$340 billion a year. That’s the annualized pace of US large-computer imports as of March 2026, a figure that arrived quietly in a trade data release and didn’t get nearly the headline it deserved. Not because it’s boring — because it’s almost too clarifying. The AI boom isn’t primarily a software story. It never was. It’s a logistics story, a real estate story, a power story. The chips have to live somewhere, and getting them there costs more than most people are tracking.
That number lands differently when you read it alongside two other things that happened this week: Google and Blackstone announcing a massive joint venture to build AI cloud infrastructure, and the quiet news that $MSFT’s custom Maia chips may soon run Anthropic’s Claude models. Two different companies, two different supply-chain strategies, both arriving at the same conclusion — that the commodity in shortest supply isn’t intelligence. It’s compute. Physical, power-hungry, capital-intensive compute.
The Blackstone partnership is the more interesting signal of the two. Blackstone is a real estate and private equity firm. The fact that one of the world’s largest alternative asset managers is now co-building AI cloud infrastructure with $GOOGL tells you something about where the returns are being priced. Not in the model. In the ground the model runs on. Data center real estate. Power capacity. Cooling systems. The boring, unsexy plumbing that makes everything else possible. What Blackstone brings is capital and site acquisition expertise; what Google brings is engineering and hyperscaler relationships. Neither would say this out loud, but the subtext is: we both think the infrastructure layer is where value accretes, and we’re moving before the next wave of demand makes it too expensive to move.
Microsoft’s Maia play is subtler but points the same direction. The premise of custom silicon is that you trade generality for efficiency — you give up the ability to run anything and gain the ability to run one thing extremely well, at a fraction of the cost per token. Amazon has Trainium. Google has TPUs. Now Microsoft is bidding to power Anthropic’s workloads with its own chips. Every major cloud provider is building its own inference silicon, which means the hyperscaler war of the next five years isn’t going to be won on GPUs alone. It’ll be won on who owns the most efficient path from prompt to response.
This is where the CoreWeave vs. Nebius question comes in — and it’s a genuinely interesting one to sit with, not because either answer is obvious, but because the framing itself reveals something. We’re now at the stage where “which AI infrastructure stock” is a retail-accessible question, which means the institutional money has already moved. The smarter question isn’t which is a better buy. It’s whether the independent infrastructure play survives the next five years as hyperscalers build their own everything.
Bill Gates writes in Source Code that the circumstances shaping his career were “mostly out of my control” — the unearned privilege of being born where and when he was, with access to machines when access to machines was the whole game. The honest read on this week’s infrastructure moves is similar. Google and Blackstone aren’t smarter than their competitors. They got to scale earlier, built relationships earlier, and now those relationships are compounding into positions that will be very hard to replicate. Infrastructure moats don’t announce themselves. They just get wider.
Which brings us to the IPO signal — the one that’s actually worth watching as a leading indicator. SpaceX and OpenAI both reportedly preparing for massive public offerings. The analyst take is that mega-IPOs cluster near market tops, which is true as a historical pattern. But the more nuanced read is: what kind of top? A valuation top driven by speculation, or a structural top where the underlying build-out has run ahead of near-term demand? Those are different problems with different time horizons. The former reverses fast. The latter takes years to resolve.
The market can stay irrational longer than anyone expects — but the imports number keeps running. $340 billion a year in large computers suggests the people building this infrastructure aren’t betting on a blip. They’re betting on a decade. The IPO chatter is noise about sentiment. The chip imports are signal about conviction.
What most observers are missing is that the infrastructure wave and the frothy IPO narrative are both true at the same time, and they’re not in contradiction. The build-out is real and capital-intensive and durable. The valuations placed on top of that build-out may or may not survive contact with reality. You can believe in the foundation and still be skeptical of the price being charged to stand on it.
The infrastructure IS the argument — for the companies building it, for the capital flowing into it, for every model that runs on top of it. The software gets all the headlines. The plumbing gets all the returns.
Leave a Reply