The AI infrastructure rally is moving beyond easy faith. Japan's cablemakers show what happens when investors price every supplier like a permanent winner.
The selloff in Japanese cable stocks is not just a local market wobble. It is a reminder that the AI boom has turned ordinary industrial suppliers into high-expectation growth stories, and those expectations can reverse faster than the underlying demand disappears.
Companies such as Fujikura, Furukawa Electric and Sumitomo Electric became natural proxies for the data center buildout because artificial intelligence needs more than Nvidia chips and cloud contracts. It needs fiber optic cable, power cable, connectors, cooling materials and the physical network that moves data between servers. For a while, that was enough. Investors wanted exposure to anything that looked like a bottleneck in the AI stack.
That trade now looks less forgiving. Trading Economics reported that Fujikura was among the leading losers in the Nikkei 225 on May 20, falling 8.54% on a day when the index dropped 1.33%. The move came after a fierce run in AI-linked Japanese industrial names, where the market had started to treat cable capacity almost like chip capacity: scarce, essential and worth a premium.
The problem is not that data center demand has vanished. It has not. Fujikura reported on May 14 that operating profit for the fiscal year ended March 31, 2026 rose 39.2% to 188.7 billion yen, while net sales climbed 20.7% to 1.18 trillion yen. The company pointed to strong data center infrastructure demand in its information and communication electronics business, raised its dividend payout ratio to about 40%, and announced plans for a 40 billion yen plant investment at its Sakura site.
Those are not weak numbers. That is exactly why the reaction matters. When investors sell after good results, they are usually objecting to the price, not the business. The market is starting to ask whether the best part of the earnings surprise is already in the stock.
As Bloomberg recently reported, a narrow set of Japanese cable and industrial suppliers became stand-ins for optimism around AI data center construction. That is a powerful setup on the way up because the story is easy to understand. Hyperscalers spend more, data centers multiply, cablemakers sell more high-value components. The line from model demand to factory orders feels direct.
But proxy trades carry their own danger. They compress a complex supply chain into a few tickers. A company may have a real product advantage and still be vulnerable if investors have already priced in years of flawless execution. That matters especially in cable and power infrastructure, where capacity expansion takes time, margins can be exposed to copper prices and logistics, and customers are large enough to push hard on terms.
Furukawa Electric's own materials show both sides of the story. The company has pointed to continued robust demand for data center-related products, including rollable ribbon cables, MT ferrules and laser chips, while also flagging issues such as copper price pressure in some segments. Its functional products business includes heat dissipation and cooling products, copper foil for high-frequency boards and semiconductor process tapes. These are useful AI infrastructure exposures, but they are not immune to cost cycles.
Sumitomo Electric is making a similar case from a broader base. In its data center growth strategy, the company said the spread of AI is pushing data processing demand ahead of earlier projections and that market growth is expected to continue beyond 2026. It also cited server spending forecasts and rising global data center power demand as support for optical fiber-related products.
That is the reason investors were interested in the first place. The question now is how much of that future was pulled into today's share prices.
The AI stack is bigger than chips
For readers in the U.S. and Europe, Japan's cablemaker rout is useful because it shows how the AI trade migrates. First, capital chased model builders and chipmakers. Then it moved into servers, networking, memory, cooling, utilities, construction companies and electrical equipment. Each step made sense. Each step also widened the circle of companies being valued on AI demand rather than their older industrial cycles.
That is where the next phase of the market gets harder. A supplier can be essential without being defensible. Fiber optic cable is critical, but the economics depend on product mix, capacity, customer concentration and how quickly rivals can add supply. Power infrastructure is also critical, but utilities, grid equipment makers and contractors face permitting delays, commodity costs and balance sheet constraints. These are not software margins hiding inside industrial businesses.
There is also a timing issue. Hyperscalers such as Microsoft, Amazon, Google and Meta are still spending aggressively on data centers, but equity markets do not wait for final demand to crack before repricing suppliers. They move when the rate of positive surprise slows. That is why order guidance, backlog quality and capex plans can matter more than a headline earnings beat.
The cleanest AI infrastructure companies will be those that can prove pricing power, not just demand exposure. Investors will want to see long-term customer commitments, differentiated products, disciplined capacity additions and margins that survive higher input costs. The rest may still grow, but growth alone may not support the valuation premiums attached during the first wave of excitement.
Japan's cablemakers are not a rejection of the AI buildout. They are a test of its valuation discipline. The physical internet of AI still has to be built, and companies that supply the right parts will have real work ahead of them. But the market is moving from asking who benefits to asking who keeps the economics. That is a much tougher question, and it is the one infrastructure investors should be watching next.
Also read: Wall Street banks are paying AI experts $25,000 a day • A disputed METR graph is testing AI's benchmark economy • ClickUp's layoffs show AI is rewriting productivity software.
