Green Daisy
back to blog
the half-trillion dollar compute bet
agents
founders
gpu
architecture

the half-trillion dollar compute bet

Brian Craighead

brian craighead

ai architect & cto, green daisy

Big tech earnings just wrapped, and the market has reached peak arms-race hysteria. Hyperscalers are pacing toward a combined half-trillion dollars in annual AI infrastructure spend.

Five hundred billion dollars. To put that in perspective, that is roughly the GDP of Austria, funnelled directly into custom silicon, server racks, and nuclear-backed data centres. The conversation is no longer about which model tops the Hugging Face leaderboard. It is about power purchase agreements, transmission lines, and whether the municipal grid can survive a server farm pulling multiple gigawatts.

It turns out the digital revolution looks remarkably like 19th-century railway construction.

The Capital Tsunami

When balance sheets representing trillions of dollars in aggregate enterprise value pour that kind of liquidity into raw compute, two economic inevitabilities follow.

First, the cost of raw inference collapses. Infrastructure at scale must be monetised, triggering a ruthless race to the bottom for token pricing. Commoditised compute is getting cheaper by the week, subsidised entirely by the cloud oligopoly.

Second, the moat has shifted. Owning a foundation model is no longer the flex it was in 2022. It is becoming basic utility infrastructure, like municipal water or broadband. The real, defensible alpha belongs to the operators who wire these commoditised models into complex, mission-critical workflows that end users will actually pay for.

At Green Daisy, we see this for what it is: a massive margin transfer from the infrastructure layer down to the application tier. You do not need a ten-billion-dollar capex budget or Nvidia's priority queue to build enduring software. You need sector expertise, high product taste, and the discipline to turn raw intelligence into finished utility.

Where the Alpha Lives

Every founder staring at this frenzy needs to ask one question: when the foundational layer is heavily subsidised by Big Tech's capex, where does the margin pool settle?

It settles in the messy, high-friction edges. It accumulates in vertical enterprise agents that resolve revenue-blocking tickets, proprietary industry workflows, and deep domain tools that generic chatbots cannot touch.

Every capital supercycle in modern financial history ends the same way: overbuild, margin compression, and oversupply. The railroads overbuilt, telecom overbuilt fibre in the late nineties, and hyperscalers are doing it now.

When the dust clears, history will not care how many H100s you hoarded. It will reward the companies that turned cheap tokens into indispensable enterprise software.

Are you burning capital trying to out-build the infrastructure giants, or are you riding the half-trillion-dollar wave they are funding for you?

share:

want to talk about this?

book a free clarity session and let's discuss how AI can work for your business.

let's chat