AI · AI Infrastructure · May 26, 2026

Share

A 141-Year-Old Cable Maker Just Punched a Hole in the AI Trade.

In four trading days between May 14 and May 20, 2026, investors wiped roughly ¥5.6 trillion — about $37 billion — off the market value of Fujikura Ltd. (TSE: 5803), the 141-year-old Japanese optical-fiber maker that had become one of the most celebrated “picks-and-shovels” plays on the global AI build-out. The stock fell roughly 40% from its all-time high of ¥7,855 (May 13) to ¥4,156 (May 20) before staging a partial rebound. The catalyst was not a fraud, an accounting restatement, or a lost contract. It was a three-year forecast — issued by management itself — that came in well below what analysts had penciled in.

Fujikura’s new “Accelerate X” medium-term plan targets operating profit of ¥315 billion (~$2 billion) in the fiscal year ending March 2029 — a number the company is proud of, and which would still represent a sharp ramp from current run-rate. The problem is that the QUICK Consensus of sell-side analysts had been modeling roughly ¥455–460 billionby the same year. A ¥140 billion gap between management’s plan and the Street’s expectations is what the market actually repriced — and the speed of that repricing is the story.

The selloff matters far beyond Tokyo. Fujikura’s shares had risen roughly 1,400% in two years and 160%in 2025 alone on the thesis that hyperscaler data-center buildouts — Microsoft, Meta, Alphabet, Amazon, plus Nvidia’s ecosystem — would absorb every meter of high-density optical-fiber cable the company could ship. When a company that owns roughly 90% of one critical segment of the AI supply chain tells the market it expects to earn one-third less than the consensus three years out, the implication is unavoidable: somebody has been wrong about the durability of AI capex. The question now is which side.

• ¥5.6Tmarket cap erasedin four trading days (May 14–20, 2026) as Fujikura fell roughly 40% from its all-time high — about $37B USD— KuCoin, Bloomberg, Reuters
• ¥315BFY29 operating profit targetin the new 'Accelerate X' plan vs. the ¥455B QUICK Consensus — a ¥140B / ~30% miss against the Street— Bloomberg, Investing.com / Reuters, May 19, 2026
• 1,400%two-year stock gainbefore the rout: Fujikura was up roughly 14x over two years and 160% in 2025 alone on AI data-center demand— Bloomberg (Dec. 2025), FastBull
• $725B2026 hyperscaler capexcombined Microsoft + Alphabet + Meta + Amazon capex — up ~77% YoY — the spend Fujikura's plan implicitly says it cannot fully absorb— Fortune, CNBC, Tom's Hardware

Fujikura is one of Japan’s three big “electric-wire” houses — alongside Furukawa Electric (TSE: 5801) and Sumitomo Electric (TSE: 5802) — and the global leader in ultra-high-density data-center fiber cable, anchored on its proprietary SWR® (Spider Web Ribbon) and WTC® (Wrapping Tube Cable) products. Through most of 2024 and 2025 it was, by total return, the single best performing stock in the Nikkei 225. Then came two earnings releases in six days that the market did not want.

For context: the stock had gained more than 400% in 2024, added another 160% through late October 2025, and topped ¥7,855 on May 13, 2026. Fujikura split its shares 6-for-1 in early April 2026 — a move companies typically make when management itself thinks the stock has run a long way. The selloff erased close to two years of gains in five sessions, then partially recovered. A 50%-peak-to-trough swing in the same week is the kind of move usually reserved for biotech earnings misses, not 141-year-old industrial firms.

Bloomberg’s framing of this story — that Fujikura’s move “exposes cracks in the AI infrastructure rally” — is the loudest interpretation, but it is not the only one. There are three credible reads of what just happened, and they have very different implications for the rest of the AI trade.

Fujikura president and CEO Naoki Okadapublicly anchored on Read 3 the same week the stock collapsed. In the company’s investor presentation, he framed the medium-term target as production-constrained rather than demand-constrained. His direct quote, per the company’s investor materials and wire-service reporting:

“Demand is enormous. We aim to expand production capacity while simultaneously introducing innovative manufacturing technologies to secure a robust competitive advantage.”

Naoki Okada · Director, President & CEO, Fujikura Ltd. · 'Accelerate X' investor presentation, May 19, 2026

That framing did not save the stock on the day. But the medium-term argument has analyst support. Goldman Sachs maintained a buy rating after the plan release. Morgan Stanley flagged Fujikura’s reliance on 15–20% external fiber sourcing as a potential bottleneck — a comment that, again, points at the supply side, not the demand side. Across the sell-side coverage, eight analysts currently rate the stock a buy, two a hold, and zero a sell, with a consensus 12-month target around ¥5,551 — roughly +14% above the post-rout close.

To read the Fujikura selloff against the actual AI capex curve, it helps to start with the underlying numbers, all of which come from Q1 2026 hyperscaler earnings:

That is not a slowing capex curve. It is, by any historical benchmark, the largest single-purpose infrastructure investment cycle of the modern era — roughly 2.5× the late-1990s fiber overbuild on a comparable share-of-GDP basis. Fujikura’s medium-term plan implicitly says the company will absorb a smaller share of that than analysts had assumed, but it does not — and cannot — say the underlying capex curve is bending.

