The Full Picture The Demand Is Not in Question. The Layer Is. |
For an hour or so on Wednesday afternoon, the headline writers had a clean run at it. |
Revenue of $109.9 billion, up 22% year-over-year. Earnings nearly double consensus. Google Cloud growing 63% — faster than Microsoft Azure or AWS in their most recent quarters. |
The bear case that AI chatbots would erode Google Search looked thinner by the hour. |
Then the after-hours desk got hold of the second number. Capex guidance for 2026 had been raised to $180–190 billion. The stock paused. |
Microsoft, Meta and Amazon reported within hours, and the same pattern repeated. Microsoft guided to roughly $190 billion in 2026 spending, citing $25 billion in higher component pricing alone. Meta lifted its range to $125–145 billion and watched its shares fall 6% after-hours. Demand was not the issue. Cost was. |
Combined, the four hyperscalers will spend up to $725 billion on AI infrastructure this year. That figure is larger than the GDP of Switzerland. |
The reaction was not a verdict on AI demand. Demand is settled. Google Cloud's contracted backlog nearly doubled quarter-over-quarter to more than $460 billion. |
The reaction was a verdict on margin compression — on whether the build will earn its cost of capital at the prices currently being paid for the inputs. |
Most portfolios are not positioned for what comes next. |
The AI trade, as it is typically expressed, lives in the application and model layer: the chatbots, the productivity suites, the consumer interfaces. That tier is real, and it is growing. |
It is not where the binding constraint sits. The constraint sits one layer below — in silicon, data centres, dispatchable power, and the physical networks that move the data. |
Where the Capital Is Actually Flowing |
Of that $725 billion, very little ends up at software firms. |
It flows to chip designers, foundries, memory suppliers, grid operators, data-centre real estate, cooling specialists — and, increasingly, to the companies that own launch capacity and orbital bandwidth. Capital is moving to the parts of the chain that are physically constrained, not the parts that are conceptually exciting. |
Partner Perspective Elon Musk is about to take SpaceX public as part of his plan to unlock the full power of artificial intelligence. Elon is predicting this will help unleash a $1 quadrillion new wealth wave. That would be enough to send a check for $2.8 million to every single man, woman, and child in America. That's how big this opportunity is. Click here to get the details and see how to claim a stake… starting with just $500. |
|
Within that infrastructure tier, one set of assets has been treated as adjacent rather than central: launch systems and the satellite networks they enable. |
That treatment is worth re-examining. |
Inference workloads no longer sit only in a handful of U.S. data centres. They are being distributed across geographies where fibre is thin, and where enterprise customers will pay a premium for connectivity that does not depend on a single national grid. The economics of getting payloads into low-earth orbit have collapsed by an order of magnitude over the past decade. |
SpaceX, the dominant private operator in commercial launch, is reported to be moving toward a public listing. |
Whether the timing holds is one question. Whether the layer it occupies is investable is another — and that one is no longer in serious dispute. |
A useful frame, in closing. |
Every previous capital cycle of this scale — railroads in the 1870s, electrification in the 1920s, fibre in the 1990s — produced more durable returns for the owners of the physical layer than for the operators built on top of it. The pattern is unusually consistent. The names that captured the public imagination were rarely the names that compounded over decades. |
The application companies tend to be the most visible. The infrastructure compounds. |
Investors who treat AI as a software story will get a portion of the trade right. Those who recognise it as primarily an infrastructure story — and who can identify where in that infrastructure supply is genuinely constrained — are likely to capture the more durable share. Google's quarter did not change the calculus. It only made it harder to ignore. |
