TSEM

Tower Semiconductor Ltd

$226.45 USD 25.4B market cap 2026-04-15 (Refresh of 2026-03-27 analysis)

Tower Semiconductor Ltd TSEM BUFFETT / MUNGER / KLARMAN SUMMARY

1 SNAPSHOT

Price$226.45

Market CapUSD 25.4B

EVUSD 25.2B

Net DebtUSD ~0B

Shares0.112B

2 BUSINESS

Tower Semiconductor is the world's leading specialty foundry for silicon photonics (SiPho), manufacturing photonic integrated circuits for optical transceivers used in AI datacenter interconnects. Also produces SiGe drivers/TIAs, RFSOI front-end modules for mobile, power management ICs, and CMOS image sensors across fabs in Israel, USA, Japan, and Italy. SiPho/SiGe RF infrastructure is 27% of revenue and growing 75% YoY, driven by 1.6T optical transceiver demand.

Revenue: USD 1.57B Organic Growth: 9.0%

3 MOAT NARROW

~80% foundry market share in silicon photonics for optical transceivers. Switching costs from 12-18 month qualification cycles. 8+ years of volume SiPho manufacturing know-how. Sole supplier to 4 of top 5 optical transceiver module makers. NVIDIA partnership. Customer prepayments and capacity reservations through 2028. Technology roadmap leadership into 3.2T/6.4T and CPO platforms. Moat transitioning narrow-to-wide as SiPho volumes scale, but nascent market ($228M SiPho rev 2025) has not yet attracted serious TSMC/GF competition.

4 MANAGEMENT

CEO: Russell Ellwanger (since 2005, 21 years)

Bold $920M CapEx for 5x SiPho capacity expansion, 70%+ customer-reserved through 2028. No dividends or buybacks -- all capital reinvested for growth. Fortress balance sheet (D/E 0.14, near-zero net debt). Extended Newport Beach fab lease ($105M). Intel Fab 11X in dispute/mediation. CRITICAL CONCERN: 0.004% insider ownership -- negligible skin in game.

5 ECONOMICS

12.4% Op Margin

7.6% ROIC

USD -0.04B FCF

0.0x Debt/EBITDA

6 VALUATION

FCF/ShareUSD -0.36

FCF Yield-0.1%

DCF RangeUSD 135 - 155

Growth phase 2026-2028: 35% annual NI growth to $750M target. Mature phase 2029-2033: 10% growth. Terminal 3%. Discount 10%. Bear $95-115, Bull $180-215. At $226.45, stock trades 45-65% ABOVE probability-weighted DCF fair value. Even bull case implies downside from current price.

7 MUNGER INVERSION -63.9%

Kill Event	Severity	P()	E[Loss]
Valuation compression from 116x P/E	-45%	40%	-18.0%
Competitive entry from TSMC/GlobalFoundries in SiPho	-35%	25%	-8.8%
SiPho overbuild / AI datacenter demand slowdown	-50%	18%	-9.0%
Cyclical semiconductor downturn	-35%	22%	-7.7%
Geopolitical risk -- Israel conflict	-40%	10%	-4.0%
Tariff / trade disruption (NEW)	-20%	15%	-3.0%

Tail Risk: Combined AI CapEx pause + competitive entry + valuation compression could see 65-75% decline to $55-80. Probability ~7-10%. Israeli fab disruption adds non-correlated tail risk.

8 KLARMAN LENS

Downside Case

Hyperscaler CapEx moderates after 2027, SiPho demand stalls at $500-600M (vs $900M+ target), utilization drops to 60%, margins compress to 15% net. Stock re-rates to 20x P/E on $300M earnings = $53/share (-76%).

Why Market Wrong

Market may underestimate SiPho demand durability given permanent EML-to-SiPho shift. 80% share with sole-supplier positions creates quasi-monopoly. Incremental margins of 42-55% could make 2028 model conservative if demand exceeds expectations. Fortress balance sheet enables through-cycle investment.

Why Market Right

At 116x P/E (up from 97x three weeks ago), market prices in 2028 model PLUS material upside. Stock rose 30% in 3 weeks with zero new data. 546% gain in 5 years. Semi cycles are real. Competition will arrive. 80% share is unsustainable over decades. Management owns 0.004% stock. Even bull case DCF ($180-215) implies downside from $226.