The countervailing data is harder to dismiss. Q1 2026 AI capex of $174B — driving an estimated 42% of U.S. GDP growth in the quarter and equal to roughly 2.4% of total U.S. GDP — represents the largest single-sector infrastructure commitment in modern American economic history. Bloomberg Opinion has separately documented that big-tech AI buildouts now face a meaningful chip-and-component inflation problem: Microsoft’s CFO openly attributed $25B of its 2026 capex to memory-chip price increases. The story is not that the buildout stops. It is that the buildout costs more, and margin distribution along the supply chain is shifting in ways the market has not yet priced.

Japan’s three major optical-fiber houses each ride the same AI-infrastructure wave but with very different exposure profiles. How the other two trade after Fujikura’s revaluation will tell us whether the market is repricing the whole AI-cable sub-sector or just one company’s capacity story.

The point worth dwelling on: Furukawa hitting an all-time intraday high during the same monthFujikura collapsed is hard to reconcile with a story about the AI-fiber sub-sector breaking. It is much easier to reconcile with a story about Fujikura’s specific medium-term guidance disappointing a Street that had extrapolated 2024–2025 growth rates indefinitely.

The Tokyo Stock Exchange recorded a daily trading turnover of ¥12.04 trillionon the day Fujikura’s “Accelerate X” release dropped — a record. The breadth was equally telling: 869 stocks declined and 664 advanced, with broad-based profit-taking spreading from Fujikura into other AI-themed Japanese names that had run hard into 2026. The Nikkei 225 had touched a fresh all-time high earlier in the same week and then reversed: closing 1.99% lower on the Friday of the rout session (May 15) at 61,409.29.

By May 25, 2026, the Nikkei had recovered enough to push back above 65,000 on optimism around U.S.-Iran nuclear talks, but the AI-stock sub-index remained mixed: Kioxia Holdings (-3.5%), Fujikura (-4%), and Advantest (-4.9%) all declined that day, while SoftBank Group ran +6% on news that two of its portfolio companies — OpenAI and SB Energy— were nearing U.S. IPOs. SoftBank’s four-session gain ran nearly +50% — a vivid reminder that AI sentiment in Japan is, at this point, less a sector trade than a ticker-by-ticker scramble through whatever inventory the market is willing to underwrite that week.

On the same May 14, 2026 board meeting that delivered the conservative FY27 guidance, Fujikura also formally approved its ¥300 billion ($1.9B)capacity-expansion plan — an amount that exceeds the company’s entire historical investment base for this product line. The breakdown:

The timing of the announcement is what kept reasonable-minded analysts on the supply-side reading of the story. Fujikura did not simultaneously cut capex and cut profit guidance — the classic late-cycle distress signature. It substantially raised capex (the ¥300B plan is the largest in company history) while delivering a profit-target ladder the market judged too conservative. Companies that genuinely fear demand erosion do not commit ¥260B to a U.S. greenfield subsidiary on a five-year payback horizon. Companies that face capacity constraints in the here-and-now, by contrast, do exactly that. The market sold the number; the supply chain still bought the build-out.

The Fujikura sequence — vertical rally, vertical drop, partial rebound — is a near-perfect compressed model of the broader AI-trade debate. The bear case, articulated by Bloomberg, Moody’s, and multiple Wall Street strategists, is that valuations have run far ahead of cash-flow realization: Moody’s now maps a scenario in which AI-linked equities draw down 40%in the coming months; Barclays sees Meta’s free cash flow falling nearly 90% under current capex trajectories; Amazon is on track for ~$17B of negative free cash flow in 2026. The Fujikura selloff is, on this read, the first credible crack — the first time a company at the supply-side core of the trade has handed the market a number the market refused to validate.

The bull case is shorter and partially circular: capex is real, backlogs are large (Jefferies pegs ~$2 trillion of cumulative cloud-services backlog), and the buildout has at every prior historic analog — railroads, electrification, fiber — produced infrastructure that became foundational regardless of how many equity holders got wiped out in the rerating. Wells Fargo’s position, articulated bluntly in May 2026, is that AI capex is a “euphoric” bubble and that the right move is to ride it anyway because the spend is too large to fight. That is closer to a momentum thesis than a fundamental one, but it captures the reflexivity of the current market: as long as the spend keeps climbing, the stocks tend to follow.

The honest reading sits between the two. Fujikura’s collapse is not the start of an AI-infrastructure unwind — Furukawa hit an all-time high in the same month, hyperscaler capex is still accelerating, and the company itself committed its largest-ever growth capex budget on the same day it disappointed the three-year profit target. But the collapse is genuine information about how fragile the equity narrative is around any single AI-cycle beneficiary. A 40% multi-day drawdown in a 141-year-old industrial on what is, ultimately, a guidance shortfall — not a fraud, not a recession, not a lost contract — tells you the marginal AI-trade holder is at this point underwriting a very thin sliver of execution. That thin sliver is what cracked.