Tower's Profit Doubled, Its Market Value Shrank 20%. Here's Why

The headline results:

Line the quarters up and the trajectory this year is steep:

Each quarter adds roughly $40-50 million of revenue and three to four percentage points of gross margin, while net profit grows by nearly half again. How much does a move in gross margin from 26.8% to 34.0% matter? Think of a factory that used to spend $73 to make every $100 of goods it sold and now spends only $66. Almost all of that $7 saving drops straight to profit, because fixed costs like depreciation and R&D didn't rise along with revenue. That is operating leverage at a capital-heavy contract chipmaker: profits spring upward quickly when things go well, and fall just as quickly when they don't.

The math for the full year is simple. Revenue for the first three quarters totals $414 million + $460.1 million + $526.2 million = $1.40 billion. To hit the $2 billion full-year guidance, Q4 has to deliver $600 million, a 14% sequential increase that lands exactly on the guidance midpoint. The company has left itself no cushion.

One number in management's own words needs unpacking. The CEO spoke of "closing the year at $2 billion annual revenue and a $2.4 billion annualized revenue run rate" (company 6-K, 2026-11-02). The $2.4 billion is simply Q4's $600 million multiplied by 4, an annualized run rate, not formal guidance for full-year 2027. So reaching $2.4 billion in 2027 requires holding $600 million in every quarter next year, and the first quarter is usually a slow season for chips. In the same release the company wrote, "We target to begin 2027 avoiding first quarter's typical industry seasonality," meaning it aims to break the usual Q1 slowdown. The one number to watch in the next quarterly report is whether Q1 2027 revenue guidance comes in at $600 million or higher. If it does, the $2.4 billion run rate turns from slogan into trajectory. If it comes in below $570 million, that is an admission that the slow season is business as usual, and the market will immediately redo its math for next year.

The demand evidence centers on silicon photonics (SiPho for short: using standard silicon manufacturing to build the tiny light channels and modulators on a chip, the core component inside the optical modules that move data around data centers):

Operating cash flow was $185 million and capital spending was $191 million, leaving free cash flow for the quarter at negative $6 million. A company with a 23% net margin is still burning cash. Is that a red flag? Not so fast. Let's take it apart.

Free cash flow is hovering around zero because expansion spending is eating all of the operating cash flow, while the profits themselves really are turning into cash. The real question is the bill still to come. The dual-track expansion plan the company announced in July totals about $3 billion (company 6-K, 2026-07-14). The terms of the subsidy from Japan's Ministry of Economy, Trade and Industry (METI) are listed separately among the risk factors, which tells you the timing and conditions of that money are not fully settled. At a spending pace of $190 million per quarter, that's $760 million a year. If the $3 billion is to be spent within three to four years, spending will need to step up another notch.

Right now there are four places the money can come from:

On dilution: Q3 GAAP net profit of $120 million divided by diluted EPS of $1.05 implies roughly 114 million diluted shares. None of this year's three quarterly releases shows any new share issuance, so existing shareholders have not been diluted to fund the expansion. That's unusual for a capital-heavy company in the middle of a build-out. In effect, the prepayment model has customers fronting part of the cost of Tower's new capacity. The trade-off is that revenue and capacity end up tied to a handful of big customers, which the next chapter covers in detail.

On earnings quality, the gap between GAAP net profit of $120 million and adjusted net profit of $131.3 million is $11.3 million, mainly stock-based compensation and amortization of acquired intangibles. That is less than 10% of net profit, so the two measures are close. Adding depreciation and amortization back to operating profit gives a rough Q3 EBITDA of about $218 million (calculated from the company 6-K cash flow statement).

Tower's place in the chain is that of a manufacturer for hire: customers bring the chip designs, and Tower builds the chips using its own production processes.

