Tower's Silicon Photonics Math Now Outruns Its Own Business Model
A record quarter, contracts extended into 2029, and a SiPho run rate heading for $2.8 billion — on a call where the CEO declined to concede his own $3.6 billion target is already stale.
TSEM · Earnings Call · 2026-11-02
The quarter that arrived at the wrong moment
Tower Semiconductor reported record revenue of $526 million for the September quarter, up from $396 million a year ago, with net profit of $120 million and a 23% net margin against 14% a year earlier. Fourth-quarter guidance of $600 million at the midpoint would put the full year at roughly $2 billion versus $1.57 billion in 2025. The CEO framed the exit rate as the real event: “Achieving a $2.4 billion revenue run rate is most significant as our demand and start plans do not indicate the typical industry first quarter seasonality.” — Russell Ellwanger, Chief Executive Officer · 2026-11-02 Management then went further, targeting a first-quarter 2027 starting point more than 45% above the first quarter of 2026. (A small crack worth noting: the prepared remarks cite 33% growth from the CEO and 36% from the CFO's script; the two revenue figures actually given imply 33%. No one corrected it.)The awkward part is the tape. Over the last 90 days the market's AI data center basket has been one of the worst-performing theme groups anywhere in the coverage set — dozens of names in the red, with co-packaged optics down more than a quarter on fourteen decliners and high-speed interconnect themes down further still. Tower is asking investors to underwrite a near-vertical capacity ramp into precisely the corner of the market the market has spent a quarter de-rating.
The math that outran the model
Here is the arithmetic that matters. Silicon photonics contributed roughly $240 million of revenue in the third quarter alone. Capacity added by year-end is double the start rate that supported it, which management says puts the fourth quarter at a SiPho annualized run rate above $1.1 billion. Then, with the Arai facility in Japan, capacity triples versus the level behind those $240 million shipments — something on the order of $2.8 to $2.9 billion of annualized SiPho capacity by third-quarter 2027 start rates.Set that against the company's own business model: $3.6 billion of revenue, $1.38 billion of operating profit, $1.2 billion of net profit at a 33% net margin. The model was lifted in the summer from a February version of roughly $2.84 billion and $750 million of net profit — a $760 million revenue raise that management says carries 67% incremental gross profit and 59% incremental net profit. Silicon photonics alone, then, would supply nearly four-fifths of the 2028 revenue target, before a single wafer of BCD, RF, image sensor or display ships.
Cody Acree of StoneX put the obvious question directly, suggesting the new model might already be too conservative. The answer was “No comment. No, I think it's an aggressive target. But certainly, we wouldn't be giving a target to the Street that we don't have very, very high confidence that we will meet.” — Russell Ellwanger, Chief Executive Officer · 2026-11-02 That is a CEO with a model he cannot yet defend walking off the page, and the discipline is deliberate: the $3.6 billion figure assumes 85% utilization across every fab, and SiPho revenue ramp math is a start-rate argument, not a shipment one.
The margin engine behind it also has a lag. This quarter's incremental gross margin hit 62%, but the CFO flagged that 65% of the $920 million SiPho/SiGe capital program has been paid, meaning the depreciation on the rest — plus the Japan tools — is still ahead: “We repeat our same forecast of 55% to 58% incremental gross profit.” — Oren Shirazi, Chief Financial Officer · 2026-11-02 Gross margin was 34% in the quarter; the model asks for 45%.
What is genuinely new this quarter
Three things. First, the long-term agreements extended from 2028 into 2029: “We have secured customer demand commitments for both 200-millimeter and 300-millimeter production and now having extended it to include 2029.” — Russell Ellwanger, Chief Executive Officer · 2026-11-02 That is one additional year of contracted visibility, and on the BCD side those agreements carry forecast blended prices 20% above 2026 levels — a striking ask for an analog foundry.Second, the laser. Management announced initial production integrating indium phosphide lasers directly into the silicon photonics die through a die-to-wafer attach of InP chiplets, with thin-film lithium niobate and 400-gigabit-per-lane modulators queued behind it. The arc here is exquisite: in May the CEO described indium phosphide as a bottleneck — “indium phosphide right now is constrained ... the starting material is the constraint” — Russell Ellwanger, Chief Executive Officer · 2026-05-13 — and by November the constrained material has become an integrated product feature. It is also a theme the supply chain is confirming independently: both AXT and Aixtron leaned on indium phosphide wafer and capacity economics in their own recent calls.
Third, near-package optics. More than ten top customers are prototyping NPO, all of them existing pluggable customers, and management argues it can roughly double the optical transceiver market while pulling SiGe TIA and driver content in as DSPs get displaced. This is the right sequencing — pluggable share funds the transition — but the revenue expectation is a second-half-2027 event, exactly where the existing capacity ramp also lands.
For Kevin Costner in the Field of Dreams, if you build it, they will come as may be true. For us, we need very strong customer partners ... when we build it, it will be sold because we are dealing with the leaders.
What management stopped saying
The vocabulary shifted. Last quarter Track 1 and Track 2 were the top two phrases in the company's own keyword set; this quarter both vanished from it, replaced by blunt multipliers and "the Arai facility, formerly Fab 6." Advanced packaging — a recurring bridge topic through the summer — also fell away, its role absorbed into the laser die-attach discussion. When a management team retires its own project nomenclature, it usually means the story no longer needs scaffolding.The more interesting silence is on competition. In May, asked where share goes as GlobalFoundries and TSMC scale, the CEO was dismissive: “There's other people that claim very high market shares, I don't see how it's possible at all.” — Russell Ellwanger, Chief Executive Officer · 2026-05-13 In August he conceded he could not handicap the industry's capacity additions at all. This quarter he handed the question to President Marco Racanelli, who offered “it's hard always to pinpoint market share, particularly in the market that is growing as quickly and rapidly as this is growing” — Marco Racanelli, President · 2026-11-02 — before adding that with growth this fast "there's obviously room for more than one player." That is a materially softer posture than six months ago, and investors should decide whether it reflects diplomacy or a field that has genuinely filled in.