There was a lot of good news this week, yet the AI trade shrugged it off. TSMC and ASML both guided September revenue higher than the Street. IBM’s poor results (down 26% this week) were actually a positive read on how early we are in AI, showing a new dynamic in AI infrastructure spending: it’s so powerful that it is pulling dollars away from IBM’s traditional hardware and software businesses. Separately, AAPL shares received a boost, up 20% over the past three weeks vs. the Nasdaq up 1%, on early signs that the new Siri, which marks the start of personalized AI beta, finally shows that the company has chops in AI.

All that said, with the exception of Apple, the market did not reward all the good news.

The Innovator Deepwater Frontier Tech ETF (LOUP), which has a thematic AI weight of 70%, was down 6% this week, TSM was down 9%, and ASML was up 1%, compared with the Nasdaq, which was down 2%. The exception was Apple, up 4%. The reason for the disappointing stock reaction: the trend that emerged in April continues. Beats and raises are no longer enough to move AI stocks higher. The question is what will change the tide.

TSM:

TSM’s June revenue was already reported earlier in the month, so the earnings report was all about the guide. September guidance implies 42% y/y growth versus the Street’s 36% expectation. Full year 2026 guidance moved from 37% to the low 40s, and the math behind that raise implies the December quarter accelerates from the 42% consensus to growth of 50% or more.

The negative is that margins are topping out. The recently reported 68% gross margin compares with the September quarter margin guidance of 66% (midpoint), which suggests the first margin decline since June 2025 and the first decline of 1% or more since December 2023. In other words, for the past year, margins have been moving higher. Now they’re moving lower. My sense is that they will actually print 67% margins in September, given the favorable pricing environment, which would be positive compared with current expectations calling for 66%.

The company also raised capex for CY26 by 15%, which means it’s now expected to grow this year by 59%. Some investors read that as a negative because it points to endless spending. We see it as a positive, given that increased TSM production capacity, especially in the US, makes it next to impossible for a competitor to gain share. Management also guided for capex to be up “significantly” over the next few years. While the word “significantly” lacks the detail that would be helpful to model the spend, the Street’s current estimates, calling for capex growth in CY26 of 59%, CY27 of 23%, and CY28 of 18%, imply that total capex in the years CY23 to CY25 will double compared with CY26 to CY28. That, in my estimate and the Street’s, feels in line with “significantly.” I’m not expecting a bombshell update to capex that will change Street estimates when the company reports in September. I do expect the capex growth numbers to inch higher. In other words, the steepest part of the curve is behind us.

IBM:

IBM was an early indicator of legacy software budgets being redirected to fund AI development. Software, at 45% of revenue, grew 11% in the March quarter, was expected to grow 10% in June, and grew 5%. In the world of IBM results, which have been steady over the past 20 years, that small miss was actually massive. Shares of IBM traded down 25% the following day, marking their worst day of trading. That miss tells us two things: the desire among Fortune 500 companies to build AI infrastructure is growing, and the risk that traditional software will be displaced is also growing. At its core, software provides humans with an abstraction layer that allows them to interact with the machine. Bots or agents don’t need the layer; they just talk to each other. Large enterprises are actively sacrificing core, historic products to build out their AI future.

ASML:

Despite a naturally lumpy business driven by machines that sell for $250M to $400M each, ASML grew June quarter revenue by 21% versus expectations of 16%, an acceleration from 13% growth in March. September guidance implies 60% y/y growth versus the Street’s 51% expectation. Full year 2026 guidance moved from 22% to 38% y/y, and the math behind that raise implies the December quarter accelerates from the 40% consensus to growth of 54%.

The stock still fell 1% following comments in the earnings call that ASML expects to build around 85 EUV machines in 2027, below analyst expectations of 90 to 95. But ASML didn’t say demand is fixed at 95 units they can’t meet. CFO Roger Dassen said 85 reflects ‘the balance between demand and supply as we see it today,’ and that if customers ask for considerably more, ASML would revisit the supply chain to see what else can be done, just as they’ve done in recent quarters. So 85 is management’s current best estimate, not a hard ceiling. The big picture still holds: even a cautious, supply-balanced number like 85 points to continued strong growth at the company that makes the machines that make the chips, suggesting the hardware buildout is still in its early stages.

