We just got an answer to the most important question in the stock market…
Where is the AI revenue?
Answer: it’s at Microsoft and OpenAI.
Both companies just released revenue numbers that are blowing the lid off the misguided argument that we are still pre-revenue when it comes to AI.
We know we have the “I”, and we’re now seeing the “R” when it comes to ROAI (“Return on AI Investment”). News outlets have been crowded with stories on how much money is being spent on AI. We see it in the professional-athlete-sized contracts being offered to AI researchers. We also see it in the headlines about tech companies building Manhattan-sized data centres that cost tens of billions of dollars.
But now, we finally have a solid look at the other side of the picture: revenue generation and cost-cutting.
Last week, we got the ever-important megacap earnings period that offers a boots-on-the-ground look at how the market narrative is materializing on a company-by-company basis. Let’s break it down.
Mega Data Centres: Capex Spend Ballooning
Let’s start with the “I”: the investments being made in AI.
Capex spending is going up. The numbers were too low. In Q4 2023, CEOs across the board jumped on their quarterly conference calls and said, in one way or another, that, “We would rather overspend on AI capacity buildout than get left behind.”
They followed through. Look at the 2023-2024 numbers below. Each of these CEOs has also recently said that this platform shift is larger than any they’ve ever seen, including the internet. They’re spending like cowboys to build it out, and some (like Meta) are betting the entire ranch.
The earnings period showed just how quickly these goalposts are moving. Take these Evercore charts, for example. The chart on the left is from before Q2 earnings, while the chart on the right is from after Q2 earnings (these charts were published ONE MONTH apart). Look at that increase in Capex projection growth rates: from 51% to 72% for 2025 and from 10% to 24% for 2026. The goalposts just moved in a huge way.
But I’ll tell you what: I think these numbers are still too low. I still think that these revised numbers can go higher for 2026. Sure, we might not get three-in-a-row on 50%+ Capex YoY growth, but I believe this buildout is far from over. I’ll give you two reasons; one is just thinking through it, while the other is a bit more mathematical.
The first reason I think that we are still too low on 2026 and beyond is that we haven’t fully implemented AI into workloads – not even close. I’m in a bit of an echo chamber in that I live and breathe this stuff, so adoption to me looks high because I use ChatGPT a hundred times a day. But when I talk to two equally reliable sources (Fortune 500 IT departments and my beer league hockey dressing room), I can tell that we’re very early in the adoption curve. We haven’t scratched the surface on what it looks like to fully implement these tools and unlock greater productivity.
While a popular analogy for baseball-loving Americans is, “What inning are we in?”, I’ll use a hockey one for Canadians: we haven’t even elbowed our way through the crowd at Union Station to line up at Gate One for a Leafs game.
The second reason is that signed capacity doesn’t line up with the current Capex consensus. Cantor recently put out this cool thinkpiece that links total signed capacity from hyperscalers to the Capex required to build these facilities. In order to build out a data centre, you need an immense amount of power. In order to get this power, hyperscalers need to line up signed capacity as the offtaker. While we don’t have full transparency on all the megacaps, we can use Google as an example.
Let’s take a look at the breakdown:
Look at that gap! According to this buildout, the street is 31% too low in 2026 and 83% too low in 2027. Sure, a lot can change in the meantime, but if this thinking is directionally correct, we are still too low in Capex projections for the next two years.
What does this mean for NVIDIA? While NVIDIA has probably captured 95%+ of the revenue/profitability of the AI hype cycle to date, I still think they have more room to run because of one key thing: product cycle cadence.
NVIDIA has shortened their time between new series of chips and has created a flywheel where each chip gets magnitudes better at each turn. CEO Jensen Huang isn’t going to stop; his competitors are failing, and he will still collect software margins on hardware for a while longer. Do I think it will end? Yes. Soon? No.
What NVIDIA is doing here is owning the computing ecosystem. We’ve seen this before in each compute cycle: mainframe, PC, smartphone, and now parallel processing. When a company owns the whole ecosystem, it can capture a ton of value. While this may be an ambitious target, Evercore’s Mark Lipacis makes the case that NVIDIA can make up 12-16% of the entire S&P500.
Talent War: Opex Is Going Up, But the COMPOSITION is Important.
On the same side of the ledger is headcount (Opex, or operational expenditure).
Opex will net go up, but it will not look the same. Meta CEO Mark Zuckerberg is going full founder mode, and personally WhatsApp messaging AI researchers who are already in the top jobs at the top labs. In 2021, NBA legend Steph Curry finalized a four-year, $215 million contract extension. Zuck recently offered 24-year-old Matt Deitke $250 million over four years to come work at Meta’s superintelligence lab, and he accepted.
But in an interview, Zuckerberg framed this expense very eloquently. The cost for the best in the business may look insane, but it still pales in comparison to the Capex spend. When you’re spending $90B+ on Capex next year, a couple of $100M AI researchers here and there are well worth building the best products on the best semiconductors.
What we’re seeing at the same time is headcount cuts in the middle. There are no longer managers reporting to managers reporting to managers. Rather, there are high-agency managers who are asking the right questions and using an army of geniuses (AI) to get things done. The low-hanging fruit for automation was software engineers and customer service agents, but this AI is coming for nearly every knowledge-based job.