Catalysts

Q1 2026 earnings (May 2026) confirming $412M+ revenue trajectory. Tool qualification milestones for 5x capacity expansion. Customer prepayment disclosures. 3.2T technology demos. Intel Fab 11X resolution. Negative: any hyperscaler CapEx deceleration signal, TSMC/GF SiPho announcement, tariff escalation.

9 VERDICT WAIT

B+ T2 Resilient

Strong Buy$95

Buy$122

Sell$215

Tower Semiconductor is an exceptional company with dominant SiPho market share and a compelling 2028 growth model ($2.84B revenue, $750M NI). However at $226.45 and 117x trailing P/E, the stock is 45-65% above probability-weighted DCF fair value ($135-155). The 30% surge in 3 weeks with no new data is pure momentum. Even the bull case DCF ($180-215) implies downside from here. Wait for $115-130 (accumulate) or $90-100 (strong buy). Aschenbrenner's 2% SALP position validates thesis quality but likely entered at $35-85 in 2025.

🧠 ULTRATHINK Deep Philosophical Analysis

TSEM - Ultrathink Analysis (Refresh)

The Core Question

The question has not changed since three weeks ago, but the urgency has intensified: Tower Semiconductor has appreciated 30% in 20 days -- from $174.68 to $226.45 -- without a single new data point. No Q1 2026 earnings. No new customer announcements. No updated financial model. The stock simply went up because it had been going up, and because the narrative of silicon photonics dominance in AI infrastructure is powerful enough to override any concern about price.

This is the exact mechanism Charlie Munger spent decades warning about. The stock is not trading on fundamentals. It is trading on a reflexive loop where price appreciation validates the thesis, which attracts more capital, which drives more price appreciation. George Soros would recognize this immediately. The question is not whether the business is excellent -- it clearly is. The question is whether reflexivity has divorced the price from any defensible estimate of intrinsic value.

At 116x trailing earnings, the market is now requiring Tower to not only execute flawlessly on its 2028 model ($2.84B revenue, $750M net income), but to significantly exceed it, in order to justify the current price. The bull case DCF -- which assumes model outperformance and assigns a generous terminal multiple -- tops out at $180-215. The stock trades above this. The market is pricing in a future that exceeds the most optimistic reasonable projection.

The Cisco Parallel, Revisited

Three weeks ago, I compared Tower to Cisco circa 2000. That comparison has become more apt, not less. Cisco at its peak traded at about 200x trailing earnings. Tower at $226.45 trades at 117x. Both companies controlled critical networking infrastructure during a genuine technology buildout. Both were correct about the demand trend. Cisco's internet traffic growth continued long after the stock peaked. The transceivers that Tower manufactures will absolutely be needed in ever-larger quantities over the coming decade.

But Cisco peaked at a $555B market cap and took 15 years to approach that level again, despite revenues growing from $18B to $49B during that period. The business executed. The stock did not, because the starting valuation already discounted 15 years of growth. Tower, at $25.4B, is a fraction of Cisco's peak cap. But the principle is identical: when the stock price has already captured the growth, the growth no longer generates returns for new buyers.

There is one important difference. Cisco's internet routing market was already large and diversified in 2000. Tower's SiPho market is still tiny ($228M Tower revenue in 2025) and concentrated. This cuts both ways. The tiny base means growth rates can remain high for longer. But it also means the business is fragile -- more dependent on a smaller number of customers, more vulnerable to technology shifts, and more exposed to competitive entry as the TAM grows large enough to matter to TSMC.

The Moral Hazard of Momentum

The most insidious risk at this valuation is behavioral, not fundamental. When a stock rises 546% in five years and 30% in three weeks, the investor is no longer making a decision about business quality. They are making a decision about whether to bet on continued momentum. And the data says momentum works until it does not, and the reversal is always sudden and savage.

Tower's Q4 2025 results were genuinely impressive: $440M revenue, $80M net profit, 18.2% net margin. The incremental margin story is real. But $80M of quarterly profit, annualized to $320M, supports a market cap of perhaps $6-8B at reasonable multiples (20-25x). The stock is at $25.4B. The gap between what the business earns today and what the market prices it at is $17-19 billion dollars of expected future value. That is the accumulated weight of narrative that would need to be sustained.

