks, the societal benefit will be gigantic.
OpenAI’s proposed trillion‑dollar infrastructure plans rely on huge loans. Meta and others are borrowing heavily for new data centres. This only works if returns appear soon, and here’s why.
The assets utilized to build out data centers will depreciate very quickly. If the earnings made from the monetization of AI do not offset the declining value of the assets built out to support AI, you can get into a bit of difficulty earnings-wise.
The “this time is different” narrative is back.
The AI Capex Explosion
When people ask me about the state of the markets and AI in general, I find myself coming back to how much money is being poured into artificial intelligence infrastructure.
The top tech firms like Microsoft, Meta, Amazon, and Alphabet plan to invest more than $300 billion in 2025 on AI projects. The bulk of this will be spent on data centers.
That’s roughly 1.2% of US GDP, a scale that will remind many experienced investors of the telecom buildouts during the dot-com bubble.
Interestingly enough, if you isolate out data center buildouts from US GDP, some economists state there is only 0.1% growth. They are literally the only driving force in the economy right now.
The logic is similar to that of the telecom fiber buildouts in the 90s.
Build capacity now and hope demand catches up. The only difficulty? The majority of fiber companies back in those days were not wrong about the future potential demand. They were just way too early, and many of them went bankrupt.
I’m not drawing comparisons to say that certain companies will go bankrupt. The hyperscalers won’t go bankrupt. The comparison is more so being drawn from the fact that fiber was, and still is, a huge driver for society. However, much like those telecom companies being too early back in the 90s, we could be too early here in regards to data center ramp-ups.
The difficulty here is profitability. Once the servers are running, power costs, cooling, and depreciation all hit the margins. These companies need to be able to generate short-term profits from the buildout.
This characteristic is unlike the fiber buildout of the 90s. Many of those fiber assets are still used today. The “guts” of these data centers, however, that being GPUs, networking switches, etc., only have a lifespan of around 3-5 years before the tech will advance beyond their useful life.
Which leads me to my next important point.
The 95% Problem: Most AI Projects Still Don’t Pay Off
There was a recent MIT study showing that 95% of enterprise AI pilots fail to produce measurable returns.
I tend to go back to Constellation Software’s comments about how a lot of AI tools are:
“solutions looking for a problem”
Sure, the use case for many pieces of AI looks promising. It does what it’s supposed to do. The issue is? Nobody really has the problem it’s looking to fix, so nobody uses it.
The takeaway? AI doesn’t fail on its own. It fails when companies develop AI solely for the purpose of developing AI, whether it is to satisfy shareholders or to try to maintain relevance.
The issue is that when AI doesn’t help you solve real issues and it is instead just a novelty, eventually the money stops flowing into it.
Retail Investors Are All-In But Institutions Are Stepping Back
As I mentioned previously, retail participation in the markets is at all-time highs. Trading data shows individual investors driving a large share of stock volume, often chasing artificial intelligence names.
The irony, but not something that should be all that surprising, is that institutions seem to be easing off. As you can tell by the chart, they were heavy sellers in the summer, while retail investors continued to buy. Although this chart is a few months old, the data still remains true today.
Reports show retail investors buying while large funds have sold into this strength. Retail investors being late to the party is about as guaranteed as the sun coming up in the morning, and it is happening again.
This is not a panic exit from institutions. It looks to me like institutional investors are just seeing stretched valuations and heightened risk.
Hyperscaler Margins Are Under Pressure
The hyperscalers are pouring record sums into AI infrastructure. The numbers are absolutely staggering, zero doubt.
Yet profits haven’t kept pace. Operating margins are thinning as investment cycles stretch out longer and monetization isn’t coming, at least not yet.
At this point in time, the cash outlays are not resulting in meaningful revenue generation. This is perfectly fine right now, but if it is a trend that continues, there will be immense pressure on these companies.
As shareholders, we’re entitled to our portion of the free cash flow a company generates. If that money is wasted on assets that don’t generate meaningful returns, we pay the price.
This is the catch at this point in time. High spending doesn’t guarantee high returns.
Look to a company like Oracle. Once a strong cash flow generator, the company is literally going all in on artificial intelligence, turning free cash flow negative and spending an extensive amount of money on capital expenditures.
When capex outpaces cash flow, even giants like Microsoft, Amazon, and Meta feel the pressure. Yes, they have their core businesses to fall back on, which makes this different from the dot com bubble.
But once these fast-depreciation assets hit the income statement, they’d better be making profits on them, or it will hit earnings.
