The Trillion-Dollar Machine Hiding Behind AI
Every time you ask an AI assistant a question, an enormous physical machine goes to work on your behalf. Not a metaphorical machine. A literal one: a warehouse full of server racks, each one packed with more than 4,500 chips built from roughly 20,000 individual slivers of silicon, priced between $1.5 million and $4 million per rack. A leading AI data center can house up to 10,000 of them.
A new joint report from Deloitte and the Semiconductor Industry Association takes that machine apart, piece by piece, and what it reveals matters to anyone thinking about where the economy is headed over the next decade.
The scale is difficult to overstate.
The report projects that revenue from semiconductors deployed in AI data centers will exceed $1.2 trillion by 2028, an almost tenfold increase over five years. Behind that number sits a projected $4.0 trillion in cumulative AI data center investment from 2023 through 2030, with up to $2.8 trillion of it flowing to
semiconductors and hardware. NVIDIA's recently announced financing ecosystem, built to support up to $500 billion in AI infrastructure investment with backing from firms including BlackRock, Blackstone, Goldman Sachs, KKR, Apollo, and Brookfield, gives a sense of how seriously the world's largest capital providers are treating the buildout.
The value hides in unexpected places.
Semiconductors account for more than 95% of an AI server rack's content value, and the famous AI accelerator chips capture the largest share. But the report's teardown shows a machine that depends on far more than its stars. Memory and storage account for more than 20% of a rack's semiconductor value, because
modern AI moves oceans of data and starves without fast access to it. Networking chips knit thousands of processors into one coordinated system. Power and cooling components keep the whole thing alive. And here is the report's most striking inversion: over half the chips in an AI server cost less than $10 each, while just 3% of high-value chips drive the majority of the machine's value. The trillion-dollar buildout runs on both.
The workload is changing shape.
Most AI chip demand so far has come from training, the phase where models learn. The report describes a shift now underway toward inference, the phase where trained models actually do their work, answering questions and powering
applications. Industry estimates cited in the report project inference growing from roughly 20% of AI compute demand in 2024 to as much as 80% by 2032. Different work favors different chips, which means the beneficiaries of the next phase may not be identical to the beneficiaries of the last one.
What AI infrastructure investment means for long-term investors.
At Dillow Wealth Management, we read reports like this one for the same reason
we encourage our clients to: not to chase a headline, but to understand the structure beneath it. A few observations we take away:
Transformations of this scale are measured in decades, not quarters. The report frames today's buildout as the foundation of a long infrastructure cycle, the kind of shift that reshapes industries gradually and then all at once.
Value in a technological transformation rarely lives in one place. The same machine that makes headlines for its most expensive chip also depends on memory, power, networking, and cooling, an entire ecosystem of businesses most people have never heard of.
And concentration cuts both ways. When 3% of components carry most of the value, the opportunity and the fragility live close together. That is precisely the kind of tension a disciplined, diversified, long-term strategy is built to navigate.
We are living through one of the most productive periods of innovation in modern history. The wisest response is neither euphoria nor anxiety. It is understanding.
Want the full picture?
We are sharing the complete 45-page Deloitte/SIA report with clients and friends of the firm. Email us at tdillow@dillowwealth.com with the subject line "AI Report" and we'll send it over, along with our summary of the key takeaways.
This material is provided for informational and educational purposes only and does not constitute investment advice or a recommendation of any security or strategy. Projections referenced are those of the cited report and its sources, not of Dillow Wealth Management. Dillow Wealth Management, LLC is a registered investment advisor. Please visit dillowwealth.com for important disclosures.