Management has upgraded their own model from $500M to $750M net income in a single quarter -- a 50% increase in the target. This is either evidence of conservative sandbagging being corrected (bullish) or evidence that management is getting swept up in the same narrative euphoria that drives the stock (bearish). I note that insider ownership remains at 0.004%. When management upgrades guidance by 50% but does not buy a single share of stock, what does that signal?

What I Am Really Waiting For

I am not waiting for Tower to become a worse company. I am waiting for the price to reflect reality instead of fantasy. There are exactly three scenarios that create an entry point:

Scenario 1: Market Correction. A broad equity selloff of 20-30% would likely take Tower down 35-50% given its extreme multiple. A move to $115-150 would bring the stock into the fair value range on the 2028 model. Probability in next 18 months: 25-30%.

Scenario 2: Execution Miss. One quarter of "merely" meeting guidance -- say $412M revenue instead of $440M+ -- would likely trigger a 25-35% selloff as momentum traders exit. The stock at $145-170 after such a selloff would be approaching fair value. Probability: 15-20%.

Scenario 3: Semiconductor Cycle. The semiconductor industry has experienced meaningful downturns every 3-5 years without exception since the 1970s. The current up-cycle is now 2+ years old. A cyclical downturn would hit Tower's non-SiPho business (73% of revenue) hard, and even SiPho could see demand moderation if hyperscalers pause CapEx. A cyclical trough could bring the stock to $80-120. Probability in next 3 years: 30-40%.

In any of these scenarios, Tower's fundamental business quality remains intact. The SiPho moat does not evaporate because the stock drops. The customer relationships do not dissolve. The 2028 model may be delayed but not destroyed. And at $100-130, the risk/reward becomes compelling: 20-25x a delayed but achievable $500-750M in net income, with a fortress balance sheet providing downside protection.

The Soul of This Business, Updated

Tower's soul remains that of a craftsman foundry -- quietly building mastery in analog/mixed-signal technologies while the world focuses on leading-edge digital logic. Russell Ellwanger's 21-year stewardship has created something genuinely rare: a specialty semiconductor company with dominant market share in a critical growth market, a conservative balance sheet, and deep customer trust.

But the soul of a business and the soul of its stock price are different things. The business soul says: this is an excellent company that will likely be worth significantly more in 2028 than it is today, if measured by intrinsic value. The stock price soul says: the market has already awarded that future value, and then some, to current shareholders.

Buffett's most underappreciated insight is that the quality of the business and the quality of the investment are separable. You can own the best business in the world and still lose money if you pay too much. Tower Semiconductor is, quite possibly, the best-positioned specialty foundry in the world. At $226.45, it is also one of the most expensive semiconductor stocks in the market on a trailing earnings basis.

The patient path is clear. Wait. Let the price come to you. The business is not going anywhere. The SiPho moat deepens with every quarter of production. But the stock, at some point, will correct -- either through a broad market drawdown, a single disappointing quarter, or simple exhaustion of the momentum trade. When that correction comes, Tower Semiconductor will be worth buying aggressively. Today, it is worth admiring from a distance.

The 30% move in three weeks, with zero new information, is the market screaming that this is not a rational price. Listen to it.

Executive Summary

3-Sentence Investment Thesis

Tower Semiconductor is the world's dominant specialty foundry for silicon photonics (SiPho), commanding approximately 80% market share in a segment experiencing explosive growth driven by AI datacenter optical interconnect demand. The company's 2028 financial model targets $2.84B revenue and $750M net profit (26.4% net margin), representing a 50.5% net income CAGR from 2025, supported by $920M in capacity CapEx with 70%+ of SiPho capacity pre-reserved through customer prepayments. At $226.45, the stock has appreciated 30% in three weeks and now trades at approximately 116x trailing earnings and 34x the 2028 net profit target -- pricing in flawless execution with no margin of safety and significant downside risk to any stumble.

