Key takeaways
- AI spending is still accelerating: The buildout of AI infrastructure is driving massive capital expenditure across semiconductors, networking, and data center cooling. This isn’t just a 2024 story. The companies supplying the picks and shovels for this wave are posting revenue growth rates that most sectors can only dream about.
- Diverse moats across the stack: What makes this group compelling isn’t just riding one trend. These companies span chip design, semiconductor manufacturing equipment, high-speed networking, and power management, meaning they’re not all competing for the same slice of the pie. That kind of diversification within a theme gives you multiple ways to win.
- Valuations demand real execution: Many of these names are priced for significant forward growth, which means any stumble on earnings or guidance can trigger sharp pullbacks. Geopolitical risk around chip export controls, customer concentration in a handful of hyperscalers, and the cyclical nature of semiconductor demand are all things you need to watch closely before buying at these levels.
I’ve been investing in US tech for years, and the one thing that consistently separates the winners from the losers isn’t hype. It’s infrastructure. The companies that build the picks and shovels for every major technology wave tend to compound far longer than the ones grabbing headlines. Right now, that principle is being tested at scale with AI spending, and the results are staggering.
Think about what’s actually happening. Every major cloud provider and hyperscaler is pouring tens of billions into data center buildouts. That capital doesn’t just go to one chipmaker. It flows through an entire ecosystem: the companies designing processors, the ones manufacturing the equipment that etches circuits onto silicon, the firms building the networking infrastructure to connect it all, and the power and cooling systems keeping everything running. That supply chain is where I think the best long-term AI stock opportunities live.
What makes this cycle different from past tech booms is the revenue backing it up. These aren’t speculative companies pitching a vision. Most of the names I’m covering are posting record earnings, expanding margins, and generating serious free cash flow. The demand signal from their customers isn’t slowing down either. If anything, it’s accelerating as AI workloads get more complex and power-hungry.
Valuations aren’t cheap across the board, though. Some of these stocks have had massive runs, and a few are priced for near-perfect execution. That’s a risk. I’m not going to pretend every name here is a screaming bargain, because they’re not. The goal was to find companies where the growth is real, the competitive position is durable, and the price still gives you a reasonable shot at outsized returns. If you’re already getting broad exposure through NASDAQ ETFs, individual picks let you overweight the specific names you have the most conviction in.
I focused on businesses across semiconductors, networking, and data center infrastructure, because that’s where the earnings momentum is most concentrated. Some are mega caps you already know. Others are mid-cap specialists flying under the radar with dominant positions in narrow but critical markets.
In This Article
- Micron Technology Inc. (MU)
- Arista Networks Inc. (ANET)
- Rambus Inc. (RMBS)
- Lam Research Corporation (LRCX)
- Vertiv Holdings LLC (VRT)
- NVIDIA Corporation (NVDA)
- ASML Holding N.V. (ASML)
- Cirrus Logic Inc. (CRUS)
- Advanced Micro Devices Inc. (AMD)
- KLA Corporation (KLAC)
Micron Technology Inc. (NASDAQ: MU)
Micron Technology Inc. is a leading semiconductor company specializing in the design and manufacture of memory and storage solutions...
Competitive Edge
- HBM (High Bandwidth Memory) is the critical bottleneck for AI accelerators from NVIDIA and AMD. Micron's HBM3E qualification with major GPU vendors creates a supply-locked position where customers cannot easily switch mid-design cycle, providing 12-18 month revenue visibility.
- Memory is an oligopoly: Samsung, SK Hynix, and Micron control over 95% of DRAM. Post-2023 industry discipline on capex has reduced the chronic oversupply cycles that historically crushed margins. Fewer players means more rational pricing behavior.
- CHIPS Act subsidies for Micron's Idaho and New York fabs reduce effective capital intensity by billions. This government co-investment lowers Micron's hurdle rate on capacity expansion while competitors building in Asia receive no equivalent U.S. subsidy.
- The segment reorganization into CMBU/CDBU/MCBU/AEBU in FY2025 signals management is structuring the business around AI-centric end markets rather than product type. This allows more targeted pricing and margin optimization per customer vertical.
- Automotive memory content per vehicle is growing 3-5x as ADAS, infotainment, and EV systems proliferate. AEBU's $4.8B revenue base provides a secular growth floor even if consumer electronics cycles weaken.
By the Numbers
- CMBU revenue exploded 257% YoY to $13.5B with operating income surging from $244M to $6.1B, a 45% operating margin. This single segment now represents 36% of revenue versus 15% a year ago, completely reshaping Micron's margin profile upward.
- Forward P/E of 13.7x against a PEG of 0.06 implies the market is pricing in massive earnings growth but still not fully valuing it. Consensus EPS jumps from $7.59 trailing to $33.74 in Y1, a 345% increase, yet the stock trades at only 13.7x that figure.
- Interest coverage at 51x with net debt/EBITDA of just 0.09x means the balance sheet is effectively unleveraged despite $12.4B in gross debt. OCF/debt ratio of 1.83x means Micron generates nearly twice its total debt in operating cash flow annually.
- DRAM revenue grew 62% YoY to $28.6B while NAND grew only 18% to $8.5B. DRAM now represents 76% of product revenue, up from 71% last year. This mix shift matters because DRAM carries structurally higher margins than NAND.
- U.S. revenue surged 83% YoY to $24.1B, now 64% of total revenue versus 53% last year. This geographic concentration toward the U.S. reduces China regulatory risk exposure while capturing hyperscaler AI infrastructure spending domestically.
Risk Factors
- FCF-to-net-income conversion is just 0.39x, meaning only 39 cents of every dollar of reported earnings converts to free cash. Capex consumes 80% of operating cash flow, and capex/depreciation at 2.1x signals spending far exceeds maintenance levels, compressing real cash returns.
- P/FCF of 50x versus P/E of 21x is a massive divergence. With capex at 43% of revenue ($16B annualized), the market is pricing earnings power that requires sustained heavy capital investment. Any capex cycle downturn would compress returns quickly.
- MCBU revenue growth decelerated sharply from 58% YoY to just 1.6%, while AEBU slowed to 2.6%. These two segments represent 44% of revenue and are effectively stagnant, meaning all incremental growth depends on data center segments sustaining triple-digit rates.
- SBC at 2.5% of revenue looks modest, but against FCF margin of only 11%, stock-based compensation consumes 22% of free cash flow. Combined with buyback yield of just 0.23%, dilution is not being meaningfully offset.
- Mainland China revenue declined 13% YoY to $2.6B while Europe fell 24% to $625M. These declines in non-AI-driven markets suggest underlying demand weakness in traditional memory end markets that the AI boom is masking.
Arista Networks Inc. (NYSE: ANET)
Arista Networks Inc., founded in 2004, is a provider of cloud networking solutions primarily serving large data centers and high-performance computing environments in the telecommunications and computer networking sectors. The company specializes in scalable, high-speed networking infrastructure designed for modern cloud environments...
Competitive Edge
- Arista's EOS (Extensible Operating System) runs a single software image across all platforms, creating deep switching costs. Once a cloud titan standardizes on EOS, ripping it out means retraining network engineers and rewriting automation scripts.
- Customer concentration in Meta, Microsoft, and other hyperscalers is also a moat. These buyers need proven 400G/800G spine-leaf architectures at scale, and Arista's track record in mega-datacenter deployments is unmatched by Cisco or Juniper.
- The campus networking push (targeting Cisco's $15B+ enterprise LAN market) opens a second growth vector. Arista's approach of extending EOS into campus switches leverages existing R&D rather than requiring a separate product stack.
- Jayshree Ullal is among the most respected CEOs in networking, with a 20-year track record of disciplined execution. Arista has never made a dilutive acquisition or chased low-margin volume, a rare discipline in hardware.
- AI/ML training clusters require massive east-west traffic within datacenters, exactly where Arista's high-radix spine switches dominate. Every new GPU cluster from NVIDIA or AMD needs a back-end network, and Arista is the default choice.
By the Numbers
- ROIC of 51% on zero debt means returns are entirely from operations, not leverage. FCF margin of 46.3% exceeds net margin of 39%, confirming earnings quality is pristine with FCF-to-net-income conversion at 1.19x.
- Service revenue grew 34.7% in FY2024 then 27.7% in FY2025, now 15.9% of total revenue vs ~19% historically. Service gross margin is ~82% vs product gross margin of ~60.7%, so this mix shift structurally lifts blended margins over time.
- EMEA surged 50.1% YoY and APAC 45% YoY in FY2025 after near-stagnation in FY2024 (6.3% and 4.3%). International now represents 20.9% of revenue, up from ~18.2% in FY2024, reducing Americas concentration risk.
- SBC at 4.9% of revenue is modest for networking/tech. With buyback yield of 0.93%, repurchases likely offset most dilution, meaning SBC is not materially eroding per-share economics.
- Operating expenses (R&D + SG&A) total just 21.2% of revenue, an unusually lean cost structure. This leaves 42.8% operating margin, and with capex at only 2.2% of revenue, nearly all operating profit converts to free cash flow.
Risk Factors
- Cash conversion cycle of 233 days is extreme, driven by 230 days of inventory. Inventory turnover of 1.59x suggests Arista is stockpiling components, likely to secure supply, but this ties up ~$3.8B in working capital and creates obsolescence risk if demand shifts.
- DCF base case target of $70.71 implies 48% downside from $136.07. Even the aggressive target of $100.70 sits 26% below current price. The market is pricing in sustained 25%+ growth for years, leaving zero margin of safety.
- Product gross margin was 60.7% in FY2025 vs 59% in FY2023, a modest expansion. But quarterly product gross profit grew just -0.3% QoQ in Q2 before recovering 6.5% in Q3, suggesting margin pressure from 800G transition costs or competitive pricing.
- APAC revenue dropped 33.3% QoQ in the most recent quarter to $153M despite the strong annual figure. This volatility suggests lumpy enterprise orders rather than a stable installed base, making APAC a swing factor for quarterly beats/misses.
