After a record-breaking financial report, it plunges 11.85%: Is TSEM expensive?

NEXT PICK · Market Insights

After a record-breaking financial report, it plunges 11.85%: Is TSEM expensive?

Wednesday, August 5, 2026

A company just delivered a record-breaking quarterly financial report—$460 million in revenue, up 24% year-over-year, net profit of $91 million, nearly doubling year-on-year, a record gross margin of 30%, and a new $520 million guidance for next quarter—then its stock price was plunged nearly 20% over two trading days, plunging 11.85% on August 5 to close at $211.14.

Trading volume has expanded to 1.6 times the daily average over the past three months. A top financial influencer on X mentioned it three times within 24 hours, but not about "why prices are rising," but rather "whether the 2028-2029 capacity race will drive prices down."

This company, called Tower Semiconductor, code TSEM, has risen 314% over the past 12 months. When a perfect financial report turns into a big bearish candlestick, the problem often isn't the financial report, but the price that has already written the report into it.

Latest price $211.14 ▼ -11.85%

Silicon photonic aging $680 million ▲ +270%

December return +314% ▼ -15% from the 50-day moving average

An Israeli foundry that does not compete with TSMC for 3nm technology

Founded in 1993 and headquartered in Migdal Haemek, Israel, Tower is an independent wafer foundry. But it never goes head-to-head with TSMC or Samsung in the advanced process track.

What it does is "special craftsmanship"

Germanium silicon (SiGe), silicon photonics (SiPho), mixed-signal CMOS, RF CMOS, CMOS image sensors, non-imaging sensors, and integrated power management. End-users cover consumer electronics, personal computers, communications, mobile phones, automotive, industrial, aerospace, and medical devices.

If TSMC builds highways, then Tower is about building overpasses, tunnels, and drainage systems—each line may seem insignificant, but without any of them, the entire road can't run.

Why AI data centers suddenly can't do without it

What truly propelled Tower from a "cyclical stock" to an "AI stock" was Silicon Photonics. As data centers move from 400G to 800G and even 1.6T, the physical limits of copper wire have been reached, and light must enter the chip interior.

This requires foundries to integrate photonic devices with silicon-based CMOS circuit monoliths, with tolerances for epitaxial growth temperature, etching depth, and doping concentration all at the nanometer level. This is based on formulas accumulated over more than a decade of mass production feedback cycles, which cannot be replicated by buying a few machines.

On February 5, 2026, Tower announced a partnership with NVIDIA to advance 1.6T data center optical modules, doubling the speed of the previous silicon photonics solution. In Q2, silicon photonics revenue grew by more than 270% year-on-year and 60% quarter-on-quarter, with an annualized operating rate reaching $680 million.

Three-layer locking keeps customers soldered onto the production line

The first layer is process customization. Once a customer's chip completes tape-out verification on Tower's platform, switching factories means redoing the entire process design kit for adaptation, re-tape-out, and re-certification, with a cycle of 6 to 12 months. AI data center products have a lifecycle of only 12 to 18 months—half a year late means missing an entire generation.

The second layer is the design enablement platform and process transfer optimization services. Tower not only manufactures wafers but also sends engineering teams into the customer's early design stage. During the 18 to 30-month design cycle, engineers on both sides are constantly together. This coupling between people and systems is itself a barrier.

The third layer is the contract and advance payment. Tower has secured $1.3 billion in silicon photonics contract revenue for 2027 and received $290 million in customer prepayments to lock in capacity. So here's the question: when customers are willing to pay upfront for your capacity, is this a moat, or an illusion left by the industry's most tense moment?

Those "unavailable" valuation indicators have actually all been in place

Research materials once labeled TSEM's price-to-earnings ratio, price-to-sales ratio, and free cash flow yield as "data missing," but with the Q2 earnings released, all these anchors are now verifiable—and they don't seem cheap.

Based on a market value of $211.14 (approximately $23.8 billion), the rolling P/E ratio is 83.5 times, and earnings per share are $2.53; Rolling revenue was $1.71 billion, corresponding to a price-to-sales ratio of about 13.9 times. The forward P/E ratio has dropped to around 42 times, indicating the market expects earnings to roughly double over the next 12 months.

For reference, the price-to-sales ratio of traditional specialty foundries typically ranges from 3 to 6 times. The Tower premium isn't for its current business, but for its blueprint for 2028.

