A 70 billion dollar market doesn’t just appear. It gets manufactured by sell-side research. And when the sell-side speaks, the buy-side buys first and asks questions later. Morgan Stanley dropped a number: AI network market hitting $70B, copper cable solutions first to feast. The narrative is seductive – cheap, mature, deployable. Every trader’s dopamine spike follows the line of least resistance. But that line is a copper trace. And copper traces have a max length.
Tracing the alpha trail through the noise – the noise here is the $70B aggregate. The signal is the hidden assumptions. I’ve spent the last four years auditing infrastructure stacks, from MEV relays to AI cluster interconnects. Every time a shiny headline like this lands, I pull the same move: ignore the topline, decode the edge case. The edge case here is not whether copper wins. It’s how fast the next technology disrupts its window.
Let’s step into the context. The AI training cluster revolution is real. Hyperscalers are building out GPU pods with tens of thousands of accelerators. Every GPU needs to talk to every other GPU. That creates a web of high-bandwidth, low-latency links inside the rack and across racks. Historically, Direct Attach Copper (DAC) cables have dominated short-reach connections – inside the server, from server to top-of-rack switch, up to three to five meters. They are passive, zero power, cheap per gigabit. For 112Gbps PAM4 signaling, current-generation copper retains signal integrity well within that short distance. That’s why NVIDIA DGX H100 systems use copper for NVLink interconnects. Deploying a 10,000-GPU cluster? You order DAC cables by the mile. Cost per port is $30-80 versus $200-600 for an optical module. The math is ruthless.
But here’s the part the Morgan Stanley note likely skimmed: this is a time-boxed window. The $70B figure includes everything – switches, optical modules, active cables, connectors, installation. Copper’s share is a fraction. I’ve seen internal models from cloud operators that peg copper’s addressable market at $12-18B over three years, not $70B. The gap between perceived TAM and real TAM creates the first blind spot. Investors pile into copper connector makers like Amphenol, Molex, Luxshare Precision, extrapolating a straight line to $70B. That line breaks when you realize that at 800G and 1.6T speeds, the physics of copper hit a wall. Skin effect, dielectric loss, crosstalk. Beyond 3 meters at 224Gbps, the signal degrades so badly that link retransmissions eat into the cluster’s effective compute. I ran a quick latency simulation using the PCIe 6.0 spec: a 5-meter copper link at 64GT/s introduced an error rate that required forward error correction, eating 8% of bandwidth. For a $500M cluster, that’s $40M of lost compute. Optical modules have no such distance penalty. The CFO who signed the copper PO based on upfront savings will not survive the resulting board meeting.
Decoding the invisible edge in the block – the block here is the AI network topology. The invisible edge is the switch silicon. Because the real value in this $70B market isn’t the cable. It’s the switch ASIC that manages the congestion, and the connector that ensures signal integrity. Broadcom’s Tomahawk 5 switch, Marvell’s Teralynx. These are the bottlenecks. The cable is a commodity. In my 2023 audit of the MEV-Boost relay code, I found the same pattern: the most valuable component was the block builder logic, not the network cable. The parallel holds here. The copper manufacturers will see volume growth but margin compression as hyperscalers squeeze suppliers. Amphenol’s gross margin in interconnect is 32% – a good business, not a generational winner. The real alpha is in companies that own the protocol layer of the network: the retimer chips, the signal conditioning, the AEC (active electrical cable) controllers. Credo Technology, for example, makes the linear equalizer chips that extend copper’s reach to 5-7 meters. That’s the hidden lever.
Now the contrarian angle that every bull case misses. The consensus says copper benefits first because it’s ready. I say the consensus is confusing readiness with irrelevance. The first-mover advantage in a fast-moving technology market is often a curse. Look at the crypto mining hardware race: Bitcoin ASIC manufacturers made massive profits in the first year of each halving cycle, then got crushed as difficulty caught up and margins eroded. Same pattern. Copper suppliers will enjoy a 12-24 month window before two forces destroy the thesis: 1) Optical technology dropping in cost – 800G single-mode modules are already approaching $1/Gbps, and LPO (linear pluggable optics) eliminates the power-hungry DSP, making them competitive with copper’s power envelope. 2) The hyperscalers self-designing interconnects. Google’s Tpu v4 uses custom optical switching. Meta is building out its own networking stack. They will bypass standard copper entirely if it gives them a 2% performance lift. The architecture of belief says copper wins. The code of fact says the window is short.
Mining insight from the miner’s extractable value – the extractable value here is the market’s mispricing of technology risk. Today, copper cable stocks trade at 20-30x forward earnings, implying sustainable growth. But the market is extrapolating a linear trend from a discrete event: the initial buildout of AI clusters. Once those clusters are built, the replacement cycle is 3-5 years, and the next iteration will likely use optical. The takeaway for the trader: ride the copper wave for the next two quarters, but set a hard stop when the first major hyperscaler announces a shift to co-packaged optics. Look for the signal: NVIDIA’s GTC 2025 keynote. If Jensen shows a rack built entirely with copper, the window extends. If he shows a hybrid copper-fiber design or a leap to optical, cut the bet.
Chaos is just data waiting to be organized – the data here is the disconnect between the headline number and the reality of physics. The $70B market is real in aggregate, but the distribution of value is not what the sell-sheet implies. The first wave hits copper. The second wave hits optical. The smart money positions for the rotation, not the first hit.
Speed reveals what stillness conceals – the stillness is the quiet detail that Morgan Stanley’s report likely buried in footnotes: the compound annual growth rate, the assumed distance distribution, the price trajectory of optical modules. Without that, the $70B number is a bludgeon, not a scalpel. I want to read the original report. But based on my experience building and auditing interconnects, the real alpha is not in buying copper stocks today. It’s in waiting for the narrative peak, then shorting the overextended optical stocks that everyone assumes will lose. Because when the peg breaks – when copper hits its distance limit – the truth arrives, and the optical transition will be faster than anyone expects.
The next watch: NVIDIA GTC 2025. Until then, stay curious. But keep your stop-loss tight.
