The Copper Truth: Why AI's $70B Network Is Still Too Smart for Smart Money

Charts lie. Liquidity speaks. Hook The AI trade is a ghost chase. Over the last six months, the Magnificent Seven returned a modest 12%. AI tokens like Render or Fetch.ai -- 50x revenue multiples, no earnings. Meanwhile, a quiet cluster of stocks climbed 30%+. Amphenol. Luxshare. TE Connectivity. Copper connector makers. The crowd stares at GPUs. The real alpha is in the cable tray. This is not a coincidence. It's a structural mispricing rooted in a single Morgan Stanley projection: the AI networking market will hit $70 billion by 2027. And copper cables will eat first. Let's follow the data. Not the narrative. Context The Morgan Stanley report isn't public. But the thesis is clean: AI training clusters are scaling at an unprecedented rate. Clusters of 10,000+ GPUs require immense interconnects. The bottleneck isn't compute -- it's how data moves between chips, servers, and racks. The report breaks the networking market into three segments: copper (DAC/ACC/AEC), optical (pluggable transceivers), and co-packaged optics (CPO). The blunt conclusion: copper will claim the majority of the $70 billion in the first 12-24 months. Why? Technical maturity. Cost. Power. Deployment speed. Today's AI clusters, like NVIDIA's DGX H100 or the upcoming B200, rely heavily on copper. The NVLink system uses copper for direct GPU-to-GPU connections inside the server. The top-of-rack switches connect to servers via copper DAC cables -- passive, cheap, reliable at short distances. Optical modules exist but at 2-3x the price per port. For hyperscalers building out tens of thousands of nodes, the cumulative savings in power and CapEx are enormous. Morgan Stanley's claim is credible: the early phase of AI infrastructure buildout will be a copper fest. Core Let me dissect this from a trader's lens. I've spent years analyzing on-chain data, but the same principles apply to industrial supply chains: flow matters. The $70 billion figure is a TAM estimate. But what portion is copper? Morgan Stanley likely breaks it down into subsegments: 1) Passive DAC (direct attach copper) for <3m interconnects, 2) Active copper cables (ACC/AEC) for 3-7m, 3) Optical modules for >7m and campus interlinks. My reading: copper accounts for 30-40% of the TAM in the first two years, or roughly $21-28 billion. That's not small. That's a 3x growth from today's AI copper revenue. Technically, copper's viability hinges on signal integrity at 112Gbps PAM4. Engineering papers confirm that well-designed copper achieves bit error rates below 1e-15 at 3 meters. In an AI cluster running 1000 GPUs per rack, the total copper length per rack is around 200 meters. That's a lot of copper, but passive cables add near-zero latency and zero power. Compare to optical: a 400G QSFP-DD consumes 8-12 watts per port. A cluster with 2000 ports burns 20kW just on module power. Copper saves that entirely. At $0.10/kWh, that's $17,500 per year savings per rack. Scale that to 1000 racks, and you're talking $17.5 million annually. That's real money. Hyperscalers optimize for TCO. Copper wins the first round. During DeFi Summer 2020, I built my first arbitrage bot on Uniswap. I lost 20% in an hour due to slippage. That taught me execution risk. Networking in AI is analogous: latency and error rates are execution risks. Copper at short distances offers deterministic low latency. Optical modules introduce retiming and equalization delays. For synchronous training (all-reduce), deterministic latency trumps raw bandwidth. That's why NVIDIA's NVLink uses copper. They optimize for determinism. Now, the supply chain. Who wins? Amphenol (APh) is the gorilla. They make the connectors and cable assemblies for every major server OEM. Luxshare Precision (Luxshare) is the Chinese equivalent, deeply embedded with Apple and now NVIDIA. TE Connectivity supplies the backplane connectors. Hon Hai (Foxconn) owns the manufacturing muscle. These companies are not AI darlings. They trade at 15-20x P/E. But their AI-related revenue is growing 40-60% year-over-year. The market discounts them because they're "old economy." That's the mispricing. Let's quantify the opportunity. Assuming 40% of the $70 billion is copper-related hardware, and the top 5 suppliers capture 70% market share, that's $19.6 billion revenue pool. Amphenol's total revenue is $12 billion. AI copper could add $4-5 billion incremental in two years. At 20% margins, that's $1 billion additional operating income. The current market cap is $70 billion. That implies a 7x EV/EBIT. Too cheap. But there's nuance. Not all copper is equal. Passive DAC has low margins (10-15%). Active copper (AEC) with retimers has margins 25-30%. Companies like Credo Technology (CRDO) supply the retimer chips for active copper. They are a higher-beta play. The market currently prices Credo at 40x earnings, pricing in perfection. But if active copper adoption accelerates, Credo could double. If adoption disappoints, it crashes. That's a binary trade. I prefer the mechanical suppliers – lower risk, steady growth, no hype. Data from recent earnings calls supports the thesis. Amphenol's Q2 2024 AI revenue grew 50% YoY. Luxshare reported a 60% jump in data center interconnects. TE Connectivity's data and devices segment grew 8% but guided 20% next quarter. The order books are swelling. Yet the stock prices haven't reflected it. Why? Because retail money is chasing AI tokens and NVIDIA from the top. Smart money accumulates copper quietly. This is classic market structure. In 2017, during the ICO mania, I analyzed the code of The DAO. The aesthetic symmetry of its smart contract architecture drew me in. But the market ignored flawless code for flawed tokens. The same dynamic: the crowd ignores the aesthetic efficiency of copper's simplicity. They want the shiny optical promise. They forget that deployment timelines matter. A copper cable is plug-and-play. An optical module requires certification, cleaning, polarity checks. In a buildout where time to revenue is critical, copper wins. Contrarian The mainstream narrative is bullish on optical. Every crypto AI token pitches "decentralized GPU networks" and "optical interconnects." The marketing is beautiful. But look at the on-chain data: the top AI token by market cap (Render) has 30% of its supply staked. That's not real demand – it's a yield play. Meanwhile, the actual hardware supply chain is getting funded with real cash flow. Regulation also plays a role. Hong Kong's virtual asset licensing push is often framed as pro-innovation. It's not. It's about stealing Singapore's status as Asia's financial hub. Similarly, the copper vs. optical debate is framed as legacy vs. future. It's not. It's about near-term economics vs. long-term vision. Smart money plays both sides, but the entry point is now for copper. My ISFP nature leans toward the aesthetic – clean code, clean signals. Copper is dirty. It's heavy. It's not photogenic. But the data is clean. The on-chain ledger of industrial orders shows a clear upcurve. Ignore the Discord noise. Trust the data. Takeaway FOMO is a tax on the unobservant. The tax here is buying optical stocks at peak hype. The alternative: accumulate copper connectors with strong order books and reasonable multiples. Watch for the moment when AI tokens pump 20% on a tweet – that's the signal to rotate into the picks and shovels. If you're long-term, the copper window is 12-24 months. After that, optical modules (especially CPO) will take over. But in trading, you don't marry the bag. You respect the cycle. The current cycle is copper. Charts lie. Liquidity speaks. The liquidity is in copper supply chains. Listen. Based on my experience auditing smart contract aesthetics and running quant strategies, the same pattern emerges: infrastructure plays first, applications later. The $70 billion AI network market is real. The copper players are undervalued. The trade is simple: buy the connectors, ignore the hype, collect the alpha.