Earnings Call Highlights: Tower Semiconductor (TSEM) | 2026 Q2

2 days ago

7 min read

• This was Tower's "operating leverage proof quarter": revenue hit a record $460.1M, up 24% YoY, but more importantly gross margin jumped to 29.9% and net margin to 19.7%, both company records. It shows the product mix upgrade driven by silicon photonics (SiPh) is converting each dollar of incremental revenue into profit with very high efficiency.

• Silicon photonics is now the main engine, not a narrative: SiPh revenue grew over 60% QoQ and over 270% YoY, the Q2 annualized run rate has passed $680M, and the company is targeting a $1B annualized run rate in Q4 2026. The large RF Infrastructure segment grew 140% YoY, pushing Tower from "specialty process foundry" to "key AI optical interconnect supplier."

• Management moved the entire 2028 model up: the new model is raised to $3.6B revenue, 45% gross margin and 33% net margin, driven by dual-track 300mm expansion in Japan. Next-quarter guidance is $520M, putting the annualized run rate above $2B.

This article is for technology and industry trend analysis only and does not constitute investment advice.

1. Three signals to remember first

First, the jump in earnings quality matters more than the record revenue. Revenue rose 11% QoQ, but gross profit, operating income and net income captured 58%, 55% and 55% of incremental revenue respectively. That "incremental drop-through" means the new capacity is selling high-ASP, high-margin silicon photonics wafers, not filling fabs by cutting prices.

Second, this is not a one-off good quarter but "the first step of sustained expansion". Management repeatedly stressed that Q2 profitability is only the start of margin expansion, and locked in that slope with the raised 2028 model.

Third, growth and efficiency are happening together. As revenue scales, operating expenses are being pushed down to about 7% of revenue (30% lower than today), while R&D is set to rise by over 40%. The savings come from SG&A and the additions go to technology, which is how a speed-first tech company should look.

2. Revenue and financials

This quarter's hard numbers (GAAP/comparable basis, USD):

• Revenue $460.1M, up 24% YoY (year-ago $372.1M) and 11% QoQ (prior quarter $413.6M), a record high.

• Gross profit $137.8M, gross margin 29.9%, up 72% YoY; gross margin rose about 8.4 percentage points YoY (year-ago about 21.5%). This is the weightiest number in the whole report.

• Operating income $90.3M, operating margin 19.6%, 2.26x the year-ago operating income.

• Net income $90.8M, net margin 19.7%, up 95% YoY (year-ago $46.6M).

• Diluted EPS $0.79 (basic $0.80), versus about $0.41 diluted a year ago, nearly doubling.

• The balance sheet remains solid: shareholders' equity hit a record $3.1B and the current ratio is about 4.9x; in Q1 the company received about $290M in prepayments from SiPho customers, mostly tied to 2027 capacity reservations.

A framework worth remembering: Tower's gross margin and operating margin are now only about 10 percentage points apart, a very efficient structure for a company still investing heavily in future growth.

3. Technology and business highlights

Silicon photonics is lifting the whole segment table. RF Infrastructure made up 49% of revenue this quarter, up about 43% QoQ and over 140% YoY; within it, silicon photonics itself grew over 60% QoQ and over 270% YoY, with a Q2 annualized run rate above $680M. The company was explicit: once this capacity investment completes its full wafer-start ramp in Q4 2026, wafer-start capacity will be more than three times Q2 shipments, with the full financial impact showing up in Q2 2027. In other words, the $1B annualized run-rate target is backed by capacity and contracts.

Dual-track 300mm expansion in Japan is the real engine behind the model upgrade. Three weeks earlier, Tower announced a dual-track expansion in Japan with support from the Ministry of Economy, Trade and Industry (METI):

• Track 1: convert the Arai fab (formerly Fab 6) into a 300mm silicon photonics and advanced packaging line and maximize 300mm output at Fab 7 (Uozu), with volume production expected in Q4 2027. This track drives the raised 2028 model, and its capacity is "already reserved and committed by several lead customers."

• Track 2: build a new 300mm fab next to Fab 7, lifting Japan's 300mm capacity to about 4x, focused on silicon photonics, silicon germanium (SiGe) and advanced optical packaging, targeting NPO and future co-packaged optics (CPO). Tool installation is targeted for completion in Q4 2028 to support growth from 2029 onward.