Cash flow is real, but it's about to be eaten up by the factory

On the cash flow side, this is actually good news. The company has about $300 million in free cash flow, $1.48 billion in cash on hand, and total liabilities of only $142 million, close to $1.3 billion in net cash; Shareholders' equity reached a record high of $3.1 billion, with a current ratio of 4.9 times.

But $300 million in free cash flow corresponds to a $23.8 billion market cap, with a free cash flow yield of only about 1.3%. More importantly, this cash flow is about to enter a large-scale capital expenditure: Japan's dual-track expansion net investment capped at $3 billion, with an additional $650 million planned for silicon photonics capacity, of which $300 million will be used to expand the Migdal HaEmek base in Israel.

Management has clearly stated that the expansion will rely on internal cash generation, with no equity dilution. This is good news for financial discipline, and it also means that free cash flow will continue to be drained by production lines over the next two to three years.

Is it a bit optimistic to use a 1.3% free cash flow yield as a margin of safety during the ramp-up phase?

The drop on August 5 was due to news or fundamentals

It must be made clear

This crash found no negative news at the company level. The financial report is a record, the guidance is a record, and the 2028 model has even been raised. The direct cause of the decline was the overall correction in the semiconductor sector, rising US Treasury yields and energy prices suppressing risk appetite, combined with concentrated realization after a year of the photon rally.

In other words, this is a typical case of "all the good news being exhausted." On the day of the August 4th earnings report, the stock price had already fallen nearly 6%, and on August 5th, it continued to drop 11.85%—the market uses two bearish candlesticks to tell you that this record-breaking report had already been bought once in a 314% increase.

Companies in breakout periods encounter mean reversion prices

From a fundamental perspective, TSEM is in the standard "breakthrough" phase. Revenue acceleration is directly confirmed by 6-K's original text: Q3 guidance of $520 million, a record high, up +31% year-on-year and +13% quarter-on-quarter, with a guidance step ratio of 1.26x, clearly cited as a surge in demand and shipment volume, rather than one-off factors.

Silicon Photonics has long moved beyond the concept and pilot phase—annualized operation rate of $680 million, $1.3 billion in contract revenue in 2027, and $290 million in advance payments, making it a commercialized product with real backlogs. In its Q2 financial report, GlobalFoundries confirmed that AI optical network demand is expected to grow +62% year-on-year, with silicon photonics expected to more than double in 2026, forming an independent upstream cross-validation.

But the price structure tells a different story. The stock price remains above the 200-day moving average by 25.1%, but has fallen below the 50-day moving average by about 15%, with a 30-day retracement of 23.3%, and the RSI (14) has fallen to the neutral weak zone at 44. Fundamentals are breaking out, prices are returning—the intersection of these two lines is often the real buying point, not the present.

Three-pronged attack: scale, capacity, and capital

Tower leads in the depth of specialty manufacturing processes, but its competitors are all bigger than the last.

GlobalFoundries is the most direct competitor, also betting on differentiated nodes rather than advanced logic, covering RF SOI, germanium silicon, FD-SOI, and image sensors. Its revenue base and customer diversification far exceed Tower's, but Siliconocnic's revenue volume has yet to be clearly disclosed—Tower's $680 million annualized operating rate is currently the strongest figure in this sector.

STMicroelectronics has a deep foundation in silicon germanium and advanced packaging, with its Agrate plant in Italy expanding capacity by 300mm, directly matching Tower's expansion in Japan. But a larger proportion of STM's revenue comes from its own brand products, and OEM is only part of the business. Customer priority and resource allocation may not be as good as pure contract manufacturers.

Samsung is the most resource-rich, with annual capital expenditures dozens of times greater than Tower's $3 billion expansion plan, and possesses full-chain integration capabilities from wafer manufacturing to advanced packaging. However, Samsung's special foundry culture is not yet fully developed, and internal priorities for storage and logic businesses often override foundry services.

2028 is the real test

@TheValueist made a less popular judgment in its foundry competition analysis: the silicon photonics capacity race is already unfolding simultaneously among GFS, STM, and Samsung, and from 2028 to 2029, both average selling price and capital return risk will be downside.

Tower's defense is multi-quarter contracts plus advance payments—locking in both price and volume during capacity shortages. But contracts always expire. When all three competitors' new capacity launches simultaneously in 2028, how much of Tower's pricing power of "customers are begging to buy capacity" will remain?

The ceiling is not about demand, but about how much capacity it can build

Tower's growth boundaries are less driven by market demand and more by its own production lines—because in the foreseeable future, demand will always be on the side where supply exceeds supply.