Upstream, the bottlenecks are in three places. Equipment: a 300mm SiPho production line needs lithography, etching and epitaxy tools from a small number of global equipment makers, and their delivery times will determine whether the Japan expansion ramps on schedule. Facilities: the Japan expansion involves converting the Arai plant and building a new facility next to Fab 7, and the release specifically lists the Arai conversion as a risk. People: the lines in Israel, Japan and Agrate in Italy all need specialized process engineers, though the release does not disclose any specific shortfall. As for raw materials, silicon wafers and specialty gases are standard products, and we found no public disclosure of supply shortages.

Downstream is where the story really lives. SiPho chips are sold to chip designers and module makers building 800G and 1.6T optical modules, and the ultimate buyers are cloud companies building AI clusters. The market widely regards Marvell as one of the most certain customers of Tower's SiPho platform (market analyst view, not confirmed by the company). The company's own releases say only that the $1.3 billion of 2027 agreements come from its "largest customers," without naming them or disclosing any single customer's share.

The contracts need to be read in two layers. The $1.3 billion is "expected revenue," an estimate under a framework agreement, not a firm, non-cancellable order. What actually binds customers is the portion of prepayments tied to minimum capacity commitments. That is real money customers have already handed over, up a net $305 million in the first nine months. In other words, firm obligations are in the hundreds of millions of dollars, the headline ceiling is over a billion, and the gap between the two has to be filled by customers actually shipping optical modules.

The weakest link is customer concentration. If one or two big customers account for most of SiPho revenue, and their own optical designs are shifting from pluggable modules toward CPO (co-packaged optics, where the optical engine sits right next to the switch chip in the same package), then any design change or the addition of a second-source supplier would hit Tower's factory utilization directly. GlobalFoundries, STMicroelectronics and Samsung have all publicly announced SiPho capacity expansions, and this risk becomes more concrete in 2028 to 2029.

Third-quarter revenue growth across a group of chip companies puts Tower on a bigger map (aggregated peer filings, reporting period 2026Q3):

Gross margin tells you more than revenue. Tower's Q3 gross margin was 34.0% and UMC's was 34%, essentially identical, while TSMC is guiding to 65% to 67% for next quarter (aggregated peer filings, reporting period 2026Q3). So Tower's revenue growth is already running eight percentage points ahead of UMC's, yet its profit structure still looks like a mature-technology contract manufacturer. The high margins of SiPho aren't fully showing up in the consolidated numbers yet. This is the most important card in the long-term bull case: if SiPho keeps growing as a share of the business, gross margin has room to move above 40%. Conversely, if gross margin gets stuck around 35%, it means SiPho pricing power is weaker than hoped.

Two big companies that have already reported this quarter show how the market is pricing similar report cards (peer earnings reactions, data as of 2026-11-02):

Neither stock moved as much as options had priced in, and TSMC even closed lower after raising guidance. That suggests the market already expects strong results from the AI supply chain, and beating estimates no longer earns a premium on its own. That said, the historical link between these two stocks' earnings reactions and Tower's is of limited use because the sample is small, so treat this as background on market mood, not a reliable pattern. Tower rose 5% the day after its report, a stronger move than either of its bigger peers. But for a stock whose market value has shrunk by more than 20% since the summer, that looks more like a rebound from oversold levels.

The price-to-earnings ratio is usable here, but only on a run-rate basis. Tower's profit is rising every quarter, so trailing twelve-month EPS badly understates its earning power. For a capital-heavy company in expansion mode, enterprise value to sales (EV/Sales), read alongside the rising gross margin, is the steadier yardstick.

What does 35x mean? If Q4 profit grows another 20% from Q3, the forward P/E naturally drops below 30x, so this is a multiple that can digest itself if the next quarter delivers. At 7.2x EV/Sales, though, a contract manufacturer with a 34% gross margin isn't cheap. It means the market has already paid in advance for two things: SiPho revenue reaching $1.3 billion or more in 2027 as planned, and gross margin continuing to climb. Compared with what was priced in at a $24 billion market value in the summer, today's price has already baked in some worry about SiPho's ceiling, but it is nowhere near valuing Tower like an ordinary mature-technology foundry.

