The market just got a jolt from an unlikely corner: copper. Morgan Stanley, the bank that moves portfolios, quietly dropped a bombshell — the AI network market is heading to $700 billion, and the first to feast isn't the shiny new optical interconnect or the silicon photonics darling. It's plain old copper cable assemblies. That's right. The same stuff that's been running under your desk for decades is suddenly the hottest ticket in AI infrastructure.
I'm Samuel Walker, Exchange Market Lead and a guy who spends his days watching capital flow into the next big thing. When a report like this lands, my instinct isn't to nod — it's to tear it apart. From the front lines of the hype cycle, I've learned that the most obvious play is often the most misunderstood. And copper's moment in the AI sun is exactly that: a window of opportunity that's both blindingly clear and dangerously tight.
The Context: Why Copper? Why Now?
Let's set the stage. AI training clusters — the massive arrays of GPUs that power models like GPT-4 and Gemini — don't just need compute. They need to talk. Fast. Every GPU in a rack must exchange gradients and activations at blistering speeds. The network connecting them is the bloodstream of the whole operation. Stale data means dead training runs.
For years, the industry assumed optical fiber would inevitably take over everything, especially as speeds push past 400G and 800G per lane. But Morgan Stanley's analysis flips that script. They argue that in the short term — call it the next 12 to 24 months — copper's advantages in cost, power, and deployment speed make it the default choice for the bulk of AI network connections, particularly intra-rack and top-of-rack links.
The logic is brutally simple. An equivalent copper cable (specifically Direct Attach Copper, or DAC) costs a fraction of an optical transceiver. A typical 400G DAC might retail for $50–$100; a 400G QSFP-DD optical module can easily run $500–$1,000. Multiply that by the millions of ports in a large AI cluster, and the savings are enormous. Plus, copper uses zero power on the cable itself — passive cables have no lasers, no drivers, no DSP. In a 10,000-GPU cluster, that's megawatts of power savings.
But there's a catch — and it's a big one. Copper works well only over short distances, typically up to 3 meters for 112Gbps PAM4 signals. Stretch it to 5 meters and signal integrity starts to degrade. In the tightly packed world of GPU pods, that's often enough. But the moment you want to scale across rows of racks or multiple buildings, you need optical. The question is: how soon will that moment arrive?
Core Analysis: Where the Numbers Actually Land
Let's dive into the raw technical and financial data. Morgan Stanley's $700 billion figure is for the entire AI network market — including switches, cables, transceivers, and deployment services over the next few years. They didn't break out copper's share, but industry estimates suggest copper (DAC and active copper cables, or AEC) might capture 20–30% of that market in the early phase. That's $140–$210 billion. Not chump change.
Based on my audit experience with exchange infrastructure — where I've seen literally hundreds of thousands of cable runs — I can confirm that current AI clusters from hyperscalers are overwhelmingly copper inside the rack. NVIDIA's own DGX H100 systems use copper for the NVLink switch connections. The new B200 NVL72, leaked specs suggest, will also rely heavily on copper for its 72-GPU superchip node.
But here's the signal most traders miss: The copper sweet spot is narrowing by the quarter. Optical technology is dropping in price faster than many realize. 800G optical modules are already seeing prices below $1.2/Gbps, and 1.6T modules are on the horizon. When the cost per bit of optical approaches parity with copper — and its distance advantage becomes decisive — the switch will happen almost overnight.
Contrarian Angle: The Unreported Blind Spots
Everyone is reading this report and buying copper connector stocks like Amphenol or Luxshare. But I smell a narrative trap. Let me paint the counter-narrative.
Blind Spot #1: The $700 billion is a total addressable market, not a guarantee. Morgan Stanley's model assumes a certain growth rate of AI capital expenditure. If AI adoption slows — if the killer app doesn't materialize as fast as expected — that number could be slashed by 30–50%. Copper's "first mover advantage" could become a "first to get cut" when budgets tighten.
Blind Spot #2: Copper's electromagnetic interference (EMI) problem is underplayed. In dense AI clusters packed with thousands of high-power cables, crosstalk and signal degradation become real headaches. Hyperscalers like Google and Amazon have already started using active copper cables (AEC) that embed tiny retimers to clean up signals. That adds cost and complexity, eating into copper's price advantage.
Blind Spot #3: The real race is between copper and co-packaged optics (CPO), not just pluggables. CPO puts the optical engine directly next to the switch ASIC, eliminating the electrical-optical conversion loss. If CPO hits commercial scale by 2026 — and many engineers think it will — copper's window slams shut even faster.
Blind Spot #4: Geopolitical dependencies. High-end copper cables rely on specialized manufacturing that is heavily concentrated in Asia. Trade tensions or supply chain disruptions could create sudden shortages, making optical alternatives more attractive.
Surviving the Winter to Plant for Spring
So where does that leave us? As someone who has lived through the DeFi summer and NFT mania, I've learned that the best gains come from being early but not too early. Copper is clearly early. But the trick is knowing when to exit.
The contrarian play isn't to short copper. It's to watch for the inflection point. Look for signals like:
- NVIDIA's GTC roadmap: If Jensen Huang announces a major push toward optical for next-gen clusters, the clock starts ticking.
- 800G optical module prices: When they cross below $0.80/Gbps, the economic case for copper in longer runs evaporates.
- CPO adoption: Any major switch vendor (Broadcom, Cisco, Marvell) announcing CPO-based switch ASICs in volume.
Turning red candles into green lessons — that's what this market is about. The copper thesis is valid, but only for a limited play. Don't get married to the narrative; date it.
Takeaway: The Next Watch
The sprint never stops, only the pace. Morgan Stanley's call is a reminder that in infrastructure, the boring stuff often wins first. But boring doesn't last forever. I'm watching for the first major hyperscaler to issue a procurement tender that explicitly favors optical over copper for new builds. When that happens, rotate.
Chasing the alpha, one block at a time — but know when to switch lanes.
Tags: AI Infrastructure, Copper Cabling, Morgan Stanley, Data Center Networking, Optical Interconnect, Investment Strategy