Apple:

This week, the iOS 27 beta, released in the US and featuring the newly overhauled Siri, is the first example of personalized AI that is a must have, and I believe it will drive a rerating of AAPL shares. I’ve been using it for the past three days, and while I find that the new Siri is slow and takes extra juice, the personalized insights between messages and mail are more powerful than I expected. I feel it’s an unlock in terms of the utility of the phone. This all matters because it finally reveals to investors, who have been waiting two years (since WWDC 24) for Apple to show that it has some AI muscle, that this muscle can be applied to Apple’s pursuit of a unique opportunity in personalized AI. I expect Apple’s progress with personalized AI will start a multiyear hardware upgrade cycle inspired by AI from CY27 through CY30. Shares of AAPL rallied on the news and finished the week up 4% vs. the Nasdaq, which was down 2%.

Now all eyes turn to big tech earnings over the next two weeks.

The framework for assessing the health of the AI trade has become a well-traveled road, centered on Cloud and capex outlook. Both carry equal weight in the conversation. We expect consensus CY27 hyperscaler capex growth estimates to increase from the current 23% y/y to 37%. As for Cloud, we expect upside in June for Google Cloud and AWS growth, and an in-line print for Azure. The bigger question is less about the magnitude of the upside and more about whether investors will reward the AI trade based on the improved outlook. My sense is yes, given these higher expectations should drive AI infrastructure earnings growth that more than offsets multiple compression related to the law of large numbers. P.S. Oracle, the emerging cloud entrant with 4% market share, won’t report its August quarter until mid-September.

Within the hyperscaler capex conversation, the key metric is next year’s growth. As it stands, the Street is looking for 23% growth in CY27. That said, whisper expectations have moved higher following Google’s $85B and Amazon’s $25B capital raises, which will be spent predominantly late this year and into next year.

For the sake of simple math, if you assume $80B of Google’s $85B raise goes to next year’s capex and $20B of Amazon’s $25B raise goes to capex, hyperscaler CY27 capex growth lands at 37% y/y. Additionally, I expect a modest increase in Meta’s CY27 capex outlook, which should push its growth to 25% from the current expectation of 21%. The biggest narrative risk in the capex conversation is Microsoft, where I believe management will guide investors to an unchanged 26% growth rate for next year. If three of the four hyperscalers raise their outlook, I believe the AI infrastructure narrative will remain intact. The buildup is outlined below:

The likely standout in June cloud reporting will be Google Cloud, as the company has been the most aggressive among the big three at building out infrastructure, and separately, leveraging its family of models to draw in customers. This translates to our expectation that Google Cloud revenue will hit a staggering 70% y/y in June, compared to Street estimates calling for 64%.AWS will likely continue its streak of outperformance, driven by demand still outpacing supply despite continued buildout, yielding 34% growth y/y, versus the Street at 32%.As for Microsoft, the trend of them posting the least upside should continue as we expect growth to come in line with expectations, up 40% y/y.The bear case is that these results will be viewed as positive, but not enough. My sense is that commentary around September will reassure investors that we’re still early in the AI cloud growth story:

Over the past three years, the bar for AI infrastructure companies has kept rising, and those companies have continued to blow past increasingly high expectations. Each time the bar moves higher, however, it makes the following year’s growth outlook more difficult.

Take Nvidia as an example. A year ago, analysts expected the company to grow revenue by roughly 20% in CY26. Today, they expect revenue to grow by 81% this year. That sharp increase in growth expectations has been met with NVDA shares rising 17% over the past year, compared with a 22% gain for the Nasdaq.

In other words, Nvidia’s multiple has declined over the past year. A year ago, the company traded at 30 times out-year earnings, based on CY26 estimates. Today, shares trade at 16 times out-year earnings, based on CY27 estimates.

The core issue is that investors are bracing for slower growth two to three years from now. That raises the question: What should investors be willing to pay for that slower growth?

For starters, I agree with the premise. Eventually, the law of large numbers will catch up with these companies, and annual growth will decline from roughly 100% to closer to 15% in the years ahead. Those declining growth expectations suggest that valuation multiples will continue to compress.

The good news is earnings. I expect earnings to continue growing by more than 20% annually for the next several years. If that proves true, it would lay the groundwork for AI infrastructure stocks to continue working, as earnings growth more than offsets the decline in valuation multiples.

Disclaimer