Tower Semiconductor ("Tower") is a tier two semiconductor foundry focused on niche-y products and technologies. Tower is an Israeli company with its headquarters and ~25% of production capacity located in Migdal HaEmek in northern Israel. Tower agreed to sell itself to Intel in early 2022 for $53 per share or $49 per share ex net cash. Today you can buy Tower stock for $13 ex net cash, i.e. for 1/4 of the price Intel was willing to pay. I believe it's a compelling long at this price and that the stock can 3x over 3 years. The stock was written up on VIC almost a decade ago and rather remarkably, despite having made huge operational progress since then, Tower's EV is only a few hundred million dollars higher today than it was then. 

The opportunity exists for several reasons. Tower is poorly followed by Wall St (from the large US firms, only Jefferies covers the stock today). Tower was radio-silent for ~18m during the pendency of the Intel deal. This silence is more problematic than it might otherwise be given that Tower is in the midst of deploying fairly large amounts of money towards various capacity expansion projects. Tower is an Israeli company with employees and assets in Israel, which obviously adds hair in the current environment. Tower's historical financials don't paint a particularly rosy picture of the business' quality and certainly don't make underwriting normalised earnings power a particularly easy task. 

History 

It's worth a recap of how Tower ended up where it is today. 

Tower got its start in 1993, when its founders acquired National Semiconductor's 150mm wafer fab (aka Fab 1) in Migdal HaEmek. Tower listed in 1994. Tower built Fab 2 (a 200mm facility) alongside Fab 1 in Migdal HaEmek in 2001. 

The co-CEOs of Tower, Yoav Nissan-Cohen and Rafi Levin, resigned in 2002. The company also dismissed the manager of Fab 2 after several delays in its construction. I've not dug too deeply into what was going on at the time but it seems clear from a cursory review that building and ramping Fab 2 was not smooth sailing and ultimately, heads rolled. The company appointed former Trade Ministry chief scientist Carmel Vernia as CEO. The Israeli government was subsidising the construction of Fab 2 to some degree so presumably this appointment was partially intended to keep those stakeholders happy and the grant money flowing. These management decisions were heavily influenced by Idan Ofer and ILCO which, at the time, owned 20% of Tower. Carmel Vernia resigned in 2005. Idan Ofer's frusrtations were clear as he told the press "we gave him every means, but he did not produce the hoped-for results". Tower appointed Russell Ellwanger (an American) as its next CEO. Ellwanger had previously worked at Philips Semiconductors, Novellus and Applied Materials. He took over in May 2005 and has been in the role ever since (he was 49 when he took the job and he's now 67).

Tower acquired Jazz Semiconductor in 2008. The deal was all-stock and resulted in Jazz shareholders owning ~9% of the new company. Jazz's principal asset was a 200mm facility in Newport Beach, California (now Fab 3).

Tower acquired a fab in Nishiwaki City, Hyogo, Japan in from Micron on June 5, 2011. The deal that nearly doubled Tower's production capacity. Tower paid $40m in cash and issued 19.7m shares to Micron (worth ~$20m at the time). The fab had ~60k wspm of capacity.

Tower acquired 51% of TowerJazz Panasonic Semiconductor Co. (“TPSCo”) on April 1, 2014. Tower consolidated its Japanese operations following the deal, closing the Nishiwaki fab. The deal provided Tower with incremental revenues of ~$400m. The three fabs TPSCo owned had ~65k wspm of capacity. Tower issued 870k of shares, valued at  $7.5m, in exchange for the 51% stake. Tower still owns 51% of that JV today. It is fully consolidated in Tower's results.

Tower acquired a 200mm fab in San Antonio, Texas from Maxim in 2016 for 3.3m shares (worth ~$40m) of Tower stock. That left Maxim owning ~3% of the pro forma shares. The deal closed on February 2, 2016.

Tower announced in 2021 that it would partner with STMicro to accelerate the ramp-up of their new Agrate fab. Tower has access to ~1/3 of the fab's cleanroom space in which they can deploy their own tools as and when they see fit. The company has hinted that there is likely to be some form of ongoing volume/revenue-based royalty to STMicro as the builder and owner of the shell itself but they have not commented on the quantum. This capacity is due to come online in 2024.

