683 Companies Out: Japan's Biggest Topix Shake-Up Since 1969 Opens a Door for Foreign Partners — and This Week's AI Models Make the Homework Cheap

The Topix is Japan's broadest major stock index. For most of its life it simply held almost every company on the Tokyo Stock Exchange's old First Section, at one point about 2,200 of them. On Wednesday, October 7, Japan Exchange Group (JPX) published the results of its new rules, and Bloomberg called it the biggest reshuffle since the index started in 1969. 683 companies will be phased out and 35 will be added, so the index falls from 1,636 stocks to 986, a cut of about 40%. Additions include McDonald's Holdings Japan, the workwear retailer Workman, the taxi-app operator GO and the semiconductor-materials maker Ferrotec. Removals include well-known names such as the staffing group Pasona.

Two numbers now decide whether a company is in the index. The first is the annual trading value turnover ratio, which shows how actively the shares change hands. The second is free-float market capitalization: the value of the shares that can actually be traded, after excluding blocks held by parents, partners and founders. A stock must rank within roughly the top 97% of total free-float market value and meet the turnover test. Just as important, index membership is now separate from the market segment. A Prime-listed company that fails the tests is out; a Standard or Growth company that passes them is in.

The change is slow by design. Weights of the removed companies are cut in eight quarterly steps, starting October 30, 2026 and ending in July 2028. A company that meets the rules at the reassessment in October 2027 stops the process. From October 2028 the index will be reviewed every year, so no company can count on staying in by default. The money involved is large: Daiwa Securities estimated in September that about ¥166 trillion (around $1.1 trillion) in passive funds tracks the Topix, including the Government Pension Investment Fund. Foreign investors were already repositioning: in the week to October 2 they sold a record ¥2.24 trillion of Japanese equity futures while buying a net ¥956 billion of cash shares.

It is tempting to read the list as a list of losers. That would be a mistake. The new tests measure how much of a company the market can trade, not how good the company is. Many of the 683 are profitable, mid-sized manufacturers, distributors and service firms whose shares rarely trade because most of them sit with a parent company, a group of banks and business partners holding cross-shareholdings, or a founding family. JPX itself says the sector mix and valuation of the index barely change after the reform. In plain terms, the exchange did not remove bad companies; it removed companies that the public market cannot easily own.

That creates a clock. A company that wants to stay in the index has until the October 2027 review to raise its free float and its trading volume. The usual levers all change who owns the company. Unwinding cross-shareholdings puts blocks of shares up for sale, and someone has to buy them. A parent can sell down a listed subsidiary. A family can bring in a strategic investor. A company can also decide that the listing is no longer worth it: management buyouts and take-privates were already running at record levels before this week. Japan logged 4,086 domestic M&A deals in 2025, an all-time high, worth ¥11.2 trillion. Inbound deals by foreign buyers rose to 372, worth ¥6.2 trillion, and private-equity investment in Japan more than doubled to a record $51.8 billion.

For a foreign company, the important point is that the pressure is not only about the stock. A board that must explain to shareholders how it will grow, raise return on equity and justify its listing is a board that is open to new distribution channels, joint ventures, licensing deals and minority investors. Since 2023 the Tokyo Stock Exchange has asked listed companies to publish plans for management that is conscious of the cost of capital and the share price. Many of the 683 have published such plans with promises they have not yet kept. A credible foreign partner that brings new revenue outside Japan is exactly the kind of story those plans need.

The hard part of finding a Japanese partner has never been the list. It has been reading. Each of the 683 companies files an annual securities report (yuho) of often more than a hundred pages, a mid-term plan, a corporate-governance report and, in many cases, a cost-of-capital disclosure. Most of it is only in Japanese. Until recently, a foreign company screening a few hundred candidates needed a team of bilingual analysts for weeks. Two launches on the same day as the Topix announcement change that math.

On October 7, OpenAI began rolling out GPT-6 in ChatGPT, with GPT-6 Sol for Plus, Pro, Business and Enterprise users and GPT-6 Luna for Free and Go users from October 8. It comes with Intelligent UI, which can answer with charts, tables, forms and buttons inside the chat instead of plain text. That is useful when you want to compare twenty companies side by side. The same day Anthropic released Claude Haiku 5.5, a small model built for high-volume work, at $0.10 per million input tokens and $0.50 per million output tokens for prompts under 100,000 tokens — about a tenth of Haiku 4.5's standard rate.

Here is our rough estimate. If you extract the key sections of each company's filings — business overview, segment results, shareholder structure, cross-shareholdings and the cost-of-capital plan — you get roughly 80,000 tokens per company. Reading all 683 then means about 55 million input tokens, or around $5.50 with Haiku 5.5, plus well under $1 for a structured two-page summary of each. Even allowing for retries and a second pass with a stronger model on your top 50, the model bill stays below the cost of one hour of a bilingual analyst. Two cautions apply. First, check every number that will drive a decision against the original filing on EDINET, the Financial Services Agency's disclosure site; models still misread tables. Second, keep any material received under a non-disclosure agreement out of consumer chat tiers and use a business or API account with data-retention controls.

Days 1 to 30: build the long list. Download the JPX list of companies scheduled for removal and filter it by the sectors where you sell, buy or build. For each company, have an AI model pull five facts from its filings: what it sells and where, its share of overseas revenue, who holds its largest blocks of shares, what it promised in its cost-of-capital plan, and whether that promise is on track. Then have a person who reads Japanese check the top 30. You will usually find that a handful of companies have a strong product, weak international sales and a plan that names overseas growth as a goal. Those are your targets.

Days 31 to 60: design the offer. A Japanese board facing the October 2027 review wants something it can show shareholders within a year. Depending on your size, that can be a distribution agreement that puts their product into your European or American channel, a joint venture for a new product line, a technology licence, or a minority stake that buys part of a cross-shareholding block and so raises free float. Write the proposal in Japanese, show the numbers in yen, and link your offer directly to the targets in their own published plan. A proposal that quotes a company's own mid-term plan back to it is read very differently from a generic pitch.

Days 61 to 90: make the approach the Japanese way. Cold emails to investor-relations inboxes rarely work. An introduction through the company's main bank, a trading house, a regional chamber of commerce or an adviser who knows the board is far more effective, and the first meeting should be about mutual understanding, not terms. Expect nemawashi — quiet consensus-building inside the company — to take months, which is why starting now matters: the companies that act early will have chosen their partners well before the 2027 review. This is the work we do at Medusa Japan from Osaka: bilingual screening, Japanese-language proposals and introductions that open the right doors on both sides.