Tower sits on two cycles at once. Its traditional business (radio-frequency chips, power management and image sensors) follows the mature-technology cycle of industrial and automotive demand, which bottomed in 2025 and is recovering modestly in 2026; NXP's and Texas Instruments' third-quarter growth of 19% to 28% confirms that. The SiPho business follows the speed-upgrade cycle of data center optical modules, which is now at the stage where 800G is shipping in high volume and 1.6T is starting to ramp. That cycle has far more momentum than the mature-technology one.

Three types of readings help locate where we are:

Our view: the optical connectivity cycle that SiPho belongs to is in the middle of its upswing. Demand and capacity are both expanding, but there is no sign of overbuilding yet. The mature-technology business is in the early part of its upswing. The turning point to really watch for is the second half of 2027 into 2028, when new capacity from Tower in Japan and SiPho expansions at GlobalFoundries and STMicroelectronics come online together.

Two trackable early signs that the cycle is peaking:

This chapter deals with week-to-week signals that govern timing over the next few weeks. Readers planning to hold for a year or longer can skip it.

The overall market is leaning bearish. As of 2026-11-03, our market-wide bull/bear reading stands at negative 33 (on a scale of plus or minus 100), the S&P 500 ETF has slipped 1.79% below its 20-day moving average, and the cross-asset rotation signal reads risk_off. Macro liquidity is tightening too: net liquidity as measured by FRED data is $5.75 trillion, down 1.0% week over week, and the 10-year Treasury yield is 4.91% (all data as of 2026-11-03).

The most direct hit to Tower comes from the chip sector itself. Among the three sub-sectors with the heaviest money outflows on 2026-11-03, the semiconductor ETF (SMH) ranks second with an outflow intensity of negative 65, behind only lithium batteries at negative 74. Money leaving the technology sector (XLK) is moving into real estate (XLRE, intensity 4.62) and materials (XLB, intensity 4.49), heading toward defensive and hard-asset areas. The biggest inflows are going to exchanges and financial data, storage REITs and chemicals, none of which has anything to do with the AI supply chain.

Specialty contract chipmaking has no sector index of its own whose money flow can be tracked, so the read has to come from the broader semiconductor sector. The rotation themes currently forming or already confirmed are consumer discretionary, industrials and healthcare, and semiconductors are not among them. In trader's terms: institutions are cutting their chip holdings. Tower's 5% jump the day after earnings was its own report card pulling it partway out of the sector, so the share price has temporarily run ahead of the money, and the money hasn't come back yet.

Long term (quarters to years): worth following. Revenue up 33% year over year, gross margin up ten percentage points in a year, net profit doubled, Q4 guidance above consensus, a 2028 revenue target raised from $2.8 billion to $3.6 billion, customers locking in capacity with prepayments, and no dilution of existing shareholders so far during the expansion. This set of fundamentals ranks in the top tier among chip companies. A forward P/E of about 35x on a run-rate basis and EV/Sales around 7x already price in a good deal of SiPho success, but as long as Q1 next year holds at $600 million in revenue, profit growth will naturally grind those multiples down. There are three costs to accept: heavy concentration in downstream customers, possible SiPho overcapacity around 2028, and free cash flow stuck near zero for the length of the expansion. The way to hold it is to follow the checkpoints: whether Q4 revenue reaches $600 million, whether Q1 2027 guidance is at or above $600 million, and whether gross margin breaks above 35%. As each one is delivered, keep holding. If any one falls short, lower your expectations for the position.

Short term (days to weeks): don't chase; wait for the money to come back. Semiconductors are among the sectors seeing the heaviest outflows, the broad market leans bearish, liquidity is tightening, and Tower is still about 20% below its 50-day and 200-day moving averages. The post-earnings bounce is just a recovery within a downtrend. The moving averages haven't turned up and the money hasn't returned, so buying now is a bet against money draining out of the sector. Wait for one of two things: semiconductors dropping off the list of heaviest-outflow sectors, or the stock climbing back above its long-term moving averages without breaking below its pre-earnings low.

What counts as money coming back, which levels are reasonable for building a position in stages, and where to admit you're wrong are covered in the next two chapters.