Tower Semiconductor of Migdal HaEmek reported its second quarter today - and this is not one record but a record on every line of the report.
What was reported
| Quarter | Year ago | Change | |
|---|---|---|---|
| Revenue | 460.1 | 372.1 | +24% |
| Gross profit | 137.8 | 80.0 | +72% |
| Gross margin | 29.9% | 21.5% | +8.4pt |
| Operating profit | 90.3 | 39.9 | +126% |
| Operating margin | 19.6% | 10.7% | +8.9pt |
| Net profit attributable | 90.8 | 46.6 | +95% |
| EPS (basic) | $0.80 | $0.42 | +90% |
In millions of dollars
The gap between 24% and 72% is the whole story
Revenue rose 24%. Gross profit rose 72%.
When profit grows at three times the rate of revenue, it means one thing: the product mix changed. Tower is selling different things today, at different margins.
The gross margin widened from 21.5% to 29.9% - a jump of 8.4 percentage points in one year.
And what changed in the mix
CEO Russell Ellwanger gives the answer in a single number:
"SiPho presented triple digit year over year revenue increase to $680 million annual run rate in the second quarter of 2026 from $180 million annual run rate in the second quarter of 2025."
3.8x within a year.
And the next target is stated: "We plan to cross the $1 billion of SiPho revenue annual run rate in the fourth quarter of 2026, with continued significant growth throughout 2027."
And why silicon photonics is precisely the AI story
This is the point worth explaining, because it is not self-evident.
An AI data center is not a collection of individual chips - it is one large machine. Thousands of accelerators need to talk to each other at enormous speed, and the bottleneck stopped being compute a while ago; it is moving the data between the chips.
Copper reaches a physical limit. As distance and rate rise, an electrical connection loses signal and generates heat. Silicon photonics replaces the electrical signal with light - carrying it through fibre, at a higher rate and lower energy.
Tower does not make the AI chip. It makes the component that lets the chips talk.
And that is the third point on the same chain we are covering today: Caterpillar sells the excavator and the generator, Cipher builds the structure and leases it, and Tower makes the optical component running inside it. Three companies, three layers, the same reporting day.
The guidance, and the 2028 target
For the third quarter Tower guides to $520 million - another record, with a range of 5% up or down. That is 31% growth year over year and 13% quarter over quarter.
And Ellwanger added a far bigger statement:
"We are raising our 2028 target business model to $3.6 billion of revenue with $1.2 billion net profit, being fully spoken for by our customers."
$3.6 billion of revenue against a current annual run rate of roughly $1.84 billion. That is close to a doubling within two and a half years. And $1.2 billion of net profit on $3.6 billion of revenue is a 33% net margin - very high for the foundry industry.
And here you have to ask: on what basis.
We checked the claim against the balance sheet - and it holds
Customer advances are the proof
A statement about future demand is usually a claim. At Tower it appears on the balance sheet.
| 30 Jun 2026 | 31 Dec 2025 | |
|---|---|---|
| Deferred revenue and customers' advances (current) | 176.0 | 25.6 |
| Long-term customers' advances | 145.4 | 1.9 |
| Total customer advances | 321.4 | 27.5 |
In millions of dollars
From $27.5 million to $321.4 million within six months - 11.7x.
This is money customers have already transferred to Tower to secure future manufacturing capacity for themselves. Not a contract, not an intention, not a backlog - cash paid up front.
And it also explains the cash flow: first-quarter operating cash flow was $510 million, of which $285 million was an increase in customer advances.
The balance sheet, and the investment
Cash and short-term deposits of $1.48 billion, against total debt of just $141.7 million - net cash of roughly $1.34 billion.
And equity: $3.07 billion, out of total assets of $3.77 billion. A company with almost no leverage.
And the line worth noting
Capital investment in the quarter came to $187 million - more than the $177 million of operating cash flow.
Which means slightly negative free cash flow this quarter.
This is not a problem - it is a decision. The company is expanding manufacturing capacity aggressively to meet demand, funding it from the enormous cash it holds and from the advances. Net property and equipment rose to $1.65 billion from $1.46 billion at the end of 2025.
But it is worth knowing that this growth costs cash, rather than only producing it.
Where Tower manufactures
A point worth knowing, because it bears on geographic risk: Tower has one operating facility in Israel (200mm), two in the U.S. (200mm), and two in Japan (200mm and 300mm) through a 51% holding in TPSCo, and it shares a 300mm facility in Agrate, Italy with STMicroelectronics.
That is a footprint spreading the risk across four countries.
My Angle
A personal opinion of Ilan Abramov - not advice, not a recommendation
Of all the Israeli reports this week, this is the one hardest to argue with.
And not because of the records. Record revenue can be achieved with an eroding margin too. What catches me here is that revenue rose 24% while gross profit rose 72% - that is not growth, it is a change of mix. The company is selling a more expensive, more profitable product today than it was a year ago.
And the thing I checked rather than simply accepting is the 2028 target. When a CEO says a $3.6 billion target is "fully spoken for by our customers", that is exactly the kind of statement you cannot take without verification.
And I found the verification on the balance sheet: customer advances that jumped from $27.5 million to $321.4 million in six months. A customer wiring cash up front to secure capacity points at something real, far more strongly than any order backlog does.
And what I do keep in proportion: a 2028 target is a target. And investment already exceeds cash flow - meaning this growth is being built ahead of the revenue arriving. That is right and justified while the demand exists, but it also means any slowdown in demand catches the company after it has already spent the money.
Bottom line: an Israeli company making a component without which an AI data center does not work, at a widening margin, with $1.34 billion of net cash and almost no debt. That is an uncommon combination.