Tower's Japanese JV has evolved over time. Panasonic announced the sale of its 49% stake in the JV to Nuvoton for $250m in 2019. Nuvoton is a listed subsidiary of Winbond, another specialty foundry that plays in different parts of the market. The JV shut down its Arai fab in 2022, leaving it with two working facilities in Uozu and Tonami relativey close to one another on the west coast of Japan.

Intel announced that it would acquire Tower for $53 per share or $5.4b on February 15, 2022. The deal was eventually abandoned on August 16, 2023 after failing to receive Chinese approval. Intel paid Tower a $353m break fee. Intel and Tower announced a new foundry agreement a few weeks later under which Tower agreed to invest up to $300m on tools and equipment that will go into Intel's 300mm New Mexico fab. Intel will install, qualify and operate the tools to produce advanced analog products for Tower.

Business

Tower is a foundry which means it provides outsourced semiconductor manufacturing services to firms who want to design but not make chips. Tower breaks its business down a few different ways, by fab, by product and by end market. By product, the key three buckets are RF, image sensors and power.

The RF segment made up ~30% of 2022 revenue and encompasses SiGe (Silicon Germanium) products that typically go into mobile infrastructure and RF-SOI (RF Silicon on Insulator) products that typically go into smartphones. These split ~1/3 (infrastructure) and ~2/3 (smartphone).

The SiGe products typically go into advanced optical transceivers which are used in telecom and data centre infrastructure. This is one of the stronger franchises within Tower where they have ~60% market share.

The RF-SOI franchise primarily supplies RFFE products like switches, antenna tuners and LNAs to vendors like Qorvo, Skyworks and Broadcom.

The image sensor made up ~15% of 2022 revenue. The sensors that Tower makes end up in cars, factory automation systems, robots, intelligent traffic systems, medical devices and cinematography equipment. Their two largest customers here are ON Semi and Intel.

The power franchise made up ~35% of 2022 revenue. This is a mixture of ICs and discrete products. The former is involved in more complex power management functionality. For example, a battery management system that measures the current, passes that information to an MCU and then decides which battery cell to drain. The latter is a more simple and commoditised component that serves a single function. The most common discretes are transistors (used to amplify or switch electronic signals) and diodes (allows current to flow in one direction but not the other). Tower's customers here include companies like Infineon, ON Semi and Vishay. In 2022 the ~35% of revenue split ~20% ICs and ~15% discrete.

The last ~10% of 2022 revenue is made up of "mixed signal" CMOS  products that include both analog and digital components. These tend to serve computing and specialty memory applications.

Tower's competitors vary across product lines. In RF, Global Foundries is another key player while TSMC and UMC are also active in this space. In sensors, they compete with TSMC, STMicro and X-Fab as well as in-house capabilities at many of their customers. In power, the competitive set is somewhat wider as they compete with Global Foundries, TSMC, Vanguard, Dongbu (in Korea), X-Fab (in Belgium) and Hua Hong (in China).

This combination of assets has historically generated 20-30% GMs and with R&D + SG&A running at ~10% of revenue, 10-20% EBIT margins. The LTM numbers are 28% and 18%. The LQ numbers are 24% and 14%. The street (i.e. Jefferies) is at 16% EBIT margins for 2023 and 17% for 2024. From 2014 through 2022, RoIC has ranged from 8% to 25% and averaged 16%.

Valuation

Tower closed at $22 per share today. The company had $6 per share of net cash at the end of Q2. They have already received another $3 from Intel (factoring in modest tax leakage) so the pro forma number is $9. That implies you're paying $13 per share for the operating assets or ~$1.45b.

Tower had invested capital (Net PP&E + Net Working Capital) of ~$1.35 at the end of Q2. That implies you're paying ~1.1x invested capital.

Tower should generate ~$240m of EBIT this year or ~$215m of NOPAT, implying you're paying ~7x 2023 NOPAT.

There are two good arguments as to why 2023 earnings are closer to trough than peak.

Tower's capacity utilisation in Q2 2023 was 55-60%. This compares to a practical peak of ~85% so the real number is more like 65-70%. That creates room for revenue growth that should come in at nicely accretive GMs (since D&A is already running through the P&L) and with minimal CapEx required. That GM dynamic will also be helped by the fact that Tower's Newport Beach fab is currently running at the lowest utilization levels (~40%) and produces the highest ASP, highest margin product in Tower's portfolio (SiGe) so as that utilisation recovers, Tower should see a further tailwind to GM from mix-shift.