Key Metrics Dashboard

Metric	Value	Assessment
Price	$226.45	New all-time high, up 546% in 5 years
Market Cap	~$25.4B
P/E (TTM, on $1.94 diluted EPS)	116.7x	Extreme
P/E (on FY2025 reported $2.19)	103.4x	Extreme
P/E (Forward, FY2028E $6.70 target)	33.8x	Elevated even IF model achieved
P/S (TTM)	16.2x	Extreme premium
EV/EBITDA (est.)	~40x	Very expensive
ROE (2025)	7.6%	Below Buffett 15% threshold
ROE (2028E target)	~26%	Excellent IF achieved
D/E	0.14	Very conservative
FCF (2025)	-$40M	Negative due to heavy CapEx
Revenue CAGR (2025-2028E)	22%	Ambitious but customer-backed
Net Profit CAGR (2025-2028E)	50.5%	Extraordinary target
Dividend	None	Never paid
SiPho Market Share	~80%	Dominant

Decision: WAIT -- Exceptional Company, Even More Extreme Valuation Than Prior Review

Phase 0: Business Understanding

What Tower Semiconductor Does

Tower Semiconductor is an independent specialty semiconductor foundry headquartered in Migdal Haemek, Israel. Unlike leading-edge foundries (TSMC, Samsung, Intel) that pursue the smallest transistor nodes for digital logic, Tower specializes in "more than Moore" analog/mixed-signal technologies that convert the physical world into electronic signals and back.

Core Technology Platforms:

Silicon Photonics (SiPho) -- Photonic integrated circuits (PICs) for optical transceivers used in datacenter interconnects. Tower manufactures the silicon chips that convert electrical signals to light and back. This is the explosive growth driver, and Tower is "by far the majority supplier" of 1.6T silicon PICs.
Silicon Germanium (SiGe) -- High-performance analog chips for transimpedance amplifiers (TIAs) and laser drivers in optical modules, plus low-noise amplifiers for handsets.
RF SOI (RFSOI) -- Radio frequency switches and front-end modules for mobile handsets (65nm platform growing 20%+ YoY).
Power Management (BCD) -- Power conversion ICs, envelope trackers for handsets (new served market), automotive power management.
CMOS Image Sensors -- Machine vision, medical imaging, AR/VR display backplanes (first OLED display backplane PO received).

Revenue Mix (FY2025):

RF Infrastructure (SiGe + SiPho): 27% ($421M) -- up from 17% in 2024, 75% YoY growth
RF Mobile (RFSOI): 23%
Power Management: 16%
Sensors & Displays: 16%
Discrete: 11%
Mixed Signal CMOS: 7%

Manufacturing Footprint:

Fab 2: Migdal Haemek, Israel (200mm) -- qualifying SiGe/SiPho, ~60% utilization
Fab 3: Newport Beach, CA (200mm) -- SiPho/SiGe production, 85% "model full"
Fab 5: Tonami, Japan (200mm) -- Power management, 75% utilization
Fab 7: Uozu, Japan (300mm) -- RFSOI, SiGe, >85% fully utilized
Fab 9: San Antonio, TX (200mm) -- SiPho/SiGe ramp, 65% utilization
Agrate, Italy (300mm) -- JV with STMicroelectronics, RFSOI
Albuquerque, NM (300mm) -- Intel Fab 11X partnership, in dispute

The Silicon Photonics Inflection

The AI datacenter buildout requires massive bandwidth between GPUs, switches, and storage. Optical transceivers convert electrical signals to light for fiber optic transmission. The industry transition is accelerating:

400G transceivers (established, legacy)
800G transceivers (high volume production)
1.6T transceivers (fastest growing -- Tower is majority supplier)
3.2T transceivers (next generation, Tower developing 400G/lane modulators via "3 different pathways")

Silicon photonics displaces legacy EML solutions because: (a) half the external lasers needed, (b) no separate indium phosphide modulator, (c) superior performance at 1.6T+ speeds due to integrated silicon modulator, (d) manufactured on standard silicon wafers. CEO Ellwanger has called this shift "permanent."

The NVIDIA Connection and Aschenbrenner Thesis

Tower's expanded partnership with NVIDIA for optical transceiver supply (through module makers like Innolight, Broadcom) positions it as a critical but hidden supplier in the AI infrastructure stack. Situational Awareness LP (Leopold Aschenbrenner's AGI infrastructure fund) held a 2.0% position (~$85M at Q4 2025 filing). Aschenbrenner likely entered at $35-85 during 2025 -- at those prices, the risk/reward was excellent.

Phase 1: Risk Analysis (Munger Inversion)

"Tell me where I'm going to die, so I'll never go there."

