TOKYO, May 8, 2026, 05:05 JST
- Japan’s Nikkei average closed at a record 62,833 on Thursday, lifted by heavy buying in AI and semiconductor names.
- Fresh Japanese commentary has turned the rally back to a harder question: whether Nvidia’s valuation, and the wider AI-chip trade, can be justified by cash returns.
- Broadcom, AMD and Arm are now part of the same test, as investors hunt for Nvidia alternatives but face supply, financing and concentration risks.
Japan’s AI-chip rally became a Tokyo market event on Thursday, not just a Wall Street story, after the Nikkei average posted its largest-ever point gain and local market reports revived concern that the Nvidia-led boom may be running ahead of its economics. TBS reported that the index closed 3,320 points higher at 62,833, a record finish, as orders poured into semiconductor-linked shares after overseas AI and chip stocks rose during Japan’s Golden Week break.
The timing matters because the trade is narrow and global. A Sankei report carried in Yahoo Japan’s AI news feed framed the rally as being led by AI and semiconductor shares from the United States through South Korea and Taiwan, while also flagging overheating concerns.
Shikiho Online, in a piece published two days ago, said the source of returns in U.S. semiconductor stocks lies in company-specific “factors,” not just the broad Nvidia trade, and said it would dissect three leading U.S. chip names to explain why some returns have surpassed Nvidia. That framing puts Broadcom and AMD closer to the centre of the Japan retail-investor debate. Shikiho Online
The sharper warning came from economist and investor Emin Yilmaz, quoted by Bunshun Online. Yilmaz said, in remarks translated from Japanese, that “all bubbles collapse,” and called it abnormal that Nvidia’s market value had exceeded Japan’s GDP. He also said strong earnings and a bubble can coexist, because a bubble is a market priced as if current growth will last forever. 文春オンライン
The other side of the trade is real demand. Nvidia and Corning said on Wednesday they had agreed a multiyear partnership to expand U.S. optical connectivity production for AI data centers, the large facilities that house chips and servers used to train and run AI models. Corning said it would lift U.S. optical connectivity capacity tenfold, expand domestic fiber output by more than 50%, build three plants in North Carolina and Texas, and create more than 3,000 jobs.
Nvidia Chief Executive Jensen Huang called AI “the largest infrastructure buildout of our time,” while Corning CEO Wendell Weeks said the partnership showed AI was “not just a technology story” but a manufacturing one. The language was promotional, but the physical investment is the point: AI spending is moving from chips into fiber, power, land and factories.
Broadcom is the clearest peer in the custom-chip lane. Reuters reported in March that Broadcom projected more than $100 billion in AI chip sales in 2027, after first-quarter revenue rose 29% to $19.31 billion and AI revenue more than doubled to $8.4 billion. “Our visibility in 2027 has dramatically improved,” CEO Hock Tan said on a post-earnings call. Reuters
Broadcom’s role is different from Nvidia’s. Nvidia sells GPUs, graphics processors used to train and run AI models; Broadcom helps large customers such as Google and OpenAI turn in-house designs into custom AI accelerators, chips built for machine-learning workloads. D.A. Davidson analyst Gil Luria said Broadcom’s longer-term visibility was “an indication of significant growth in demand.” Reuters
But the risks are also showing up in the same supply chain. The Information reported Thursday that OpenAI’s AI-chip deal with Broadcom had hit an $18 billion financing snag, a reminder that some AI buildouts still depend on huge funding packages and future revenue assumptions. OpenAI and Broadcom had announced in October a plan to deploy 10 gigawatts of OpenAI-designed AI accelerators, with racks targeted to start in the second half of 2026 and finish by the end of 2029.
Arm offered another warning sign. Reuters reported Thursday that Arm shares fell after the chip designer warned about smartphone weakness and supply challenges for a new AI chip; CEO Rene Haas said Arm had secured capacity for the first $1 billion of demand but not beyond that.
U.S. market action was mixed, not euphoric. Reuters said the PHLX chip index lost 2.4% on Thursday, with AMD down 3.1% and Intel off 1.9%, while Nvidia and Microsoft both rose more than 2%. That split suggests investors are still paying for proven AI leaders, but are less patient with companies where supply, funding or execution questions are harder to answer.
The downside case is simple. If AI users do not pay enough, fast enough, the data-center spending now lifting chip, fiber and power names could leave suppliers with stretched valuations and customers with heavy bills. Yilmaz told Bunshun he doubted AI businesses could monetize at the speed implied by current investment plans, and said the cost-recovery math would require a very large base of paid users at high monthly fees.
In Tokyo, the next check is earnings. Iwai Cosmo Securities chief strategist Kazuaki Shimada told TBS there was “short-term overheating” and that buying had become more concentrated in semiconductors. He said Japanese corporate guidance, starting with major earnings due after the rally, would be key to whether the move holds. TBS NEWS DIG
For now, the market is holding two ideas at once. AI demand is creating real orders, plants and supply bottlenecks; it is also asking investors to believe that a small group of companies can keep turning vast infrastructure spending into profit. That is why Nvidia is no longer just a chip stock in Japan’s market debate. It is the proxy for whether the AI boom becomes cash flow, or just another expensive buildout.
