Best semiconductor stocks to buy after the 2026 rally

On May 14, 2026, NVIDIA closed at an all-time high of $1,287. Two weeks later it was $1,205 — down 6.4% on no news. The chat groups lit up: rally over.

That same morning Google disclosed a $460 billion Cloud backlog. Meta raised 2026 capex to the $125–145 billion range. TSMC said it had taken in over a year of revenue in orders. The forums were looking at the chart. The order books were already in 2028.

Both pictures are real. Both will stay real for the next six quarters at least. What you do with that depends on which kind of investor you are — the one watching the chart or the one reading the backlogs.

This guide is for the second one.

What this guide is and isn't

It is: 30 semiconductor and semi-adjacent names organized into 6 value-chain layers, with one-paragraph verdicts on each, three model portfolios at different risk appetites, and the rupee math an Indian-resident investor needs to size positions and file taxes properly.

It is not: investment advice. It is not a momentum scan. It is not bullish on everything that has "AI" in the pitch.

Numbers are sourced from the latest filings, earnings calls, and analyst notes as of early June 2026. Multiples move daily. Read the latest 10-Q on anything you act on. Caveats throughout. Read the trade-offs in the model portfolios at the end before sizing anything.

The framework

After a rally, you don't bet on momentum. You bet on businesses where the customer cannot postpone the order. Multiple compression hurts most when the buyer can wait a quarter. Insulation against that risk — not earnings growth — is what makes a semiconductor name "durable" at current multiples.

Six layers, ranked by how much insulation each tier has against multiple compression:

Layer	What it is	Insulation level
1. Equipment	Machines fabs cannot operate without	Highest
2. EDA / IP	Design tools every new chip starts in	Very high
3. Foundries / IDMs	The factories	High (with geopolitical tail)
4. AI compute & networking	GPUs, custom ASICs, switching silicon	High (but bifurcated)
5. Memory, analog, specialty	DRAM, NAND, analog, mobile, edge AI	Medium
6. Infrastructure & power	Cooling, REITs, electrical, generation	High (different cycle)

Each layer below: 1-paragraph thesis, 3–7 names with crisp verdicts, then the Top pick of the layer in a single sentence.

The verdict format throughout:

Verdict — [Action]: [The reason in one line]. [The caveat in one line].

Actions used: Core buy (size as full position), Add (build into it), Hold (own but don't add at these prices), Watch (waiting for entry), Hedge (small position as risk offset), Skip (the multiple does not compensate for the risk), Avoid (the equity itself is structurally damaged).

Layer 1 — The toll booth: Equipment

Foundries cannot exist without these machines. When fab capex compresses, equipment companies hurt. When fab capex expands — and TSMC just raised 2026 capex to $52–56B and Samsung joined them — equipment companies print.

ASML

The only EUV supplier on earth. Nikon and Canon publicly walked away years ago.

The numbers: Q1 2026 net sales €8.8B, GM 53%. Year-end 2025 order book €38.8B. 2026 sales guide €36–40B. High-NA EUV (the Twinscan EXE) ships at ~$380M per machine; ASML plans 20/year by 2028, 10–20 already booked. Forward P/E ~26–30×.

Catalyst: TSMC's N2 ramp + High-NA EUV adoption decisions from Samsung and Intel through 2026–27.

Risk: China export controls keep tightening — successive rule rounds have removed marginal revenue. If TSMC pushes N2 by two quarters, ASML's multiple compresses on the news, not the earnings.

Verdict — Core buy: EUV monopoly with a 12-month-plus backlog locked in. If you only own one semi name, this is it. Size for index-comparable weight, not for swing-trading.

Applied Materials (AMAT)

Broadest WFE portfolio — deposition, etch, CMP, ion implant. Sells across foundries, memory, and packaging.

Q1 FY26 was a revenue record at 49.1% non-GAAP gross margin. CEO publicly guides CY26 semi equipment revenue >+20%. Forward P/E ~39×. AI-driven leading-edge logic, HBM, and advanced packaging are the drivers.

Catalyst: TSMC capex flowing through to multi-quarter shipment visibility, plus the HBM cycle pulling memory tool demand.

Risk: Trade tension with China is the most concentrated risk. If memory CapEx slows in H2 2026, AMAT decelerates harder than KLAC.

Verdict — Add: Broadest exposure to the foundry capex cycle at a multiple that has already re-rated. Not the cheapest WFE name; it is the most diversified.

Lam Research (LRCX)

Etch leader, deposition strong, memory-exposed (DRAM and NAND CapEx is a meaningful driver).

