Tower makes Japan its main hub for optical chips

What happened

On September 24, 2026, Nikkei Asia reported from Tokyo that Tower Semiconductor will set up its largest production hub for optical communications semiconductors in Japan. Chief executive Russell Ellwanger told Nikkei about a planned $4 billion investment. The headline deck says that money is aimed at booming demand for AI server technology. Nikkei also notes that Tower already has three manufacturing sites in Japan, plus plants in Israel and the United States. The rest of the Nikkei piece is behind a paywall, so the breakdown below follows Calcalist the same day. Calcalist attributes the plan to Ellwanger’s Nikkei interview and to company statements. The dollars in that account are company figures, not an outside audit.

Calcalist reports that Tower is investing about $3 billion in its Japanese operations, backed by about $1 billion in grants from the Japanese government. Nikkei reported $4 billion. Read the Calcalist pair as company figures on the same plan, not as a second pile of money and not as an audited sum. The spending expands factories Tower already has in Japan rather than depending only on a new site. The company plans to convert its Arai facility, formerly Fab 6, into a 300mm silicon photonics and advanced optical packaging site, and to raise output at Fab 7 in Uozu. That first phase is aimed at full production readiness in the fourth quarter of 2027. Tower then intends to build a new 300mm fab next to Fab 7. The company says that plant would multiply silicon photonics and silicon germanium capacity several-fold and begin a significant financial contribution from 2029. Both dates are aims on the company’s calendar.

The demand figures are company figures too. Tower says it has contracts representing about $1.3 billion in silicon photonics revenue for 2027, and that customers have prepaid about $290 million to reserve capacity. Contracted wafer commitments for 2028 are described as even larger. Silicon photonics is an optical I/O approach: it moves data with light instead of only electrical links, which is why AI clusters that shove traffic between processors care about it. In the same Calcalist account, Tower has raised its targets to $3.6 billion in revenue and $1.2 billion in net profit in 2028, compared with $1.6 billion in revenue and $220 million in net profit in 2025. Those 2028 figures are company guidance, not a forecast from this page.

The Japan work runs through Tower’s local operations, including facilities held via its majority ownership of TPSCo, the former Panasonic Semiconductor manufacturing business. Calcalist adds that Tower plans deeper cooperation with Japanese suppliers, universities, and research institutions.

Why it matters

AI’s pinch is shifting from getting a processor to moving bits between processors without overheating the rack. A specialty foundry putting this much of its own money into Japan, with public grants beside it, is an industrial bet on that traffic. The 2027 contracts and the prepayments are a demand signal only for as long as you remember Tower is the one stating them. The 2028 revenue and profit targets are the same kind of statement. Israel still supplies the specialty process background. The United States and Israel plants stay in the network. This is not TSMC CoWoS, and it is not an HBM story.

What to watch is whether the Arai conversion hits the fourth-quarter 2027 readiness aim, how much of the grant money is actually drawn, whether the new fab beside Fab 7 gets a public construction date, and whether 2028 contracted wafers show up as revenue near the company’s own target. Until that first phase is actually ready, this is a funded plan.