What If the Spending Works?
I’ve gone over all of the potential negatives towards this large-scale AI spend. So now, I want to focus a bit on the positives.
I believe the markets are priced to absolute perfection right now, especially on the AI side of things. However, it’s important to understand that perfection can occur. These companies could ultimately execute well.
If the billions going into AI infrastructure pay off, the markets could end up going much higher. The real win will be a productivity multiplier that lifts corporate profits across a multitude of industries. I have never witnessed a piece of technology that impacts virtually every segment of the market. But AI does.
If machines and AI can take over more routine work, firms could see margins rise meaningfully. Unfortunately, this would probably come at the cost of human jobs. But we can keep in mind that disruptive technology has done this many times.
Let’s take the healthcare sector as an example. Faster drug discovery, modelled through AI simulations, could shorten development timelines.
The benefit compounds quickly: quicker testing means earlier revenue and fewer failed trials. That’s how small time savings, multiplied across an entire pipeline, quietly build up. And this is just one example. Many industries could see it.
Hyperscalers Could Cement Their Dominance
AI runs on computing power, and computing power costs a fortune. That’s where the hyperscalers like Amazon, Microsoft, and Google still have the edge.
Their data centers already dominate global capacity, and their scale lets them absorb the rising cost of GPUs, energy, and cooling without breaking margins. When the companies leading the charge in a particular tech revolution are not unprofitable, speculative companies, but companies with hundreds of billions of dollars in spendable cash, that is a formidable moat to match.
The business model is simple: rent out compute, storage, and AI tools built on top of technology they already control. This creates a data moat and network effects that keep customers inside their ecosystems. Once a firm builds its AI pipeline on one platform, switching costs rise fast.
In 2025, the top four hyperscalers are expected to invest more than $320 billion in AI and data centers. That kind of spending raises barriers for smaller players. In fact, it makes it almost impossible to compete.
It’s Not Just The Hyperscalers, However
Thinking outside the box is critical when it comes to this AI boom. It is not just tech companies that are benefiting.
Utilities, healthcare, construction companies, etc.
In fact, I think the utility sector is one of the most underrated out of all of them.
Ultimately, AI will not succeed if we do not have the power to supply it. With a single data center drawing as much power as a small city, this is a gigantic issue. Electricity demand is the first and most important pressure point.
I would expect grid expansion and new transmission projects to ramp up substantially as utilities race to meet this load. This is why I have been looking so closely at a company like Quanta (PWR). But the valuation at this point worries me.
I think one of the main issues we may face in this regard is the fact that no matter how quickly you want to satisfy power demand, these are slow-moving industries.
If a retail company is seeing heightened demand, it doesn’t take much to construct some new warehouses and hire employees. They could have this done in a year, possibly even less.
But building transmission lines, substations, and power generation facilities? You don’t just snap your finger today and they appear tomorrow. This will take a long time and could really put a bottleneck on the technology.
However, these utilities are firmly in the “picks and shovels” department of AI. Even if grid expansion takes half a decade to see meaningful results, it won’t be the utility providers that suffer; it would be further down the line. Which is why I think they’re an interesting play in the space.
So, What Is My Opinion on This Current AI Boom?
When a company generating $12B in annual revenue is committing hundreds of billions of dollars to other companies, and those companies (AMD) are giving it equity stakes in their company in order for them to purchase their chips, it would be difficult to argue against the idea that we are indeed in a bubble.
In fact, Jeff Bezos, who has a bit of experience when it comes to previous bubbles, has straight-up said we are in an AI bubble. He mentions that, like previous bubbles, there will be winners and losers.
He also mentioned that we should not associate the bubble conditions in the market today with the benefit of the technology, that being AI.
I tend to agree with this. Heavy speculation in AI stocks today doesn’t make AI inherently bad. But don’t confuse this with many AI investments being strong long-term investments.
Right now, valuations in the big AI-linked firms look stretched but not outlandish, and in the smaller AI pureplays, they’re at nosebleed levels. The difficulty here is that a large sum of the capital is tied up in a small portion of equities, and all of those companies are depending on each other to continue to fuel the industry.
Proof of potential profitability remains spotty. Some firms are actually showing real revenue traction from cloud-based AI tools. But the majority are simply banking on future potential demand.
Without clearer returns on invested capital, this area of the market will be fueled by speculation.
However, there are plenty of names I’m bullish on in regards to AI in the future. They’re mostly all featured here at Premium – names like ASML, Google, or Amazon. And, I will continue to try to find more.