We’re already seeing cuts at Amazon and Microsoft as each company cites hundreds of millions in savings on headcount in the middle and bottom layers. Employment growth at these companies is at 15-year lows, slowing from 45% in 2020 to 21% in 2021 and 2% in H1 2025. That is a massive slowdown in hiring.
What this is ultimately doing is creating significant operating leverage as AI companies eat their own cooking, becoming the lowest friction high adopters. Headcount savings is one component of the “R”; now, let’s get to the other one.
Here Comes the Revenue
Microsoft and OpenAI are showing that the revenue is here, and the growth is parabolic. We got two key data points last week on the revenue front. One came from a very well-known source (earnings), while the other came from my favourite web journalists (The Information). Let’s start with Microsoft.
Satya Nadella, Microsoft’s CEO, has proved that even large ships can sometimes turn quickly. Although Microsoft missed the mobile platform shift, they got to the AI party way before everyone and made the most important partnership with the most important company of this era, OpenAI. Microsoft invested $13B over the course of a couple of rounds for a revenue share, a profit share, an IP share, and, for a period, exclusivity on signing OpenAI as a customer on the cloud compute side.
This essentially saw Microsoft recoup most of their investment in a couple of years by way of OpenAI spending a lot of money on Microsoft’s cloud product, Azure. The proof is in the pudding. This quarter, Azure blew everything out of the water. Consensus estimates for growth at Azure were 34.5%; they came in at 39%.1
This is the reacceleration of a massive revenue base at scale, in the most important industry of our time. Azure’s run-rate is now $86B, growing 39% YoY. Scale. Growth.
Now we get to the fun part. OpenAI is on an absolute heater, smashing through previous revenue expectations in an epic manner. At the end of July, The Information reported:
“OpenAI roughly doubled its revenue in the first seven months of the year, reaching $12 billion in annualized revenue. That figure implies the ChatGPT maker is generating $1 billion a month, compared to about $500 million a month at the start of the year. The growth comes as the company logs roughly 700 million weekly active users for its ChatGPT products used by both consumers and business customers, up from 500 million weekly active users OpenAI said all of its products had in late March.”
This is parabolic growth at a scale I don’t think we’ve seen in the market. I even asked ChatGPT to come up with some comparisons, and while it did a good job with the list, none of them quite cut it.
Speaking of moving goalposts, as The Information reported last week: “OpenAI’s existing technology and products are proving so valuable to businesses that the company projects to be generating $20 billion in annualized revenue by the end of the year, up from around $6 billion in annualized revenue at the start of the year.” They had zero revenue three years ago. The previously reported revenue projections below looked crazy. They don’t look so ridiculous now…
If OpenAI can pull off these revenue figures, and they figure out the restructuring and IPO, I believe they will be in the top 10 market-cap companies before the end of the decade.
Bringing It All Together
These last couple of months have taught us several things:
- Capex numbers were too low – We were far too low on our 2025 numbers, which are now getting pushed up. I believe 2026 may have a similar trend, but maybe not quite at the same magnitude.
- Opex is shifting – AI researchers get professional-athlete-sized deals while the bottom/middle layer gets automated away.
- AI revenue is here – Microsoft and OpenAI have real and parabolic AI revenue growth.
- We’re still just getting started – We believe that as AI gets integrated into more and more workflows, we’ll still need more compute, which means more infrastructure, which means more power.
We believe it’s time to get on board with AI. Those who don’t adapt may end up like a hockey fan outside the stadium trying to buy a ticket to a game that’s already in the third period.
Strong Convictions. Loosely Held.
- Nicholas Mersch, CFA
Notes
- Microsoft's EBIT per employee in FY25 was ~40% higher than in FY23
The content of this document is for informational purposes only and is not being provided in the context of an offering of any securities described herein, nor is it a recommendation or solicitation to buy, hold or sell any security. The information is not investment advice, nor is it tailored to the needs or circumstances of any investor. Information contained in this document is not, and under no circumstances is it to be construed as, an offering memorandum, prospectus, advertisement or public offering of securities. No securities commission or similar regulatory authority has reviewed this document, and any representation to the contrary is an offence. Information contained in this document is believed to be accurate and reliable; however, we cannot guarantee that it is complete or current at all times. The information provided is subject to change without notice.
Commissions, trailing commissions, management fees and expenses all may be associated with investment funds. Please read the prospectus before investing. If the securities are purchased or sold on a stock exchange, you may pay more or receive less than the current net asset value. Investment funds are not guaranteed, their values change frequently, and past performance may not be repeated. Certain statements in this document are forward-looking.
Forward-looking statements (“FLS”) are statements that are predictive in nature, depend on or refer to future events or conditions, or that include words such as “may,” “will,” “should,” “could,” “expect,” “anticipate,” intend,” “plan,” “believe,” “estimate” or other similar expressions. Statements that look forward in time or include anything other than historical information are subject to risks and uncertainties, and actual results, actions or events could differ materially from those set forth in the FLS. FLS are not guarantees of future performance and are, by their nature, based on numerous assumptions. Although the FLS contained in this document are based upon what Purpose Investments and the portfolio manager believe to be reasonable assumptions, Purpose Investments and the portfolio manager cannot assure that actual results will be consistent with these FLS. The reader is cautioned to consider the FLS carefully and not to place undue reliance on the FLS. Unless required by applicable law, it is not undertaken, and specifically disclaimed, that there is any intention or obligation to update or revise FLS, whether as a result of new information, future events or otherwise.