TDK's Record Half Is Being Paid For By AI Data Centers — and Financed By Negative Free Cash Flow

Record sales and profit, a raised dividend, and a ¥100bn bet on physical AI, against a smartphone market management just cut again and a free cash flow line that swung to minus ¥50bn

6762.T · Earnings Call · 2026-10-30

TDK: The Record Half Nobody Asked For, and the Bet Nobody Has Priced Yet

Two markets, one P&L

TDK's first half of fiscal March 2027 was, on paper, the best in company history: net sales ¥1,560.6bn (+31.9%), operating profit ¥208.2bn (+41.0%), net profit ¥176.4bn (+58.4%), EPS ¥92.94. The second quarter alone set quarterly records on every profit line. But the composition is the story. Demand for the AI ecosystem is running roughly 1.7x year-on-year against an April plan of about 1.2x, while the other half of the portfolio is contracting. Yamanishi did not soften it: “Given the greater impact expected from memory supply shortages, we now project a further 5% point year-on-year decline, taking production below 1.1 billion units” — Tetsuji Yamanishi, Senior Executive · 2026-10-30 — on top of the roughly 10% decline already assumed in April for smartphones. That is a squeezed ICT device production market, and TDK guided small-capacity battery volumes down about 10% quarter-on-quarter into the third quarter as a result. Yet the company raised full-year guidance to ¥3,550bn of sales and ¥365bn of operating profit. The bridge is simple: data center demand, plus a weaker yen. Foreign exchange alone added ¥134.4bn to first-half sales and ¥19.9bn to profit, and the guidance rate moved from ¥150 to ¥155. Strip the currency out and you still have a genuine volume story in passive components (sales +25.3%, profit 2.7x) and magnetic applications (profit 2.8x, on HDD head volumes +38% and suspensions +34%).

The genuinely new theme: physical AI

What is new in this quarter's language is not “AI data centers” — that phrase has been climbing TDK's keyword stack for several quarters. It is physical AI, which appears at the top of the company's own keyword set for the first time, and it comes with capital attached rather than adjectives. Saito committed roughly ¥100bn to expand TMR sensor capacity — a two-site system across Shinanogawa and Asama — with the reasoning spelled out: “we estimate that dozens of sensors will be used for humanoid robot for joint angles and hand tactile sensing and other functions” — Noboru Saito, President and CEO · 2026-10-30. The second new thread is stranger and more interesting, because it is a genuine technology transition rather than a demand wave. TDK is putting about ¥40bn into thin-film inductors because the industry is moving voltage regulators from beside the processor to underneath it — vertical power delivery. The pitch is that shorter wiring cuts resistance and "improve efficiency by about 70%," and that TDK's coil thin-film plating process, inherited from HDD head manufacturing, is what allows the lower profiles and high current the new substrate-embedded regulators need. That is a direct read-through to the rack-level power architecture debate showing up across the global tape — other reporters this week are explicitly framing 800V data center architectures and high-voltage DC distribution as the next design cycle. TDK is not the only company talking about it, but it is one of the few arriving with a component-level answer and a capacity number. And here is the tension worth holding onto: over the last 90 days the market tape on AI data center exposure has been decisively negative — a wide basket of names down roughly 5% with far more decliners than advancers. TDK is raising capex into a theme the tape has spent a quarter selling. Either the component layer is decoupling from the system layer, or TDK is late to the party. The evidence in this call — nearline HDD units raised to 79 million, MLCC capacity up about 20%, aluminum electrolytic capacity targeted at roughly 7x by March 2029 — points to the former, but it is an assumption, not a fact.

In HDD market, continued robust demand has led us to raise our forecasts for nearline HDD production for data centers from the April estimate to 79 million units.
Tetsuji Yamanishi, Senior Executive · 2026-10-30

The bill: negative free cash flow and a quieter portfolio story

The least discussed line in the release is the cash. First-half operating cash flow fell ¥110.8bn year-on-year despite higher profit, hit by a one-off withholding tax payment and working capital build; capex was ¥101.6bn including the Linergy acquisition; free cash flow was negative ¥23bn. Full-year free cash flow guidance was cut to “negative JPY 50 billion, mainly reflecting strategic M&A investments” — Tetsuji Yamanishi, Senior Executive · 2026-10-30, with annual capex lifted ¥30bn to ¥400bn and depreciation to ¥250bn. Management insists this does not impair the midterm capital allocation plan, and it raised the dividend from ¥20 to ¥25 in the same breath — a deliberate signal that the cash burn is investment, not deterioration. The portfolio-management narrative has also subtly changed shape. Earlier quarters leaned on "structural reform" and one-time expenses; this quarter the vocabulary is business portfolio management, with a concrete tally: 15 CBUs now under improvement, up from 13, after two flash-memory product groups were moved in, and three remaining under discussion to be decided by fiscal year-end. Notably, the tariff language that populated the top of TDK's keyword list earlier in the year has vanished entirely, replaced by AI. Governance is the other quiet move — “we will transition from our current company with an audit and supervisory board company with the three committees” — Noboru Saito, President and CEO · 2026-10-30 — which reads as preparation for a next midterm plan under sharper external oversight.

The honest read

Saito's framing was unusually candid about the shape of the year: “Targets in the medium-term plan are largely on track, but we will further reinforce business portfolio management” — Noboru Saito, President and CEO · 2026-10-30. That is a CEO saying the plan is fine but the mix is not, and he is reallocating toward the AI ecosystem and physical AI while managing the smartphone decline rather than pretending it away. ROIC guidance rising from 8% to 9.7% on a ¥3,550bn sales base is a real improvement, and the ¥100bn TMR commitment plus R&D and CVC activity suggests TDK sees a window rather than a cycle. The risk is equally clear: the company is levering its balance sheet and its cash conversion to a demand curve whose terminal value the market has spent three months marking down. The record half is real. The multiple is the argument.