Q3 FY26 (March quarter): $5.84B at 49.9% gross margin — record. HBM and advanced packaging momentum. Forward P/E ~40×.

Catalyst: Continued HBM3E/HBM4 capacity additions at Micron, Samsung, SK Hynix.

Risk: Memory cyclicality. LRCX rises and falls with the DRAM/NAND cycle harder than AMAT or KLAC.

Verdict — Add: Highest leverage to the memory CapEx upcycle currently underway. The downside is the memory cycle inversion — buy expecting the up-cycle to continue another 18–24 months, not forever.

KLA Corporation (KLAC)

Process control and metrology — the quality gate of every fab. Near-duopoly with Onto Innovation (KLAC ~75% share, Onto ~7–10%).

Highest margins in WFE (60%+ gross historically). Forward P/E ~38×. "Well positioned for AI ecosystem buildout — foundry/logic, memory, advanced packaging, services" per the latest earnings call.

Catalyst: Advanced packaging (CoWoS) growth at TSMC drives metrology demand disproportionately.

Risk: Single-fab concentration — TSMC alone is ~38% of revenue.

Verdict — Core buy: Cleanest margin profile in WFE, least cyclical of the three big names, structural duopoly. If you can only own one WFE name, KLAC. Lower volatility, higher conviction.

Onto Innovation (ONTO)

KLAC's junior partner in process control. Smaller, more focused on packaging/HBM metrology.

Q1 2026: revenue +20% YoY. Forward P/E ~32×. Sub-$10B market cap means more upside if it takes share, but liquidity is thinner for retail.

Catalyst: HBM4 ramp through 2027 — Onto's tools are specifically used in HBM stacking inspection.

Risk: Customer concentration heavier than KLAC. Less depth in the foundry segment.

Verdict — Add (advanced): A real catalyst on HBM4 ramp, but it's a smaller name with thinner liquidity. Sized 1/3 of a KLAC position, only for portfolios that already own KLAC.

ACM Research (ACMR)

China-exposed cleaning and ECP equipment. Majority of revenue from mainland China.

Trades cheaper than US WFE peers (forward P/E ~10–13×) — but the discount exists for a reason.

Catalyst: China semi build-out continues to need cleaning equipment regardless of US export rules.

Risk: SEC + listing risk. PCAOB audit issues. Successive US restrictions could include ACMR. This is the trade where the discount is real and the risk is real.

Verdict — Skip (most readers): The discount is real but the listing risk is also real. The semi exposure here is achievable through ASML and TSM with much less overhang. Only for portfolios that explicitly want a China cyclical leveraged play.

Layer 1 top pick: ASML for conviction, KLAC for cleaner risk-adjusted entry.

Layer 2 — The design tools: EDA / IP

Every new AI chip — TPU, MTIA, Trainium, Maia, Anthropic's ASIC, every hyperscaler accelerator — starts in Cadence or Synopsys. The market is a near-duopoly. Siemens EDA holds meaningful share but lags on AI-workload tooling.

Cadence (CDNS)

Q1 2026: revenue $1.474B (+19% YoY). Record backlog $8.0B. $4.0B of remaining performance obligations expected over the next 12 months. 45% non-GAAP operating margin. 2026 revenue guide raised to +17% YoY.

Forward P/E ~45×. Recurring revenue, sticky tools, high switching cost.

Catalyst: AI workload chip starts are roughly doubling. Every hyperscaler now designs its own silicon — every project is a CDNS deal.

Risk: China export-control overhang (~7–10% of revenue at risk in late 2025; partly resolved, regulation remains live).

Verdict — Core buy: Every new AI accelerator program is a Cadence deal. Recurring backlog, duopoly structure, structural growth in chip design starts. The under-discussed name in the AI infrastructure stack.

Synopsys (SNPS)

Cadence's twin. Latest backlog $11.0B at April 30, 2026. Trades at a slight discount to CDNS on forward multiples. Greater IP-licensing mix than CDNS — the Arm-comparable royalty stream is meaningful.

Forward P/E ~38×.

Catalyst: Ansys acquisition closed; integration to drive cross-sell into AI / automotive design. Continued AI chip design starts.

Risk: Same China overhang. Larger acquisitions to integrate.

Verdict — Core buy: Same thesis as CDNS at a slightly cheaper multiple plus richer IP licensing mix. If you want the EDA duopoly thesis at the cheaper of the two, SNPS.

Arm Holdings (ARM)

The IP company. FY26 royalty revenue $2.61B (+21% YoY); datacenter royalty more than doubled. Q4 FY26 royalty $671M (+11% YoY). FY26 total revenue $4.92B record.

The multiple: ~38× EV/Revenue, ~100× price-to-royalty-revenue.

Catalyst: Royalty-rate step-up with each generation of Arm v9 architecture. Datacenter share expansion (NVDA Grace, Microsoft Cobalt, AWS Graviton).

Risk: At ~38× revenue, you're paying for years of compounding to hold without slip. RISC-V long-term substitution risk in low-end IoT/MCU.

Verdict — Watch: Beautiful business. Valuation prices in zero slip for years. Wait for a 25–30% pullback before sizing a real position.

Layer 2 top pick: CDNS or SNPS — pick one or barbell them.

Layer 3 — The factory: Foundries and IDMs

The companies that physically make the chips. The Taiwan tail risk concentrates here.

TSMC (TSM)

Global foundry share 70.4% Q4 2025, projected ~75% in 2026. Leading-edge (3nm/2nm) share >90%.

NVIDIA overtook Apple in 2025 as TSMC's #1 customer at ~19% of revenue. Apple is now #2 at 17%. Top 10 customers = ~76% of revenue. The hyperscaler-AI shift in customer mix favors durability — AI capex is contracted years out; consumer phone cycles aren't.

2026 capex raised to high end of $52–56B. FY26 revenue guided to >30% YoY growth (~$158B). Forward P/E ~24–28×. Gross margins recovering toward 55%+.

Catalyst: N2 ramp accelerating into 2026; CoWoS advanced packaging capacity tripling to ~85K wpm by end-2026.

Risk: Taiwan geopolitics. The single largest tail risk in the semi stack, with no analytical hedge. Arizona Fab 21 is producing 4nm at par-with-Taiwan yields but does not de-risk the corporate base case.

Verdict — Add (sized down for tail risk): Second-most defensible structural bet, but carries a tail risk you can't model. Half the position you'd want at no geopolitical overhang. Express the rest of the theme via ASML.

Intel (INTC)

The turnaround narrative is dated. 18A has three confirmed external customers: Amazon, Microsoft, US DoD. NVIDIA and Broadcom tested 18A but have not committed to production. Apple has 18A-P PDK and is running simulations. CEO Lip-Bu Tan publicly embraced external customers in 2026.

Forward P/E ~102–117× (some analysts cite 143×). Net debt high. Operating losses in foundry continue.

Catalyst: 18A external customer wins (especially NVDA / AVGO if they commit), plus CHIPS Act milestones.

Risk: If any testing customer walks, the multiple compresses fast. The turnaround is now fully priced.

Verdict — Skip (or trade only with a tight stop): The story is real and the catalysts are real, but the equity already prices a successful turnaround. Asymmetric downside if execution slips. Long-only investors should wait for a meaningful disappointment to re-enter cheap.

GlobalFoundries (GFS)

Mature-node specialist (12nm and above). RF, embedded, automotive, IoT.

Q1 2026 revenue ~$1.6B (slight YoY decline). Operating margin under pressure. Multi-year supply agreements with auto and aerospace customers provide visibility.

Forward P/E ~22×. Lower than TSM, but the growth and moat are also lower.

Catalyst: Auto silicon content per vehicle continues to rise; GFS is a key supplier.

Risk: Cyclical exposure to auto. Higher capex intensity per dollar of revenue than TSMC.

Verdict — Hold (for diversification): Defensive mature-node exposure, off the bleeding-edge multiple cycle. Useful for a 5–10% sleeve in a balanced semi portfolio, not as a core holding.

UMC (United Microelectronics)

The Taiwanese mature-node alternative. Cheaper than TSM at ~12× forward P/E.

Pure cyclical play — when consumer/auto demand recovers, UMC participates first.

Catalyst: Consumer electronics cycle bottoming. Communications inventory clearing.

Risk: Same Taiwan tail risk as TSM but without the AI-accelerator upside.

Verdict — Skip: You get Taiwan risk without the AI upside. Express mature-node exposure via GFS instead, or via the equipment makers that sell to both.

Layer 3 top pick: TSM sized for tail risk; GFS as defensive diversification.

Layer 4 — AI compute and networking

This is where the rally was loudest. It is also where the bifurcation between "structural moat" and "expensive momentum" matters most.

NVIDIA (NVDA)

Q1 FY27 (quarter ended April 26, 2026): Data Center revenue $75.2B, +92% YoY, total revenue $81.6B (+85% YoY). Data Center is ~92% of total revenue. Within DC: Hyperscale $37.9B + AI Clouds / Industrial / Enterprise $37.4B.

The counterintuitive number: forward P/E ~25× on FY27 estimates, materially below the 5-year trailing average around 70×. The stock is at all-time highs on dollar price; the multiple has compressed because earnings outran price.

Catalyst: Blackwell B200 + GB200 NVL72 systems shipping; Rubin GR200 in 2027.

Risk: ASIC threat (Google TPU, Meta MTIA, AWS Trainium). The current ~25× forward P/E is forgiving of mild deceleration. Any quarter that breaks the 70%+ DC growth pattern compresses the multiple.

Verdict — Core buy: The narrative is "expensive" but the forward multiple has actually compressed. Dollar entry at record highs, multiple at 5-year average. Concentration risk is real — size to your conviction on the ASIC-vs-merchant-silicon debate.

AMD

The second pillar of the AI compute trade. MI400X / MI500 ramp; CDNA roadmap competitive.

Q1 2026: revenue $7.4B (+36% YoY). Data Center segment +73% YoY. Forward P/E ~30× — meaningfully cheaper than NVDA.

Catalyst: Hyperscaler diversification away from sole-NVDA dependence; the OpenAI–AMD MI accelerator commitments through 2027.

Risk: AMD has historically over-promised on AI accelerator timelines. The MI300X execution was strong; MI400X execution is the next test.

Verdict — Add: The structural "alternative to NVDA" trade at a cheaper multiple. Position size 30–50% of your NVDA position — directional bet on hyperscaler vendor diversification.

Broadcom (AVGO)

Custom AI ASICs + networking + VMware compounder. AVGO is no longer "the silent partner" — both the Google TPU partnership and Meta MTIA are now publicly named.

Q1 FY26 AI revenue $8.4B (+106% YoY), Q2 guide $10.7B. AI backlog $73B. Hock Tan public 2027 target: $100B+ AI revenue. April 14, 2026 Meta MTIA deal extends through 2029 with 1 GW initial deployment. Anthropic has access to 3.5 GW via Google–Broadcom TPU capacity. AVGO holds ~70% of custom AI accelerator market.

Forward P/E ~31×. VMware integration tracking ahead of plan.

Catalyst: Meta MTIA volume ramp; new ASIC customer announcement(s).

Risk: Customer concentration (top 3 are heavy share). VMware execution risk.

Verdict — Core buy: The structural hedge to NVDA — wins under both "NVDA dominates" and "hyperscalers shift to ASICs." Cleaner backlog visibility than NVDA. The barbell with NVDA is the call.

Marvell (MRVL)

Pure custom ASIC + optical networking play. Trainium 2/3, Microsoft Maia, Google Axion all involve Marvell silicon.

Q1 FY27: revenue $2.418B (+28% YoY). Data Center segment >75% of revenue and accelerating. Forward P/E ~33×.

Catalyst: Trainium 3 ramp at AWS; new design wins at Microsoft and Google.

Risk: AVGO is the dominant force in the same market — MRVL has to win share or grow with the market. Lower margins than AVGO.

Verdict — Add: The fast-follower in custom AI silicon. Smaller, more leveraged play than AVGO. Position 30–40% of your AVGO position; redundant if you already own AVGO heavily.

Arista Networks (ANET)

Switching silicon for AI data center fabrics. 800G is the production interface for hyperscaler AI clusters; 1.6T is in the design pipeline.

Q1 2026: revenue $2.21B (+27% YoY). Operating margin 47%+. Customer concentration: Microsoft + Meta together >40% of revenue. Forward P/E ~34×.

Catalyst: 800G volume shipments; 1.6T transition in 2027.

Risk: Customer concentration. NVDA's Spectrum-X and Broadcom's switching ASICs both compete for the same socket.

Verdict — Add: The clean networking pick for AI data centers. Customer concentration is the risk — size accordingly. Sub-30% position weight in any AI-heavy semi portfolio.

Layer 4 top picks: NVDA + AVGO as the barbell. ANET for networking. AMD if you want hyperscaler-diversification leverage.

Layer 5 — Memory, analog, specialty

The cyclical tier. The names here are not "durable at peak multiples." They are situationally interesting at specific points in the cycle.

Micron (MU)

HBM is the entire story. Q3 FY26 guide: revenue $33.5B ± $750M at ~81% gross margin — peak-cycle levels. HBM3E sold out through most of 2026. FY26 capex raised to >$25B.

Stock up ~70% YTD 2026.

Catalyst: HBM4 ramp in 2027; AI accelerator HBM demand visibility.

Risk: 81% gross margin is historically the cycle top. Memory cycles compress fast when supply catches up. Capex expansion now seeds the next over-supply.