The company's own scale is very clear

The annualized operation rate of silicon photonics is expected to exceed $1 billion by Q4 2026 from the current $680 million; $1.3 billion signed by 2027; The 2028 business model after management is raised: $3.6 billion in revenue, $1.2 billion in net profit, gross margin of 45%, operating margin of 38%, and net margin of 33%—up 27% from the previous model in February 2026, with margins up by 600 to 700 basis points.

The pace of capacity implementation is equally clear

Japan's dual-track expansion will convert the original Fab 6 (Arai plant) into 300mm silicon photonics and advanced packaging, with mass production conditions ready in Q4 2027; A new factory will be built next to Fab 7 on the second track, aiming for the fourth quarter of 2028, by which time Japan's 300mm capacity will quadruple. This plan has a net investment cap of 3 billion USD, with a support framework of about 160 billion yen from Japan's Ministry of Economy, Trade and Industry.

If we really achieve $3.6 billion in revenue and $1.2 billion in net profit in 2028, then the current $23.8 billion market cap corresponds to about a 6.6x price-to-sales ratio and a price-to-earnings ratio of less than 20x—sounds cheap, but only if there isn't a single drop in the past three years.

What are the buyers betting on?

The logic for being bullish on TSEM is straightforward: it's one of the few companies that has truly turned their AI narrative into a profit sheet.

▲ Bull Case

Bears do not deny earnings, only questioning prices and 2028

Bears on TSEM rarely refute any fundamental data; they argue about how much that data is worth.

▼ Bear Case

From the November financial report to the 2028 production line

Over the next two years, TSEM's stock price rhythm will be dominated by several time points, each of which can independently verify or disprove the current valuation.

•	Q4 2026: Whether silicon photonics can achieve annualized operation rate to exceed $1 billion on schedule is a positive outlook
•	Around November 2026: Q3 earnings delivered on a record $520 million revenue guidance, with the median guidance exceeding the market consensus of about $506 million, indicating a positive outlook
•	FY2027: $13 billion in signed silicon photonics revenue began quarterly recognition, with revenue visibility turning into actual accountability, showing a positive direction
•	Q4 2027: Renovation at Japan's Arai plant completed, with 300mm silicon photonics and advanced packaging ready for mass production, with a positive direction
•	Q4 2028: A new factory next to Fab 7 will begin production, with Japan's 300mm capacity quadrupled, but competitors' capacity will also be concentrated and released during the same period, making the direction mutual

After 314%, the first decent pullback

Over the past 12 months, TSEM has been repriced from a mid-sized company classified as a "cyclical specialty foundry" to a core supplier of AI optical interconnect infrastructure, with a cumulative increase of 314% and a market capitalization pushing to $23.8 billion.

There are roughly three turning points

In February 2026, NVIDIA partnered with 1.6T optical modules, officially bringing Tower into the AI narrative; On July 14, Japan's Ministry of Economy, Trade and Industry announced a $3 billion capacity expansion, raising the capacity ceiling by one level; The Q2 earnings report on August 4 ended this narrative with a 270% silicon photonics growth rate and a $3.6 billion 2028 model.

Then there's this scene—the best news has been released, but the stock price has dropped nearly 20% in two days, with a 30-day pullback of 23.3%, falling below the 50-day moving average. Analysts' average target price remains near $318, with a consensus rating of "Strong Buy," and Suskhana even raised the target price from $180 all the way to $330. When the gap between the target price and the stock price stretches to 50%, usually only one side is correct.

⚠️ Risk Notice • Valuations heavily depend on the 2028 revenue target of 3.6 billion; any year that falls short of expectations will trigger severe revaluation • From 2028 to 2029, GFS, STM, and Samsung Silicon Photonics will concentrate their capacity release, putting pressure on average selling prices and capital returns • A $3 billion expansion erodes cash flow, with a free cash flow yield of only about 1.3%, indicating a thin margin of safety • Silicon photonics revenue has a high customer concentration, and once AI capital expenditure shifts, demand elasticity shrinks rapidly • The stock price is 25.1% above the 200-day moving average, up 314% over 12 months, and the risk of mean reversion has yet to be fully released

🟡 Neutral Fundamentals have been realized, and valuations have priced in the next three years ahead of schedule

💬 Discussion The performance is real, but the price is also truly high. Real-time buy and sell positions can be posted on TSEM via the official account.

Data source

•	Source: NextPick real-time snapshot + Tower Semiconductor's Q2 2026 earnings report and 6-K announcement + official company press release + mainstream financial media and publicly available data from market data providers.