#	Risk Event	Severity	Likelihood (5yr)	Expected Loss
1	Valuation compression from extreme 116x P/E	-45%	40%	-18.0%
2	Competitive entry (TSMC, GlobalFoundries, UMC)	-35%	25%	-8.8%
3	SiPho capacity overbuild / AI CapEx pause	-50%	18%	-9.0%
4	Cyclical semiconductor downturn	-35%	22%	-7.7%
5	Technology disruption (CPO replaces pluggables earlier)	-30%	12%	-3.6%
6	Customer concentration risk	-25%	15%	-3.8%
7	Geopolitical risk (Israel conflict escalation)	-40%	10%	-4.0%
8	CapEx execution failure ($920M plan)	-30%	10%	-3.0%
9	Intel Fab 11X dispute unfavorable	-10%	30%	-3.0%
10	Tariff / trade disruption	-20%	15%	-3.0%
	Total Expected Downside			-63.9%

Risk Deep-Dives

1. Valuation Compression (Severity: -45%, Likelihood: 40%) -- THE DOMINANT RISK

Since March 27, the stock surged 30% ($174.68 to $226.45) with NO new fundamental data (Q1 2026 not yet reported). Pure momentum/narrative. At 116x trailing earnings, more expensive than most AI darlings at peaks. Even achieving the 2028 model ($750M NI, ~$6.70 EPS) puts the stock at 33.8x 2028E -- a premium for a foundry decelerating from peak growth. Eight consecutive quarters of beats create expectations where merely meeting guidance triggers selloffs.

2. Competitive Entry (Severity: -35%, Likelihood: 25%)

80% SiPho foundry share invites competition as TAM grows. GlobalFoundries has SiPho PDK. TSMC N7 has optical capabilities. If TAM grows to $3B+, it becomes irrational for larger foundries not to invest. Switching costs (12-18 months) provide 2-3 year buffer. CEO: "It's very difficult for somebody to break into our position right now."

3. SiPho Demand Disappointment / AI CapEx Pause (Severity: -50%, Likelihood: 18%)

2028 model requires SiPho growing from $228M to $900M+. Needs uninterrupted AI datacenter buildout. Hyperscaler CapEx has historically experienced 1-2 year digestion periods. One year of moderation delays model 18-24 months while depreciation on $920M installed equipment continues.

4. Cyclical Downturn (Severity: -35%, Likelihood: 22%)

Revenue cyclicality: $1.68B 2022 peak to $1.42B 2023 trough. Non-SiPho business (73% of revenue) exposed to handset/industrial/auto cycles.

5. CPO Technology Risk (Severity: -30%, Likelihood: 12%)

Co-packaged optics may favor advanced packaging foundries (TSMC, Intel) over Tower's 200mm specialty process.

10. Tariff / Trade Risk (NEW) (Severity: -20%, Likelihood: 15%)

Escalating semiconductor tariffs risk for global manufacturer. U.S. fabs provide some insulation but Israeli HQ and Japanese operations face trade barrier risk.

Tail Risk: AI CapEx pause + competitive entry + valuation compression could see 65-75% decline to $55-80. Probability ~7-10%.

Phase 2: Financial Analysis

Revenue & Profitability (5-Year History)

Year	Revenue ($B)	Gross Margin	Op Margin	Net Margin	Net Income ($M)	EPS
2021	1.51	21.8%	11.0%	9.9%	$150M	$1.35
2022	1.68	27.8%	18.6%	15.8%	$265M	$2.39
2023	1.42	24.8%	38.5%*	36.4%*	$518M*	$4.66
2024	1.44	23.6%	13.3%	14.5%	$208M	$1.85
2025	1.57	23.2%	12.4%	14.1%	$220M	$1.94

*2023 includes $290M Intel merger termination payment.

Quarterly Trajectory (2025) -- The SiPho Inflection

Quarter	Revenue	Net Profit	Net Margin	EPS	Beat %
Q1 2025	$358M	$40M	11.2%	$0.45	+18.4%
Q2 2025	$372M	$47M	12.6%	$0.41	+10.8%
Q3 2025	$396M	$54M	13.6%	$0.55	+1.9%
Q4 2025	$440M	$80M	18.2%	$0.78	+13.7%

Q4 validates mix enrichment: $40M incremental NI on $82M incremental revenue from Q1 = 48.8% incremental net margin.

