Skip to content

Build1 publisher3 min readPublished

Tower's $4 billion optical-chip build-out in Japan starts inside an already-qualified fab

Tower Semiconductor and Japan will co-invest $4 billion to reach 45,000 300mm-equivalent wafers a month of optical-connectivity capacity by 2029. Until the new fab lands in 2029, growth comes from adding tools to a fab that already ships.

The Engineer · Build desk

Illustration accompanying Tower's $4 billion optical-chip build-out in Japan starts inside an already-qualified fab

What happened

  • Tower Semiconductor will put $3 billion and Japan's METI $1 billion into Tower's Japanese fabs, Nikkei reports, for a roughly $4 billion optical-connectivity build-out.
  • Tower targets capacity equal to 45,000 300mm wafers a month in Japan by 2029, around 40 times its 2025 level.
  • Track One converts the idle 200mm Fab 6 in Arai to 300mm for silicon photonics and optical packaging, and expands output at the 300mm Fab 7 near Uozu.
  • Track Two builds a new 300mm fab beside Fab 7 to raise chip output several times over, with a material contribution to results expected in 2029.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint Through 2028, Tower's supply growth is limited to what added tools at Fab 7 and the converted Fab 6 can deliver, because the new fab only contributes from 2029.
  • exposure With more than 80% of contract optical-chip production for servers at one company, a schedule slip at Uozu would reach most contract buyers, Marvell among them.
  • cost Tower carries three-quarters of the $4 billion itself, so a late Track One ramp lands mostly on its own balance sheet.

Track One leans on Fab 7 near Uozu [4]. The fab is already fully qualified. It is in mass production of silicon-germanium electronic chips on 65nm-class nodes such as TPS65SG, and of silicon-photonics chips on the 45nm-class TPS45PHD process [5]. Adding equipment there lets Tower raise output as each tool is installed, without waiting for a new fab and its process flows to qualify [6]. Tower has set full production readiness for Track One in the fourth quarter of 2027 [7].

Qualification sets the order of the plan. New capacity goes first onto process flows that already ship, and the building that still has to qualify comes second [6][9]. I think this sequence is right: output grows from the first installed tool, and the schedule risk sits in Track Two [9]. The layout is sensible too. The converted Fab 6 takes on optical-engine assembly, so the end state has two sites making the electronic and photonic chips and one bonding them together [10].

The risk does sit there. Tower's 45,000-wafer target falls in 2029, the same year it expects the new fab beside Fab 7 to start contributing materially [2][9]. The headline figure depends on a fab Tower has yet to construct passing qualification on schedule [9]. Working back from the 40-fold multiple, Tower's Japanese capacity in 2025 was about 1,125 300mm-equivalent wafers a month [1]. Tower plans to hire about 200 people [14].

Tower expects Track One alone to support about $3.6 billion of revenue and $1.2 billion of net profit in FY2028 [8]. The implied net margin is about 33% [2]. That forecast is Tower's claim about its own order book [8]. For it to hold, Track One has to hit its late-2027 readiness date and the added tools have to run full in their first year [7]. METI's $1 billion covers a quarter of the $4 billion total, and Tower funds the rest [3][3].

Nikkei, as reported by Tom's Hardware, says Tower holds more than 80% of contract production of optical communications chips used in servers, with Marvell among its customers [11]. At that share, Tower's capacity schedule is most of the contract market's schedule [11]. Tom's Hardware ties the project's size to fast-growing demand for optical connectivity in AI infrastructure [12]. The report does not include lead times, backlog, or evidence that data-center builds are waiting on optical parts. On this evidence, the dominant contract supplier is committing capital now against demand it expects in 2029 [2][11].

Part of the money is for a market where Tower says it trails [13]. The company concedes that Intel, TSMC and GlobalFoundries are ahead in co-packaged optics, and plans to bring CPO and its precursor technologies to Uozu [13]. It has not disclosed details of that roadmap [13].

What to watch

  • Whether Track One reaches full production readiness in the fourth quarter of 2027 as scheduled.
  • Construction and qualification milestones for the Track Two fab beside Fab 7 ahead of its expected 2029 contribution.
  • Any CPO roadmap Tower publishes for Uozu, given it concedes Intel, TSMC and GlobalFoundries lead.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories