Tower’s miss is the setup: Q2 revenue disappointed, but the RF and SiPho capacity math points to a back-half mix inflection

The clean read on this print is that the Street was positioned for a revenue recovery that arrived more slowly than modeled, while Tower delivered better earnings than that revenue line should have allowed. On the street-comparison basis, revenue was $372.1 million against $395.0 million, a -5.8% surprise, so the top-line miss was not rounding noise. EPS went the other way, with actual $0.50 versus $0.47, a +6.4% surprise, which tells us the quarter’s investable question is not whether demand was below the Street’s Q2 bar, but whether Q2 marked the trough in revenue quality. Management’s own reported basis is different from the Street basis and should not be conflated: Oren Shirazi put the company’s quarter at “$0.42 basic and $0.41 diluted earnings per share,” while the print basis shows the $0.50 beat. The market may be treating the miss as evidence that Tower’s specialty foundry recovery is still hostage to broad industrial and mobile timing; the counterargument is that the company’s internal mix targets are now specific enough to make Q2 look like the last quarter before utilization and mix begin to matter again.

That distinction between what was priced and what surprised matters because the print punished the wrong line item if investors stop at revenue. What was priced in was a Q2 revenue level close to $395.0 million, effectively pulling some of the guided Q3 level into the June quarter. What actually surprised was the gap between the Street’s timing assumption and management’s sequential roadmap: the company reported $372 million in Q2 revenue on its own basis, then guided Q3 to $395 million plus or minus 5% and targeted more than $40 million of additional Q4 revenue. Russell Ellwanger’s phrasing is important because it is not a generic “better second half” comment: “We guide our third quarter revenues to be $395 million, plus/minus 5% and additionally, target a $40 million-plus revenue increase for the fourth quarter over the third quarter.” That sentence commits to a shape, not just a direction, and the magnitude is enough to reframe the revenue miss as a pushout if the company executes by Q4.

The financial trajectory supports that interpretation, because revenue has been pinned in a narrow band for much of the last cycle while margin has only recently started to rebuild from the trough. The Q2 FY2025 revenue base of $372.1 million is still below the Q4 FY2024 level of $387.2 million, so it is fair to say the June quarter did not yet prove a full recovery. But gross margin at 21.5% improved from 20.4% in the prior quarter, while Q3 FY2025 in the quarterly history shows 23.5%, indicating the next leg is about mix and loading rather than a one-quarter expense anomaly. The EPS line also backs the idea that incremental revenue should matter: diluted EPS rose from $0.35 to $0.41 on company quarterly history while revenue moved from $358.2 million to $372.1 million, a small revenue change with visible earnings sensitivity. That is the setup the Street risks missing: Tower does not need a foundry-wide boom to re-rate; it needs the guided RF and SiPho mix to lift a revenue base that has been stuck below the full-loading ambition.

The margin story is inseparable from capacity, because Tower is spending into specific platforms while still carrying near-term depreciation and cash demands. Shirazi disclosed $80 million of gross profit and $40 million of operating profit for Q2, each higher than the prior quarter by $7 million, which is the first proof point that incremental volume is dropping through despite the current investment phase. The company is not funding a vague expansion plan: it has approved $1.15 billion of CapEx across Agrate, 11X, and SiPho/SiGe, with a separate $350 million investment aimed at high-margin SiPho and 5G capacity. The funding profile explains why free cash flow may not screen cleanly even if the thesis is working, since management described quarterly CapEx running “between $100 million to $120 million a quarter.” In other words, investors should not demand near-term cash conversion as the primary confirmation signal; the confirmation should be revenue moving through the underloaded fabs and gross margin recovering toward the mid-20s already visible later in the history.

The segment color makes the back-half revenue bridge more credible than a typical second-half seasonal guide, because the company identified multiple sources of growth with magnitudes rather than anecdotes. RF infrastructure represented 25% of corporate revenue in Q2, with over $90 million in revenue, up from 14% in the same period of 2024. RFSOI in mobile showed a Q2 to Q1 revenue increase of over 20%, and management expects close to 30% Q3 over Q2, with another increase targeted in Q4. Sensors and displays are also expected to rise about 20% in the second half of 2025 versus the prior quarters and prior-year run rate, mainly from machine vision. The critical point is that these are not all the same end market: RF infrastructure, mobile RFSOI, and machine vision do not have identical cycle timing, so the Q4 target does not rest on one customer product launch or one inventory correction.

Silicon Photonics is the sharper variant-perception lever, because it changes Tower’s customer relevance from specialty capacity supplier to beneficiary of datacenter optical speed migration. Management said current wafer starts include 400 and 800 gigabit per second volume and a ramp on 1.6 terabit per second, while separately reminding investors that SiPho revenue was circa $105 million in 2024 with an expectation of doubling in 2025. The capacity endpoint also has a concrete magnitude: Ellwanger said the second half of 2026 capacity should be 33% higher in Silicon Germanium and 2.2x larger in Silicon Photonics than the fourth quarter 2025 targeted shipments. That is the core reason we would not underwrite Tower solely on Q2 revenue versus consensus. If SiPho is moving from a 2024 base of circa $105 million toward a 2025 doubling, and capacity is being built to 2.2x the fourth quarter 2025 shipment level, the market’s June-quarter miss framework is too backward-looking.

The obvious pushback is that Tower’s guided growth still depends on utilization, and the current gross margin is not yet at peer-leading levels. That pushback is valid, but it should be aimed at the slope of recovery rather than at the existence of one. Fab 3 is already fully utilized at the company’s 85% utilization model, so at least one asset is not waiting for demand to appear. At the corporate level, the longer-term revenue ambition is far above today’s run rate: management targets $2.7 billion in annual revenue at full loading of existing fabs, including Agrate and New Mexico. The issue is timing, not theoretical capacity. If Q3 holds near the $395 million guide and Q4 adds the targeted $40 million-plus, Tower exits the year on a very different utilization narrative than the Q2 consensus miss implied.

The supply-chain read-through is most constructive for customers tied to analog, RF, power, and image-sensor specialty capacity, and more specific for photonics substrate suppliers. For NXP, Tower’s RF infrastructure contribution at 25% of corporate revenue and RFSOI growth of over 20% Q2 to Q1 point to specialty RF capacity tightening first, not broad commodity foundry strength. For ON Semiconductor, the second-half sensors and displays increase of about 20% is the relevant signal, because the read-through is to machine vision and image-sensor-related specialty manufacturing rather than consumer imaging. For STMicroelectronics, the TPSCo JV exposure makes the Japan capacity discussion relevant, particularly the 12-inch Uozu fab included in the $350 million SiPho and 5G expansion. For Soitec, the implication is direct but narrower: Tower’s PH18 Photonics-SOI platform sits in the supply chain exactly where 1.6 terabit per second wafer starts and a 2.2x Silicon Photonics capacity plan would pull through substrate demand.

The peer comparison also argues that Tower is not being valued for the right benchmark if investors compare it only with scale foundries. GlobalFoundries reported $1,634.0 million of revenue, 27.6% gross margin, and +3.1% revenue YoY, while Tower’s latest peer-table quarter shows $413.6 million, 26.8% gross margin, and +15.5% revenue YoY. The relevant point is not that Tower is close to GlobalFoundries in scale, because it is not. The point is that Tower is already within 0.8 percentage points of GlobalFoundries’ gross margin in the peer table while growing faster on that same table, and the company is layering capacity into SiPho and SiGe rather than chasing generic wafer starts. Against UMC’s 29.2% gross margin and +5.5% revenue YoY, Tower still has margin catch-up to prove, but its revenue growth profile in the table is not the weak link.

The call delivery reinforces the idea that management was deliberately trying to redirect attention from Q2 to the back-half ramp, though the tone data also flags why investors may not give full credit immediately. The tone history shows Q2 FY2025 guidance_tone at 0.83, the highest in the table, while sentiment was 0.45. That optimism came with tone_confidence of only 0.25 and uncertainty of 55.9, so the transcript was assertive on targets but not clean enough to remove execution risk. This is a useful tension rather than a reason to dismiss the guide: management gave unusually explicit revenue steps, but the language model scores say the call still carried elevated uncertainty. That matches the fundamental setup, where the investment case hinges on whether named platform ramps convert into the Q4 revenue step rather than whether Q2 was cosmetically better than expected.

That delivery pattern became more credible in subsequent call history, but Q2 remains the event where management laid down the measurable bar. The tone table’s later Q1 FY2026 call-over-call delta shows uncertainty down by -8.4 and qa_evasiveness down by -21.6, while guidance_tone improved by +0.06. Those numbers do not prove Q2 execution by themselves, but they suggest management’s forward commentary did not collapse into evasiveness as the cycle progressed. The conflict is that Q2 itself had the highest guidance_tone at 0.83 while tone_confidence was just 0.25, so investors should treat the Q2 guide as a high-conviction plan with limited transcript-level proof at the time. That is exactly why the stock reaction to a revenue miss can create opportunity: the burden of proof is real, but the target is visible.

The investment conclusion is therefore not “buy the miss” in a generic sense; it is buy the misclassification of the miss if the market is treating Tower as a slow-growth specialty foundry rather than a mix-transition story. The EPS beat of +6.4% showed that earnings power did not break when revenue missed by -5.8%, and the guide offered a path from $372 million in Q2 reported revenue to $395 million in Q3 and $40 million-plus additional Q4 revenue. The highest-quality confirmation would be gross margin moving with that revenue rather than lagging it, because the thesis depends on RF infrastructure, RFSOI, and SiPho carrying better economics than the corporate average. The highest-risk break would be a revenue guide reset without an offsetting explanation from SiPho, RF infrastructure, or sensors, because that would turn the Q2 miss from timing into demand.

What to watch next is precise. For Q3 FY2025, the first hurdle is revenue inside the $395 million plus or minus 5% guide; a result below that range would break the pushout thesis, while a result near $395 million with gross margin near the 23.5% quarterly-history level would confirm that mix and utilization are starting to flow through. The second hurdle is the Q4 target: management needs to maintain the $40 million-plus revenue increase over Q3, because that is the number that turns Q2 from disappointment into transition quarter. The third hurdle is segment evidence, not adjectives: RF infrastructure should remain anchored around the 25% of corporate revenue level or expand, RFSOI should show the expected close to 30% Q3 over Q2 increase, and SiPho commentary should stay consistent with the 2025 doubling from circa $105 million in 2024. If those figures hold on the next call, the market will have to underwrite Tower on the 2026 capacity path, including 33% higher Silicon Germanium capacity and 2.2x larger Silicon Photonics capacity. If they do not, the Q2 revenue miss was the first warning, not the last.