Tower is in the process of adding capacity in partnership with STMicro (in Italy) and Intel (in the US). They are now $200m into a $500m CapEx program with STMicro. They will spend another $100m in H2 this year and another $200m in 2024. Tower's CEO said on an earnings call - before the Intel deal was announced - that the Agrate capacity was fully spoken for through 2026. That comment was made in late 2021 in a different economic environment so only time will tell if it holds true. The company hasn't provided guidance on the expected revenue contribution at Agrate but I think a range of $200-300m is reasonable. Tower only recently announced the partnership with Intel so that CapEx ramp has yet to kick off. They will spend $300m on tools over the next 12-18m and expect to be fully qualified and selling wafers at the start of 2025. They've not given much guidance but some rough math on layers per wafer and price per wafer suggests that this could yield ~$200-250m of revenue. These are meaningful numbers in the context of a business expected to do $1,425m of revenue this year. Together, Agrate and Albuquerque could add 1/3 to Tower's revenues over the next 2-3 years.

There are a few important dynamics to consider when thinking about Tower's business today and how it's evolved over time. If you rewind the clock to 2015, Tower had created the TPSCo JV in 2014 and, as a result, Panasonic was now a 40% customer. Tower also had ~$100m of net debt, manageable but hardly a fortress balance sheet. Today Panasonic is ~15% of revenue and Tower has ~$1b of net cash. The dwindling relevance of Panasonic matters because (a) they're a drag on growth and (b) they're a drag on margins. In 2015, Panasonic contributed ~$385m of revenue and the rest of Tower's customers contributed ~$575m. In 2022, those numbers were ~$235m and ~$1.4b, respectively. If you strip out revenues from Panasonic and Maxim, Tower's revenue growth from 2015 through 2022 has averaged 14%. Tower discloses net income attributable to minority interests and the only subsidiary in which they don't own 100% is TPSCo so we can easily back into TPSCo's PAT. Tower also discloses TPSCo's debt and cash balances so we can fairly reliably back into TPSCo's EBIT. That suggests that TSPCo is making low/mid single digit EBIT margins. This would suggest that historically margins ex TPSCo have actually been far healthier than consolidated reported margins and that, as the % of revenue coming from Panasonic continues to decline, Tower should see a tailwind to margins.

Upside

I don't have enough information today to get really precise with my model so I'm going to use "CEO math". Tower is on track for $1,425m of revenue this year. That's a 4% CAGR vs 2019 which was a fairly tough year for the broader analog semiconductor industry so nothing too heroic as the baseline. If we grow that at a GDPesque 4% as a result of utilization recovering somewhat and tactical capacity expansion here and there you're looking at $1.7b of revenue in 2027. If you layer on ~$450m for Agrate and Albuquerque, once ramped, you're looking at $2.1b. If we look at EBIT margins ex Panasonic/Maxim, they've averaged ~17.5% over the last eight years. The benefits of levering OpEx against a meaningfully larger revenue base should get that close to 20% by 2027. That implies $400m of EBIT, $360m of NOPAT and $3.3 of EPS (ex interest income). If you put that on 14x ($46) and credit them for $18 per share of net cash at YE26, you're looking at a $65 share price by 12/26. That's a 3x over just over 3 years or a 40% IRR.

Risks

The big ones are pretty obvious. #1 The Israel situation gets completely out of hand. Their two fabs in Israel are ~30% of capacity today going to ~20% over time as Agrate and Albuquerque ramp up. That said the HQ is in Israel and R&D functions too. The 150km to Gaza is comforting but doesn't eliminate the risk entirely. #2 The cycle deteriorates. They've already seen this happen in RF but, so far, demand for their image sensors and power semiconductor demand has remained healthy. The structural underpinnings of both of these end markets feel pretty good to me but that doesn't preclude a cyclical correction at some point. #3 They've got $1b of net cash on the balance sheet and are a self-described "growth company", might they do something stupid with it. It's possible but I rather doubt it. Their CEO has been in the role for almost 20 years and over that time has taken Tower from a business doing negative $170m of EBIT on $100m of revenue with $500m of debt to a business doing $240m of EBIT on $1.4b of revenue with $1b of net cash on the balance sheet. He was savvy enough to sell the business to Intel at close to the top. They had to pinch pennies for a long time to survive. I doubt they forget those lessons overnight.