Near-packaged optics (NPO) is next year's new battleground. As AI clusters scale from thousands to hundreds of thousands of XPUs and electrical interconnect nears its physical limits, silicon photonics has become the main platform for 800G/1.6T pluggables; next, scale-up (in-rack and cross-rack) is NPO's turn. NPO uses the same ecosystem as pluggables, neatly sidestepping hyperscalers' concerns about reliability and serviceability. How this timeline plays out and how long the pluggable window is extended are broken down in full in CPO volume ramp pushed to 2028, but this is not bad news.

The depth of co-development with customers is the moat. The SiPho coherent optical modules from Tower's long-running partnership with Marvell crossed the "millions of units" cumulative shipment mark this quarter, proof that one of the industry's most complex silicon photonics ICs can be mass-produced at high yield. For more context on Marvell's push to bring light into the chip with silicon photonics, see Celestial AI explained: the "light into the belly of the chip" Marvell bought for $3.25 billion.

4. Management outlook

Next-quarter guidance: revenue $520M (±5%, range $494–546M), up about 13% QoQ and about 31% YoY, putting the annualized run rate above $2B. The wording was "confirmation" rather than "expectation": the company said early in the year that the second half would be strong, as previously announced capacity investments are qualified and converted into shipments.

Raised 2028 model (versus the February version):

• Revenue $3.6B, $760M (+27%) above the February model.

• Gross profit $1.63B, gross margin 45% (prior version 39%); incremental revenue brings in 67% incremental gross margin.

• Operating income $1.38B, operating margin 38% (prior version 32%).

• Net income $1.2B, net margin 33% (prior version 26%, versus 20% this quarter).

Note the layering in management's tone: $3.6B is a "target run rate, nominally achieved for full-year 2028," a target backed by contracts and capacity; Track 2's quantified contribution is explicitly "not yet given," and the company said it will update the long-term model in early 2027 alongside final construction and tool-installation plans. The model's three key assumptions were also spelled out: wafer pricing, cost and installation/qualification timing, and 85% fab-wide utilization. The CFO singled out "price per wafer" as the most variable and most important variable to watch.

5. Supply chain and customer clues

• The indium phosphide (InP) shortage has been covered with a long-term contract. Tower signed a multi-year InP epitaxial (epi) supply agreement with IQE, locking in III-V epi material while bringing key processes in-house; integrated lasers (indium phosphide) will generate tens of millions of dollars in revenue next year. Why this upstream "engine of light" is the deepest bottleneck in AI optical interconnect is mapped across the full supply chain in The "engine of light" shortage map.

• Exclusive ties with lead customers. Management said plainly that for co-developed modules, lead customers get a head start and the company asks for 100% market share, with exclusivity agreements on both sides. That is Tower's basis for holding an "exclusive or majority share" as peers such as GlobalFoundries, STMicro and Samsung expand capacity.

• Figures of merit are the real differentiator. The company named insertion loss as one of the most critical metrics: low insertion loss lets customers buy fewer expensive CW lasers, or even halve the number of lasers in a package (the earlier collaboration with InnoLight is one example). Add the 400G silicon modulator with Coherent and its TFLN and indium phosphide modulator plans, and staying one to two generations ahead is the moat.

• Utilization reveals the pace. The 200mm Fab 2/3/9 run at 80–85% utilization, Fab 5 in Japan at 75%, and Fab 7 is fully loaded and well above the 85% model assumption. Fab 7 being "overbooked" is the most direct thermometer of SiPh demand.

• RF Mobile is a strategic gear shift, not weakness. 300mm RF SOI fell 14% YoY because of the move from 200mm to 300mm and the concentration of 300mm RF SOI in Fab 7 to free up SiPh/SiGe capacity; design-win momentum has the company expecting 300mm RF SOI wafer starts to triple by mid-2027.

Conclusion

This quarter Tower made one important thing clear: silicon photonics is no longer just a growth story; it is the engine that has already rewritten the company's profit structure. Record 29.9% gross margin and 19.7% net margin, plus a new model pushing the 2028 net margin to 33%, amount to management vouching for today's expansion with the next three years of numbers.

From next quarter into next year, three things are worth watching: first, whether silicon photonics reaches a $1B annualized run rate in Q4 2026 on schedule; second, NPO's share of silicon photonics shipments (management says it will be a "double-digit percentage" in the second half of 2027); third, when Track 2's quantified numbers appear with the long-term model update in early 2027. If two of the three come through, the growth slope supported by Japan's 300mm capacity holds.

The real risk is simple: the entire 2028 model rests on three pillars, "wafer pricing, installation timing and 85% utilization," and pricing is the most sensitive. With GlobalFoundries, STMicro and Samsung all expanding supply at once, whether prices hold will directly decide whether the 45% gross margin is a castle in the air.