- At 49.5x trailing earnings and 41x P/FCF, the stock prices in consensus EPS growth from $2.75 to nearly $7.00 by Y4. Any deceleration in hyperscaler capex cycles would compress the multiple and earnings simultaneously.
Rambus Inc. (NASDAQ: RMBS)
Rambus Inc., founded in 1990, is an American technology company that develops and licenses high-speed memory interface and semiconductor intellectual property. Operating in the technology sector, the company focuses on innovations that drive the performance of electronic devices worldwide...
Competitive Edge
- Rambus owns foundational patents on memory interface signaling (DDR, HBM). Samsung, SK Hynix, and Micron must license these technologies. This creates a toll-booth model with near-100% gross margins on royalties and extreme switching costs for licensees.
- The DDR5 and HBM3E upgrade cycles are structural, not cyclical. AI server buildouts by hyperscalers require higher-bandwidth memory, and Rambus's PHY IP and silicon interface chips sit directly in that critical data path between processors and memory.
- After years of patent litigation, Rambus now has long-term licensing agreements with all major DRAM manufacturers. This removes the binary legal risk that plagued the stock for a decade and creates predictable, recurring revenue streams.
- South Korea revenue surged from $7M in FY2022 to $198M in FY2024, reflecting deepening relationships with Samsung and SK Hynix. This geographic concentration actually signals strategic importance, as these two companies control ~70% of global DRAM.
By the Numbers
- FCF margin of 47.1% massively exceeds net margin of 32.6%, with FCF-to-net-income conversion at 1.45x. This signals extremely high earnings quality, as the IP licensing model generates cash well above reported profits with minimal capex needs (capex/OCF just 7.5%).
- ROIC of 27.7% on near-zero leverage (debt/equity 0.014) means returns are entirely from operations, not financial engineering. This is rare in semis and confirms genuine pricing power on the IP portfolio.
- Product revenue accelerated sharply to 40.9% YoY growth in FY2025 (up from 9.9% in FY2024), reaching $348M. This is the fastest product growth in the dataset and signals the DDR5/HBM memory interface cycle is hitting an inflection.
- Net cash position of $737M against a $9.9B market cap provides a 7.4% cash cushion. With current ratio at 8.2x and interest coverage at 220x, there is essentially zero financial distress risk.
- Royalties grew 23.5% YoY to $279M in FY2025 on top of 50.7% growth in FY2024. This two-year compounding reflects expanding adoption of Rambus memory standards across the DRAM ecosystem, not one-time catch-up payments.
Risk Factors
- SBC at 7.7% of revenue is meaningful. With trailing net income around $230M, SBC of roughly $54M represents ~23% of earnings. Buyback yield is essentially zero (0.003%), so dilution is uncompensated. Shares outstanding are not shrinking.
- DCF base case target of $45.47 implies roughly 50% downside from the current $91.04 price. Even the aggressive target of $59.71 sits 34% below current levels. The market is pricing in sustained hypergrowth that the cash flow models don't support.
- Cash conversion cycle at 123 days is elevated, driven by 112 days of inventory. For a company that is primarily an IP licensor, carrying this much inventory relative to product revenue suggests either silicon product buildup or demand timing risk.
- Contract and other revenue has declined for four consecutive years, from $88M in FY2022 to $80M in FY2025. This shrinking segment, which includes engineering services, suggests fewer new design wins or reduced customer engagement on custom projects.
- Trailing P/E of 44x and EV/EBITDA of 31x are steep for a company with 16% three-year revenue CAGR. The PEG of 0.79 looks attractive only because forward estimates bake in 40%+ EPS growth, which requires the DDR5 ramp to sustain without pause.
Lam Research Corporation (NASDAQ: LRCX)
Lam Research Corporation is a leading provider of advanced wafer fabrication equipment critical to semiconductor manufacturing. Founded in 1980 in the United States, the company operates within the technology sector, serving a global clientele...
Competitive Edge
- Lam's dominance in plasma etch and deposition is structurally reinforced by 3D NAND layer scaling. Every new generation (200+ layers) requires more etch and dep steps, growing Lam's content per wafer even without fab capacity additions.
- Switching costs in etch/dep are extreme. Qualifying a new tool takes 12-18 months and risks yield loss. Once Lam is designed into a process flow at Samsung, SK hynix, or Micron, displacement by Tokyo Electron or AMAT is nearly impossible mid-node.
- The installed base model creates a durable annuity. With 90,000+ chambers globally, spare parts, upgrades, and relicensing generate $7B+ annually. This revenue stream compounds as the installed base grows and tools age.
- Gate-all-around (GAA) transistor architecture at 2nm and below increases etch step count by 25-30% versus FinFET. As TSMC, Samsung, and Intel ramp GAA in 2025-2027, Lam's content per wafer inflects higher structurally.
- HBM (High Bandwidth Memory) for AI requires advanced TSV etch and deposition, processes where Lam has clear technology leadership. HBM capacity is tripling through 2026, and Lam captures outsized share of this equipment spend.
By the Numbers
- FCF-to-net-income conversion is 1.00x, essentially perfect earnings quality. Combined with 30.2% FCF margin and 30.2% net margin, every dollar of reported profit converts to real cash, rare for a capital equipment company running capex at 2.3x depreciation.
- ROIC of 53.1% on zero goodwill and zero intangibles means returns are generated entirely from operating assets, not acquisition accounting. This is among the highest genuine ROIC figures in all of semis equipment.
- Customer Support revenue grew from $4.9B in FY2021 to $6.9B in FY2025, now 38% of total revenue. This recurring, higher-margin stream grew through the FY2023-24 downcycle, providing earnings ballast that pure-play equipment peers lack.
- SBC at 1.8% of revenue is remarkably lean for a tech company. Buyback yield of 1.4% meaningfully exceeds dilution, so share repurchases are genuinely shrinking the float rather than just offsetting option grants.
- Taiwan revenue doubled YoY (+106%) to $3.4B in FY2025, and foundry mix hit 45% of systems revenue (59% in the latest quarter). This signals Lam is capturing disproportionate share of TSMC's advanced node and backside power delivery buildout.
Risk Factors
- At 46x trailing P/E and 37.9x EV/EBITDA with a PEG of 4.74, the stock prices in near-flawless execution. DCF base case of $102 implies 55% downside. Even the aggressive target of $129 sits 43% below current price, a stark disconnect.
- Cash conversion cycle of 300 days is extreme, driven by 307 days inventory on hand. Inventory turnover at 1.19x means Lam holds nearly a full year of COGS in stock, creating significant obsolescence risk if customer orders shift or get cancelled.
- China revenue was $6.3B in FY2024 (42% of total) and remains $6.2B in FY2025 (34%). This concentration creates binary risk from export controls. Any tightening of BIS rules on etch/dep equipment to Chinese fabs could erase billions in revenue overnight.
- 3-year revenue CAGR of just 2.6% versus 5-year CAGR of 11.5% shows the recent growth surge is a cyclical recovery, not a new structural trajectory. The 26.8% YoY revenue jump flatters a weak FY2024 base when revenue fell 16-17%.
- P/B of 27.7x on tangible book of $7.96/share means $217 of the $225 stock price is intangible value with zero goodwill on the balance sheet. Any earnings miss or multiple compression has almost no book value floor to support the stock.
Vertiv Holdings LLC (NYSE: VRT)
Vertiv Holdings LLC, established in 2016 following its spin‐off from Emerson Electric, is a technology company specializing in mission-critical infrastructure solutions. It serves data centers, telecom networks, and commercial facilities worldwide with an emphasis on reliability and efficiency...
Competitive Edge
- Vertiv sits at the critical chokepoint of AI infrastructure: thermal management and power distribution for data centers. Every hyperscaler GPU deployment requires proportional cooling and power capacity, creating a direct link between AI capex cycles and Vertiv's order book.
- Switching costs are structurally high. Once Vertiv's thermal and power systems are integrated into a data center design, ripping them out mid-lifecycle is prohibitively expensive. The $2B+ services/spares stream (growing 14.5% YoY) monetizes this installed base lock-in.
- Unlike pure-play competitors like Schneider Electric's data center division, Vertiv's entire business is mission-critical infrastructure. This focus allows faster R&D iteration on liquid cooling and high-density power solutions, which are the bottleneck technologies for next-gen AI clusters.
- The segment restructuring (eliminating CIS and IRS, moving to Products vs. Services) simplifies reporting and likely reflects operational consolidation that should improve cross-selling and reduce duplicative overhead as the company scales.
- Customer base includes all major hyperscalers (Microsoft, Google, Amazon, Meta) plus colocation providers and telecom operators. This diversification within the data center ecosystem reduces single-customer concentration risk while still capturing the secular AI buildout.
By the Numbers
- FCF-to-net-income conversion of 1.42x signals high earnings quality. With capex at only 10.4% of operating cash flow and capex-to-depreciation at 0.71x, Vertiv is spending less on maintenance than it depreciates, meaning reported FCF is genuinely free.
- Americas operating profit grew 56.2% YoY on 41.9% revenue growth, implying significant margin expansion in the highest-growth region. Americas operating margin jumped from ~24.4% to ~26.8%, showing pricing power and operating leverage in the AI/data center buildout.
- Total backlog doubled to $15B (108.9% YoY), now representing ~1.5x trailing revenue. This is the strongest forward visibility Vertiv has ever had and virtually guarantees revenue acceleration into FY2026-2027.
- PEG of 0.58 against consensus EPS growth from $3.41 trailing to $6.01 (Y1) and $8.05 (Y2) suggests the market is still underpricing the earnings ramp. Forward P/E of 44.6x compresses to ~33x on Y2 estimates and ~27x on Y3.
- SBC-to-revenue at just 0.45% is remarkably low for a $62B market cap tech-adjacent company. Buyback yield of 0.58% more than offsets any dilution, meaning per-share economics are genuinely improving without SBC headwinds.
