Spotted: Elon Crates at a U.S. Air Force Base

When Equinix announced its Inference Exchange program on September 2, the financial press focused on a 2% move in EQIX shares. Precious metals investors should be looking at something else entirely: the copper that fills every wall, floor, and ceiling of every data center Equinix now has reason to build.

What’s Driving the Market

Equinix Inference Exchange combines Nvidia Enterprise Reference Architectures, Together AI’s inference platform, and Equinix’s global infrastructure to optimize deployment speed, flexibility, and cost efficiency. The platform, available starting in Q1 2027, positions Equinix as an inference-as-a-service provider across more than 280 data centers across 77 metros, with 230 cloud on-ramps and more than 10,500 businesses interconnected on its exchange.

Critically, this is not a software announcement. Equinix data centers are built to support Nvidia’s Blackwell Ultra systems, including DGX B300 and GB300-class deployments. Every GPU rack at that density demands more physical infrastructure than anything the colocation industry has previously installed at scale.

Unlike traditional data centers, AI-oriented facilities require significantly more copper per megawatt of capacity to handle increased power distribution and advanced cooling needs. AI training data centers have been estimated at roughly 47 tonnes of copper per megawatt, compared to roughly 21 tonnes per MW for crypto-focused facilities. Equinix is not dabbling here. In February 2026, Equinix committed up to $700 million to support Hanley Energy’s new advanced manufacturing facility in Dundalk, Ireland, aimed at accelerating production of power equipment needed for next-generation data centers.

That gap in copper intensity is not a new discovery — it has been building pressure in the physical market for months. an earlier breakdown of why copper demand from AI infrastructure sent prices to record highs traces how mining supply has struggled to keep pace with the density requirements now being locked in by announcements like Equinix’s. Understanding that backdrop makes the scale of the inference buildout’s copper pull considerably easier to quantify.

The Investment Opportunity

Copper sits at approximately $6.60 per pound as of September 5, 2026. The metal’s sensitivity to any supply disruption has rarely been sharper. Analysts have highlighted an export ban from the Democratic Republic of Congo that targets copper and cobalt concentrates, along with weaker output from Chile, where production fell 9.4% year-on-year in July due to severe weather.

The structural argument is straightforward. S&P Global has estimated copper demand from AI-powered facilities will average about 400,000 tonnes a year over the next decade, peaking at 572,000 tonnes in 2028. With mine development taking more than a decade, the AI-driven copper crunch could arrive sooner than expected. Equinix’s inference buildout accelerates that timeline by pulling enterprise workloads into colocation halls rather than hyperscaler campuses, spreading demand across dozens of markets simultaneously.

Equinix and Digital Realty (DLR) have both been positioning for higher-density deployments to accommodate GPU-intensive workloads. Both companies are copper consumers, not copper producers. The producers and royalty streamers exposed to tightening refined-copper supply are where the metals-specific opportunity lives: Freeport-McMoRan, Ivanhoe Mines, and royalty vehicles with copper streams in their portfolios deserve attention as inference buildouts at Equinix, Digital Realty, and CoreWeave-linked facilities accelerate procurement.

Of those names, Freeport-McMoRan is the most direct expression of the copper-demand thesis, and its most recent results offer a useful lens on how the company is actually performing against that backdrop. a detailed look at Freeport-McMoRan’s 24% earnings beat and the role Grasberg gold played shows which operational levers are driving outperformance — context that matters when sizing a position ahead of an anticipated copper demand surge.

Risks to Monitor

One nuance worth examining is the difference between inference workloads and training workloads in terms of electrical and physical infrastructure requirements. Training large AI models is orders of magnitude more power-hungry than running finished models at inference. If inference remains significantly less power-dense than training at scale, the copper uplift per rack may be lower than early projections suggest. Efficiency gains in chip design, particularly Nvidia’s roadmap beyond Blackwell Ultra, could also compress power draw per token over time.

Tariff uncertainty continues to encourage shipments into the US, pushing COMEX inventories toward record levels while tightening supplies elsewhere, which creates a bifurcated market that can mask underlying tightness. A resolution of US refined-copper tariff policy ahead of the scheduled January 1, 2027 start date for a potential phased duty could temporarily relieve pressure on the red metal.

Bottom Line

Equinix’s Inference Exchange is a colocation business announcing its single largest strategic pivot. The company’s 10,500-plus customers now have a reason to run denser, hotter, more copper-intensive workloads inside Equinix halls rather than elsewhere. Investors watching the AI infrastructure wave should not stop at the chip layer. The physical metal that connects every GPU to every watt of power is where the scarcity resides, and demand signals from Equinix’s buildout are as clear as any the copper market has seen this cycle.