Tokyo's verbal yen defense just handed the carry trade a credibility haircut — and AI funding flows are the tell

On October 1, 2026, Japanese Prime Minister Sanae Takaichi told Nippon Television that her administration's growth push — not yen-buying — would "ensure market confidence in the yen," and explicitly denied that her economic policy targets the exchange rate. She also disclosed that she had told US President Trump in September that the yen's undervaluation was a problem. The line was carried by Reuters and picked up by CNBC the same day.

The problem is that the yen never re-anchored. The Ministry of Finance sold an estimated $85 billion to buy yen on July 30 and 31 — Goldman and Morgan Stanley's published estimates — and the US Treasury chipped in another $5–10 billion by selling euros for yen. USD/JPY briefly dropped from 163.99 to about 155, but by Aug. 18 it had already weakened back to 159, and the pair has spent September grinding between 157 and 158.50. Takaichi's October 1 comments came with the 10-year JGB at 3.11% — a level not seen since 1996 — and the 30-year at 4.13%, an outright record.

The yen carry trade is funded on the rate gap, not on FX narrative. Cross-border yen borrowing — the closest proxy for carry-trade size — ballooned to a record ¥360 trillion ($2.35 trillion) by March 2026, the largest such buildup in three decades per BIS data analyzed by Jefferies. That position was accumulated when BOJ policy rates sat near zero and US 10-year yields sat near 4.5%, so the spread was enormous.

Morgan Stanley's analysis is blunt: unless BOJ hikes above 1.75–2% "within a short period," the market will keep viewing the central bank as "behind the curve." The implication is that verbal intervention — Takaichi's October 1 pitch included — does not change the price of the carry. It only changes the volatility around it. That distinction is the whole story for investors downstream.

The yen carry trade's most consequential line item in 2026 is US AI infrastructure financing. Japanese life insurers, banks, and pension funds are now the marginal funding pipe for hyperscale data centers. Nippon Life Insurance — a mutual, but the largest of Japan's life carriers — disclosed in mid-September that it will invest ¥2 trillion ($12.75 billion) in infrastructure loans, primarily for AI data centers in the US, by 2035. Yields on those loans average more than 2% over the cost of hedging the dollar, and they are funded on the cheap-yen side of the balance sheet.

Hyperscaler AI capex is on track for roughly $697 billion in 2026 per J.P. Morgan estimates, with another $186 billion of supporting infrastructure financing layered on top. Japanese institutions are not the largest single funder — Wall Street banks and US private credit dominate — but they are the most FX-leveraged slice, because every dollar they lend against the yen balance sheet is implicitly a short-yen position.

Japan's Financial Services Agency noticed. On September 25 the FSA announced it was stepping up scrutiny of AI data-center financing by Japan's biggest banks and life insurers, focused on risk management rather than blocking the lending. The FSA's concern is the same one Fitch flagged for the megabanks in 2022: the dollar funding cost rises when the yen strengthens, and a wave of unhedged or under-hedged US project-finance loans can swing Japanese insurers' foreign-bond hedge ratios — already cut from ~60% to ~40% over the past two years — into reverse.

The clearest winners are the Japanese megabanks, and the data is unambiguous. MUFG closed FY2026 (ended March 31, 2026) with net income of ¥2,427.2 billion, up 30.3% year-on-year and the highest in 31 years, with ROE lifting to 10.85% from 9.08% the prior year. Q1 FY2027 (three months to June 30, 2026) followed through with ¥809.4 billion of net income, up 48% on the year, and management guided to an all-time-high ¥2.7 trillion for the full year. Sumitomo Mitsui's FY2026 ROE rose to 10.59% from 8.01%; Mizuho's lifted in parallel.

But the cost of keeping the carry trade alive is concentrated at life insurers, where domestic bond holdings have racked up a record $96 billion of unrealized losses as of end-June 2026 — up 7% in a single quarter. Dai-ichi Life and Tokio Marine, the two publicly traded majors with US listings, are exposed on both sides of the trade: long foreign bonds for yield, short yen through USD asset holdings, and now the FSA pressing them on AI data-center underwriting standards.

On the other side of the Pacific, the AI buildout itself stays well-funded because the carry trade hasn't unwound — but the marginal dollar cost rises whenever yen stress spikes. Nvidia's reported $500 billion AI financing push, unveiled in early October, depends in part on Japanese institutional commitments. SoftBank Group — long-duration dollar AI bets funded in yen-denominated debt — is the cleanest single-name barometer: the company has issued a record ¥1 trillion ($6.3 billion) of Japanese retail bonds to fund AI capex, on top of $30 billion of follow-on commitments to OpenAI signed in February and roughly $30 billion of obligations due in the second half of 2026.

The single highest-impact near-term catalyst is the Bank of Japan policy meeting on October 29–30, 2026. A 25 bp hike to 1.5% is the consensus expectation and is already largely priced; the asymmetry comes from a surprise hold (yen weakens, JGB rally, carry trade widens) or a hawkish surprise that explicitly links BOJ tightening to FX stability. A former BOJ official told the Japan Times on Sep 28 that a back-to-back hike is a real possibility at 20–30%.

Over a 1–3 year horizon, the structural question is whether BOJ keeps pace with the Fed–BOJ rate convergence Morgan Stanley models. The bank's economists expect the yen's fair value to strengthen from 165–167 today to about 155 over the next 12–18 months as the Fed pauses and BOJ continues normalizing. That trajectory forces the carry trade to shrink gradually rather than blow up — which is the soft-landing scenario for AI funding, not the crisis one.

The bear case is also structural: if Takaichi's $2.3 trillion, 14-year investment plan delivers the inflation she is hoping for, BOJ may need to hike faster than markets expect, the JGB curve steepens past its October 2 levels, and the carry trade compresses into a US recession. That is the path where SoftBank's dollar bonds come under pressure, life insurers face mark-to-market on foreign bonds again, and the marginal yen-funded AI loan becomes uneconomic. The probability is not high, but it is no longer negligible — which is exactly what "credibility haircut" means in practice.