AI Is Building the Future on Chips. But What If the Bottleneck Is Hiding Under the Data Center Floor?

Marvell is talking about $10 billion in custom chips by 2029. But the real AI story may start not with the processor, but with electricity, cables, and copper.

AI looks light, almost invisible: a model in the cloud, an answer in seconds, a new service with no physical shape. But behind that “invisibility” sits very heavy infrastructure. Servers need power. Processors need to be connected. Data centers need cooling. And all of it has to run without interruption.

Marvell’s latest outlook shows the scale clearly. The company expects its custom chip business to exceed $10 billion in revenue in fiscal 2029, while data center revenue this year could grow by about 50%. Reuters also reports that major U.S. tech companies may spend more than $700 billion on AI infrastructure this year, up from roughly $400 billion in 2025.

The numbers are impressive, but they open a quieter question: can the physical world keep up with the digital one? Chips are only one part of the machine. Around them are high-speed interconnects, power distribution, transformers, cooling systems, substations, backup power, and thousands of meters of cable.

That is why the AI boom is slowly becoming more than a story about software and semiconductors. It is becoming a story about energy, electrification, grid capacity, and critical minerals. Copper does not look dramatic in this story, but without it the basic logic does not work: more computing means more electricity, more connections, and more conductivity.

This is where the supply chain becomes more important than it first appears. A new data center can be announced quickly. A new copper source cannot. It has to be found, tested, drilled, financed, and proven economically meaningful. That takes years of work, not one earnings call.

NovaRed Mining operates in the quieter part of this story — where the future supply pipeline is still being formed. Its British Columbia focus and use of MetalCore for AI-assisted mineral evaluation are interesting not as a loud AI label, but as an example of how copper exploration is also gradually moving into a data-driven phase.

Maybe the market is still looking at AI from too high up — at the cloud, the models, and the chips. But if you dig lower, beneath that story sits a much simpler question: will there be enough metals, electricity, and infrastructure to actually switch this future on?