Risk Factors
- Tangible book value per share is $0.03 vs. a $264.71 stock price. Intangibles represent 32.2% of total assets, and goodwill another 16.7%. The entire equity story rests on earnings power, with virtually zero asset-based downside protection.
- EMEA operating profit declined 14.1% YoY despite 1.7% revenue growth, meaning EMEA margins contracted sharply from ~24.5% to ~20.7%. Quarterly data shows volatile swings of -19.9% and +32.9% QoQ, suggesting execution issues or unfavorable mix in that region.
- DSO of 97.6 days is elevated for an industrial equipment company. Combined with a cash conversion cycle of 87 days, Vertiv is financing nearly three months of working capital, which will intensify as revenue scales toward $13.6B+.
- Corporate overhead costs surged 39.4% YoY to $224M, growing faster than revenue (27.7%). At this rate, corporate costs are consuming 2.2% of revenue, up from 2.0%, partially offsetting the operating leverage gains in the segments.
- DCF base case target of $92.21 implies the stock at $264.71 is trading at nearly 3x intrinsic value. Even the aggressive DCF target of $150.60 is 43% below the current price. The entire valuation depends on sustained hypergrowth well beyond current consensus.
NVIDIA Corporation (NASDAQ: NVDA)
NVIDIA Corporation, founded in 1993, is a leading technology company focused on visual computing and artificial intelligence. It operates primarily within the semiconductor industry, driving innovations in gaming, professional visualization, data centers, and automotive technologies...
Competitive Edge
- CUDA's 20-year ecosystem creates massive developer switching costs. With 4M+ developers trained on CUDA, competitors like AMD's ROCm face a cold-start problem that money alone cannot solve quickly.
- NVIDIA's full-stack strategy (chips, networking via Mellanox, software via CUDA/NIM/Omniverse) means it captures value at multiple layers. Hyperscalers buying GB200 NVL racks are locked into the NVIDIA ecosystem, not just buying commodity silicon.
- The shift from selling GPUs to selling complete data center systems (DGX, HGX, NVL racks) raises average selling prices and deepens customer integration. This is a structural margin and stickiness improvement, not a one-time benefit.
- Automotive revenue at $2.3B and growing 39% YoY represents an underappreciated option value. NVIDIA DRIVE is designed into platforms at Mercedes, JLR, and BYD, creating a recurring software-defined vehicle revenue stream.
- Jensen Huang's 31-year tenure and 3.5% ownership stake creates rare founder-operator alignment. His willingness to cannibalize existing products (Hopper to Blackwell in 12 months) prevents the innovator's dilemma that typically plagues incumbents.
By the Numbers
- ROIC of 117% against a debt/equity of just 0.07 means returns are driven by genuine operating performance, not financial engineering. This is one of the highest organic ROIC figures in the entire S&P 500.
- FCF-to-net-income conversion of 80.5% is strong for a hardware company, and capex/OCF is only 5.9%, meaning nearly all operating cash flow drops to free cash flow. Capital intensity is remarkably low for a semiconductor business.
- PEG of 0.33 against a forward P/E of 22x implies the market is still not fully pricing in consensus EPS growth from $4.90 trailing to $8.23 in Y1, a 68% jump. At this PEG, the growth premium is cheap relative to the trajectory.
- SBC/revenue at 2.96% is well-contained for a tech company of this scale, and buyback yield of 0.9% roughly offsets dilution. Shareholders are not subsidizing employee compensation to a meaningful degree.
- Data Center segment operating income grew 57% YoY on 68% revenue growth in FY2026, implying incremental margins above 60%. The segment's operating leverage is still expanding as scale benefits compound.
Risk Factors
- Data Center revenue growth decelerated from 216.7% to 142.4% to 68.2% over three years. While 68% is exceptional, the deceleration curve suggests the hyper-growth phase is normalizing. Consensus Y2 revenue estimates of $331B imply a further step-down.
- Days inventory outstanding of 120 days is elevated and cash conversion cycle of 127 days is long for a fabless chip designer. This signals either build-ahead for Blackwell/next-gen or potential demand-supply mismatch risk if AI spending pauses.
- Consensus estimates show a revenue dip from $362B (Y1) to $331B (Y2) before reaccelerating to $469B (Y3), a 8.4% decline. Similarly, EPS drops from $8.23 to $7.78 in Y2. The market may not be pricing in this mid-cycle air pocket.
- Tangible book value per share of $5.43 versus a $180 stock price means the market is pricing in roughly $4.2 trillion of intangible value. Any sustained deceleration in AI capex cycles would compress this premium violently.
- RPO of $2.3B is tiny relative to $216B trailing revenue, just 1.1%. Unlike SaaS companies with large contracted backlogs, NVIDIA has minimal revenue visibility beyond current quarter orders, making it inherently more volatile.
ASML Holding N.V. (NASDAQ: ASML)
ASML Holding N.V. (NASDAQ: ASML) is the world's leading supplier of photolithography equipment for the semiconductor industry...
Competitive Edge
- ASML is the sole supplier of EUV lithography globally. TSMC, Samsung, and Intel have zero alternatives. This isn't market share dominance; it is a literal monopoly on the enabling technology for sub-7nm chips, protected by 15+ years of R&D and thousands of patents.
