In this round of AI stock gains, the reason has changed. In the first half of the year, the market was paying for “models getting stronger.” Now, the reason for paying is the visibility of capital expenditures. Oracle ($ORCLB )’s latest quarter cloud infrastructure revenue jumped 121% year over year to $7.4 billion, with remaining performance obligations stacked up to $664 billion (or $664.0B). This fiscal year’s capital expenditure guidance points to a range of $90–95 billion—nearly double last year’s figure. These aren’t profits; they are orders that need to be spent.
Even more noteworthy is the structure. UBS estimates that global AI capital spending will rise from about $900 billion in 2026 to $1.2 trillion in 2027, and roughly two-thirds of compute demand comes from inference rather than training. PwC projects that cumulative investment in global data centers through 2050 will total $3.16 trillion. Translate it like this: this isn’t a product cycle—it’s an infrastructure cycle.
So when I ask, “What other opportunities are there?” I lean toward placing the answer upstream rather than at the model layer. Winners at the model layer will keep changing, but every card has to be powered, connected to the grid, housed in racks, and paired with switching chips. These links collect a toll; order visibility is measured in years. $NVDAB and $AMDB are both rising on the same theme, but one is the landlord and the other is the chaser—valuation tolerance is completely different. The truly fragile part is pure compute leasing: buying cards at high prices and renting them by the hour. Once supply catches up with demand, gross margin is the first thing to get squeezed.
The risk is also here. Someone has already pointed out that if capital expenditures slow down, the market is pricing a high multiple for slower growth. That’s true—except no one wants to be the first to hit the brakes right now. It’s more like a game of chicken: whoever stops first is the one that gets eliminated.
One question to leave you with: if you can only keep one AI position—models, compute leasing, or power and grid interconnection—which one would you choose? I want to hear different answers.
#AI stocks keep rising—what other investment opportunities are there?
Even more noteworthy is the structure. UBS estimates that global AI capital spending will rise from about $900 billion in 2026 to $1.2 trillion in 2027, and roughly two-thirds of compute demand comes from inference rather than training. PwC projects that cumulative investment in global data centers through 2050 will total $3.16 trillion. Translate it like this: this isn’t a product cycle—it’s an infrastructure cycle.
So when I ask, “What other opportunities are there?” I lean toward placing the answer upstream rather than at the model layer. Winners at the model layer will keep changing, but every card has to be powered, connected to the grid, housed in racks, and paired with switching chips. These links collect a toll; order visibility is measured in years. $NVDAB and $AMDB are both rising on the same theme, but one is the landlord and the other is the chaser—valuation tolerance is completely different. The truly fragile part is pure compute leasing: buying cards at high prices and renting them by the hour. Once supply catches up with demand, gross margin is the first thing to get squeezed.
The risk is also here. Someone has already pointed out that if capital expenditures slow down, the market is pricing a high multiple for slower growth. That’s true—except no one wants to be the first to hit the brakes right now. It’s more like a game of chicken: whoever stops first is the one that gets eliminated.
One question to leave you with: if you can only keep one AI position—models, compute leasing, or power and grid interconnection—which one would you choose? I want to hear different answers.
#AI stocks keep rising—what other investment opportunities are there?