2028 Financial Model (Management Target)

Metric	FY2025 Actual	2028 Target	Change
Revenue	$1.57B	$2.84B	+81% (22% CAGR)
Gross Margin	23.2%	39.4%	+16.2pp
Operating Margin	12.4%	31.7%	+19.3pp
Net Profit	$220M	$750M	3.4x (50.5% CAGR)

Model upgraded from $500M NI (Q3) to $750M NI (Q4) in one quarter. Encouraging but may reflect narrative momentum in management guidance.

Balance Sheet -- Fortress

Net debt ~$0. D/E 0.14. Current ratio 6.5x. $2.9B equity. Supports $920M CapEx without external financing.

Cash Flow

Year	Operating CF	CapEx	FCF
2021	$420M	$310M	$110M
2022	$530M	$370M	$160M
2023	$680M*	$440M	$230M*
2024	$450M	$440M	$10M
2025	$400M	$440M	-$40M

FCF negative through at least 2026-2027 during $920M CapEx program.

Valuation Comparison (Updated)

Metric	Mar 27 ($174.68)	Apr 17 ($226.45)	Change
P/E (TTM)	90x	116.7x	+27 turns
P/E on 2028E ($6.70)	26.1x	33.8x	+8 turns
P/S (TTM)	13.6x	16.2x	+2.6 turns
P/B	6.7x	8.7x	+2.0 turns

DCF Fair Value Range:

Scenario	Fair Value	vs. $226.45
Bear Case	$95-115	-49% to -58%
Base Case	$135-160	-29% to -40%
Bull Case	$180-215	-5% to -21%
Probability-weighted	$135-155	-32% to -40%

Stock trades 45-65% ABOVE probability-weighted DCF fair value. Even bull case implies downside.

Graham Number: sqrt(22.5 x $1.94 x $25.90) = $33.60. Stock at 6.7x Graham Number.

Phase 3: Moat Analysis

Moat Sources

1. Switching Costs (HIGH): 12-18 month qualification cycle. Sole supplier to 4 of 5 top optical module makers. Deep PDK/device library integration.

2. Technology Leadership (HIGH): 8+ years volume SiPho manufacturing. Developing 3.2T/6.4T/CPO next-gen platforms.

3. Customer Relationships (MODERATE-HIGH): NVIDIA partnership. Customer prepayments and reservations through 2028.

4. Scale (MODERATE): Global multi-fab footprint. Not absolute scale vs larger foundries.

Moat Rating: NARROW (transitioning toward WIDE)

80% SiPho market share with sole-supplier positions is compelling but market is still nascent ($228M Tower SiPho revenue 2025). As TAM grows toward $3B+, competitive pressure will increase.

Moat Duration: 5-8 years. Secure through 2030. Beyond that depends on CPO transition and competitor investment.

Phase 4: Decision Synthesis

Management

CEO Russell Ellwanger (21 years): Prescient SiPho bet, strong customer culture, disciplined. Critical concern: Insider ownership 0.004% -- essentially zero skin in the game.

Entry Price Targets

Level	Price	P/E (2028E $6.70)	Gap from $226
Strong Buy	$90-100	13.4-14.9x	-56% to -60%
Accumulate	$115-130	17.2-19.4x	-43% to -49%
Fair Value	$135-160	20.1-23.9x	-29% to -40%
Current	$226.45	33.8x	--

What Would Upgrade to BUY

Price drops to $115-130 without fundamental deterioration
Two consecutive Qs of $475M+ revenue with 27%+ gross margins
Customer prepayment commitments totaling $500M+
TSMC explicitly declining SiPho foundry entry

What Would REJECT

SiPho quarterly revenue declines 2 consecutive Qs
Major customer qualifies second SiPho foundry source
CEO departure without credible succession

Final Verdict

Quality Grade: B+ (potential A- if 2028 model achieved) Recommendation: WAIT Tier: T2 Resilient

Exceptional company with dominant SiPho position. At $226.45 and 117x trailing P/E, the stock is 45-65% above fair value with no margin of safety. The 30% move in 3 weeks with no new data is pure momentum. Wait for $115-130 (accumulate) or $90-100 (strong buy).

Sources: Tower Semiconductor 20-F filings (SEC EDGAR), Q1-Q4 2025 earnings transcripts (AlphaVantage), financial statements, EODHD prices. No analyst reports consulted.