- High-NA EUV (EXE:5000 series) creates a second upgrade cycle. Customers who just bought NXE:3600 systems will need EXE systems for 2nm and below. This doubles the addressable opportunity per node transition without requiring new customer acquisition.
- Export controls on China actually protect ASML's pricing power. Chinese fabs are forced to buy older DUV tools at premium prices before restrictions tighten further, while leading-edge customers face no substitution threat. Both dynamics support ASPs.
- ASML's installed base of 5,500+ systems creates a services annuity. Each EUV tool requires ~$5M/year in maintenance. As the EUV fleet grows from ~200 to 500+ tools by 2030, service revenue alone could approach €15B annually.
- The CHIPS Act (US), European Chips Act, and Japan/Korea subsidies are all funding new fabs that must buy ASML tools. Government capex subsidies effectively de-risk ASML's order book for the next 5+ years across multiple geographies.
By the Numbers
- ROIC of 28% on a capital-intensive equipment business with only 0.14x debt/equity signals genuine economic moat, not financial engineering. ROE of 49% is roughly 1.8x ROIC, but the gap is driven by high asset turnover (0.66x) and margin, not dangerous leverage.
- FCF margin of 34% nearly matches operating margin of 35%, with FCF/NI conversion at 1.18x. Earnings quality is excellent: the company generates more cash than reported profits, and SBC at just 0.6% of revenue is negligible dilution for a tech company.
- Net bookings surged 48% YoY to €28B in FY2025 after two consecutive years of decline. This is a powerful leading indicator: bookings now exceed trailing revenue by roughly 15%, signaling accelerating shipment volumes into FY2026-2027.
- Service and field option revenue grew 26% YoY to €8.2B, now representing 25% of total revenue vs. 21% in FY2021. This recurring, high-margin installed base revenue provides earnings stability through cyclical system sales troughs.
- EUV system revenue surged 39% YoY to €11.6B while unit shipments rose only 9% (44 to 48 units), implying average selling price per EUV system jumped ~28%. This reflects High-NA EXE:5000 mix-up and pricing power that directly flows to gross margin.
Risk Factors
- Cash conversion cycle of 270 days is extreme, driven by 264 days of inventory. With each EUV system costing €150M+ to build, ASML carries massive work-in-progress inventory. Any demand shock could strand billions in partially built tools.
- FCF 3-year CAGR is negative at -6.1% even as revenue grew 15.5% CAGR over the same period. Capex/depreciation of 1.53x shows capacity investment is outpacing current utilization, compressing near-term free cash flow generation.
- DUV unit volumes collapsed 25% YoY (374 to 279 units) in FY2025, with KrF down 49% and ArF Dry down 43%. While EUV mix-up is the narrative, DUV is still 49% of system revenue. A sustained DUV downturn would pressure total utilization rates.
- Quick ratio of 0.71x is below 1.0, meaning current assets excluding inventory cannot cover current liabilities. For a company with €2.7B in debt and 270-day inventory cycles, this creates refinancing sensitivity if working capital needs spike.
- Trailing P/E of 49x against a PEG of 2.22 means the market is paying over 2x fair value for each unit of growth. Even on FY2026 estimates (€38.5 EPS), the stock trades at 35x, which requires flawless execution on the ramp to €44B+ revenue.
Cirrus Logic Inc. (NASDAQ: CRUS)
Cirrus Logic Inc. is a semiconductor company founded in 1984 that develops high performance analog and mixed-signal integrated circuits for consumer, industrial, and automotive applications...
Competitive Edge
- Cirrus Logic's custom codec and amplifier designs are deeply embedded in Apple's iPhone and AirPods supply chain, creating multi-year design-in cycles with high switching costs. Replacing a custom audio IC mid-product-cycle is nearly impossible for Apple.
- The HPMS segment (camera controllers, haptics drivers, power conversion) expands Cirrus's content-per-device beyond audio. Each new function won inside an iPhone platform adds an incremental revenue stream without requiring a new customer relationship.
- Fabless model with TSMC and GlobalFoundries manufacturing means Cirrus avoids the capital intensity of owning fabs. R&D at 21.8% of revenue is the primary investment, keeping the business asset-light and FCF-generative through cycles.
- Cirrus's analog/mixed-signal design expertise is a genuine human-capital moat. These chips require deep physics knowledge and years of iteration, making it harder for digital-first competitors like Qualcomm to replicate performance at equivalent power efficiency.
- Automotive and industrial diversification efforts, while early, position Cirrus for content growth in EVs (battery management, audio) where analog precision matters and product cycles are 5-7 years, much longer than consumer electronics.
By the Numbers
- FCF margin of 31.1% exceeds net margin of 20.5% by over 10 points, with FCF/NI conversion at 1.52x. This signals exceptionally high earnings quality, as capex-to-OCF is just 2.4%, meaning nearly all operating cash flow converts to free cash flow.
- ROIC of 23.5% on a near-debt-free balance sheet (net cash of $685M) means returns are entirely organic, not leverage-amplified. This is a genuine capital-efficiency story, not a financial engineering one.
- HPMS revenue grew from $265M in FY2021 to $759M in FY2025, a near-tripling that shifted product mix from 19% to 40% of total revenue. This diversification away from pure audio reduces single-product-cycle risk.
- Buyback yield of 4.2% ($347M in TTM repurchases) against SBC of $81M means buybacks are retiring shares at roughly 4x the rate of dilution. Share count is genuinely shrinking, creating real per-share value accretion.
- Interest coverage of 566x with a current ratio of 7.65 and net cash position of $685M gives the company nearly unlimited financial flexibility through any semiconductor downcycle without needing external capital.
Risk Factors
- Three-year revenue CAGR is negative at -0.7% despite 5Y CAGR of 7.8%, meaning FY2023-FY2025 was essentially a flat period. The 7% YoY rebound in FY2025 only brings revenue back near FY2022 peak levels of $1.78B, not above them.
- Cash conversion cycle of 112 days is elevated, driven by 91 days of inventory on hand. With inventory turnover at just 4.0x, there is meaningful working capital tied up, and any demand miss could trigger write-downs in a fast-moving chip market.
- SBC at 4.1% of revenue looks modest, but relative to net income it represents roughly 21% ($81M SBC vs ~$388M net income). GAAP margins are already absorbing this, but it still represents a meaningful claim on shareholder economics.
- FCF conversion trend is flagged at -1, indicating deterioration. Despite strong absolute FCF/NI of 1.52x, the direction is worsening, which could signal working capital headwinds or timing benefits in the current period reversing.
- Rest of World revenue grew 14.8% YoY while China (59% of revenue) grew just 1.1%. The geographic concentration in China remains extreme, and the growth divergence suggests the core market is stagnating while smaller regions carry the incremental gains.
Advanced Micro Devices Inc. (NASDAQ: AMD)
Advanced Micro Devices Inc. (AMD), founded in 1969, is a leading semiconductor company that designs and manufactures high-performance computing and graphics products...
Competitive Edge
- AMD's MI300X/MI325X GPU lineup has broken NVIDIA's monopoly in data center AI training and inference. Hyperscalers like Microsoft, Meta, and Oracle are actively qualifying AMD as a second source, creating structural demand that didn't exist 18 months ago.
- The Xilinx acquisition gives AMD a unique FPGA+CPU+GPU portfolio for edge AI and embedded applications. No other chipmaker can offer this integrated stack, creating cross-sell opportunities in automotive, telecom, and industrial markets.
- TSMC's leading-edge process technology (3nm, 5nm) is equally available to AMD, neutralizing Intel's historical manufacturing advantage. AMD's chiplet architecture also delivers better yields and cost efficiency at advanced nodes.
- The PC AI refresh cycle is a genuine tailwind for AMD's Client segment. Ryzen AI processors with dedicated NPUs are winning OEM design wins at Dell, HP, and Lenovo, positioning AMD for share gains as the corporate PC fleet upgrades.
- AMD's open-source ROCm software ecosystem, while behind CUDA, is improving rapidly and attracting enterprise customers who want to avoid vendor lock-in with NVIDIA. This software moat is the key battleground and AMD is investing heavily.
By the Numbers
- PEG of 0.19 is extraordinarily low given consensus EPS estimates imply a ~156% CAGR from $2.65 trailing to $6.78 Y1 and $10.89 Y2. Even discounting for estimate optimism, the growth-adjusted valuation is compelling relative to mega-cap semis.
- FCF-to-net-income conversion of 1.57x signals high earnings quality. Cash flow exceeds reported profits because capex/depreciation is just 0.34x, meaning AMD's fabless model generates real cash without heavy reinvestment requirements.
- Net cash position of $6.7B with interest coverage at 51x and debt-to-equity of just 4.7%. This balance sheet gives AMD strategic flexibility to fund AI accelerator development and potential acquisitions without capital structure risk.
- Data Center revenue grew 93.6% in FY2024 and 32.2% in FY2025 to $16.6B, now representing 48% of total revenue vs. 29% in FY2023. This mix shift toward the highest-margin segment is structurally improving the earnings profile.
- Client and Gaming operating income surged 140.5% YoY to $2.86B on 50.8% revenue growth, implying significant operating leverage as the PC cycle recovers. The segment's operating margin expanded meaningfully from roughly 12.3% to 19.6%.
Risk Factors
- SBC at 4.7% of revenue ($1.6B+) against a buyback yield of only 0.5% means dilution is outpacing repurchases by roughly 4:1. Shareholders are funding employee comp, not receiving net capital returns. Reported margins overstate true cash economics by ~470bps.
- Data Center operating income grew just 3.5% YoY despite 32.2% revenue growth, a severe margin compression from 27.7% to 21.7%. This suggests AMD is pricing aggressively or spending heavily to compete with NVIDIA, and the profitability story is lagging the revenue story.
- Cash conversion cycle of 157 days is extremely long, driven by 151 days of inventory. Inventory turnover at 2.4x is sluggish for a fabless company and suggests either demand uncertainty or deliberate stockpiling ahead of potential supply constraints.
- Goodwill and intangibles at 54% of total assets ($53B+) from the Xilinx acquisition create meaningful impairment risk. Tangible book is just $12.94/share vs. $199 stock price, a 15x premium that depends entirely on acquired assets delivering.
- Trailing operating margin of 10.7% and ROE of 7.1% are weak for a company trading at 74x earnings. Even ROIC at 6.6% barely exceeds cost of capital, meaning current profitability does not justify the valuation without dramatic margin expansion ahead.
KLA Corporation (NASDAQ: KLAC)
KLA Corporation is a leading provider of process control and yield management solutions for the semiconductor and related industries. Founded in 1975, the company operates in the high-tech sector with a focus on advanced inspection and metrology systems...
Competitive Edge
- KLA holds near-monopoly positions in optical wafer inspection and reticle inspection, with 50-55% market share. Competitors like Applied Materials and Hitachi lack the algorithmic IP built over decades of defect library data, creating a self-reinforcing data moat.
- EUV lithography at 3nm and below exponentially increases the number of inspection and metrology steps required per wafer. Every new node tightens the relationship between KLA's tools and customer yield, making process control spending structurally grow faster than overall WFE.
- The installed base of 50,000+ tools globally creates a services annuity stream with 10%+ annual growth and likely 60%+ gross margins. Customers cannot easily switch inspection platforms mid-production without re-qualifying entire process flows.
- Advanced packaging (chiplets, HBM, CoWoS) is a secular tailwind that creates entirely new inspection layers outside traditional front-end WFE. KLA's acquisitions of SPTS and Orbotech positioned it years ahead of competitors in this adjacency.
- Unlike lithography (ASML) or etch/deposition (AMAT, Lam), process control faces no credible Chinese domestic competitor. The algorithmic complexity and decades of defect classification data create barriers that cannot be replicated through capital spending alone.
By the Numbers
- ROIC of 42.6% with FCF margin of 34.4% and FCF-to-net-income conversion of 96% signals exceptional earnings quality. This is a capital-light cash machine despite $5.9B in total debt, generating $4B+ in unlevered FCF annually.
- Wafer Inspection revenue surged 43.1% YoY to $6.2B in FY2025, accelerating from flat growth in FY2024. This single product line now represents over 50% of total revenue and is clearly the primary growth engine.
- Services revenue grew 15.2% YoY to $2.68B, marking the fifth consecutive year of double-digit growth. This recurring stream now represents 22% of total revenue, up from 24% mix but providing a higher-quality earnings floor.
- Capex-to-OCF of just 8.1% means nearly all operating cash flow converts to free cash flow. Capex roughly equals depreciation (0.98x ratio), indicating the business requires minimal reinvestment to maintain its installed base.
- Semiconductor Process Control gross margin expanded to 64.2% ($7.02B on $10.95B revenue) in FY2025, up from 64.5% prior year. Combined with 25.3% revenue growth, this segment delivered $1.4B in incremental gross profit, showing pricing power at scale.
Risk Factors
- Cash conversion cycle of 267 days is extreme, driven by 236 days of inventory on hand (inventory turnover just 1.55x). Revenue grew 17.5% YoY but if inventory grew faster, there is meaningful obsolescence risk in a technology business with rapid product cycles.
- Remaining Performance Obligations declined 20% YoY to $7.86B in FY2025, the third consecutive year of decline from the $13.1B peak in FY2022. This 40% cumulative backlog drawdown suggests the current revenue surge is partially backlog conversion, not purely new demand.
- At $1,482, the stock trades at 2.3x the aggressive DCF target of $725 and 3.2x the conservative target of $465. Even assuming consensus FY2028 EPS of ~$54, the stock trades at 27.5x earnings three years out, pricing in near-flawless execution.
- PEG ratio of 6.6x against a 3-year EPS CAGR of 12.2% shows the market is paying an enormous premium per unit of growth. The P/E of 43x requires sustained 25%+ earnings growth to justify, but the 3-year revenue CAGR is only 6.7%.
- China revenue declined 3.7% YoY to $4.04B in FY2025 after surging 46.4% in FY2024, still representing 33% of total revenue. Export restriction risk on this single geography creates a binary downside scenario for one-third of the top line.
The AI infrastructure cycle is real, but I think the market is starting to separate the names that can actually deliver from the ones trading on proximity to the theme. That distinction is going to matter a lot more over the next two years than it did over the last two. When everything in the sector rips together, stock selection feels irrelevant. When multiples compress even slightly, the companies with pricing power and sticky customer relationships hold up. The rest give back months of gains in a week.
I’m not worried about AI spending slowing down. I’m worried about investors treating every semiconductor and infrastructure name like they’re interchangeable. They’re not. Margin profiles are different. End-market concentration is different. Cyclical exposure is different. A company selling into automotive and industrial markets is going to have a completely different earnings path than one selling directly into hyperscaler data centers, even if both get lumped into the same “AI play” bucket on social media.
Be specific about why you own each name. That’s the filter that actually protects you.