Verdict — Hold (if already own); Skip (if just buying): The cycle is real but the upside is priced. Buying here is buying at the peak of margin expansion. Existing holders can stay — new positions face asymmetric downside.

Western Digital (WDC)

Post-split, WDC is the HDD-pure entity (SanDisk / SNDK spun out as the NAND business). Data center mass storage is having a real revival driven by AI workload data volumes.

Forward P/E ~9×. Wildly cheaper than the AI semis. Dividend yield ~3%.

Catalyst: 32TB+ HAMR drives ramping; data center HDD pricing power continuing.

Risk: Long-term decline of HDD in any non-cold-storage use case. Cyclicality.

Verdict — Hedge / Add (defensive): The unloved AI play. HDDs are not glamorous but data center HDD spend has been rising. Single-digit forward P/E with dividend yield + AI exposure. Genuinely contrarian.

Seagate (STX)

The other HDD name. Same thesis as WDC but bigger HAMR transition lead.

Forward P/E ~13×. Dividend yield ~2.5%. STX is further along on HAMR rollout than WDC.

Verdict — Hold: Same theme as WDC. If picking one, STX for HAMR leadership; WDC for cheaper multiple.

SanDisk (SNDK)

WDC's NAND spin-out. Lower margin than HDDs, more cyclical exposure to memory cycle.

Forward P/E ~14×. Higher beta to the memory cycle than MU.

Verdict — Skip: No clear advantage versus Micron at the current entry. If you want NAND exposure, do it through Micron.

Texas Instruments (TXN)

Analog incumbent. Industrial + automotive concentration. Multi-year capacity expansion completing through 2027.

Q1 2026: revenue $4.07B (+11% YoY). Operating margin 38%. Forward P/E ~28×. Dividend yield ~2.7%.

Catalyst: Industrial recovery. Auto silicon content rising.

Risk: Multi-year capacity expansion will weigh on FCF until 2027.

Verdict — Hold: Defensive analog. Slow grower, steady payer, low correlation to AI cycle. 3–5% sleeve in a diversified semi portfolio.

Analog Devices (ADI)

The premium analog name. Higher margins than TXN, slower growth, more industrial concentration.

Q2 FY26: revenue $2.64B (+22% YoY). Forward P/E ~28×. Dividend yield ~1.7%.

Catalyst: Industrial inventory normalization is the catalyst that already played out partly; further re-acceleration is the next leg.

Risk: Industrial cyclicality. Auto exposure.

Verdict — Hold: Slightly better quality than TXN. Same general thesis. Pick one for the analog sleeve, not both.

Onsemi (ON)

Auto-power semi specialist. SiC (silicon carbide) market leader for EV powertrain.

Q1 2026: revenue $1.42B (-22% YoY) — auto/EV correction continues. Forward P/E ~21×.

Catalyst: EV demand stabilization. SiC content per EV rising.

Risk: EV cycle continues to weigh. China auto silicon competition real.

Verdict — Watch: The EV correction may not be done. SiC story is real but timing is uncertain. Wait for two consecutive quarters of revenue growth.

Microchip (MCHP)

Embedded MCU + analog. Industrial recovery play.

Q4 FY26: revenue $1.18B (-26% YoY). Forward P/E ~24×.

Verdict — Skip: Continued revenue decline. Better embedded plays via TXN. Wait for cycle inflection.

Qualcomm (QCOM)

Smartphone modems + auto + edge AI. Apple iPhone modem business expected to fall away by 2027.

Q2 FY26: revenue $10.98B (+17% YoY). Forward P/E ~13× — among the cheapest large-cap semis. Dividend yield ~2.3%.

Catalyst: Auto and IoT diversification continues; PC Snapdragon X Elite ramping.

Risk: Apple iPhone modem revenue declining to zero. Smartphone unit volumes flat.

Verdict — Hold (cheap defensive): Real diversification value at a cheap multiple. The Apple cliff is the risk you take. Smaller sleeve, 3–5% position.

Ambarella (AMBA)

Edge computer vision and robotics SoC. The subject of much investor attention as the humanoid-robot trade developed.

Q1 FY27 revenue ~$87M (+58% YoY). Pivoting from automotive ADAS to robotics + edge AI. Forward P/E n/m (not yet GAAP profitable). Market cap ~$4.5B.

Catalyst: Robotics SoC design wins; ADAS resurgence.

Risk: Small-cap volatility. Cash burn until robotics scale.

Verdict — Add (speculative sleeve only): The cleanest public robotics CV pick. Small position only — 1–3% of an aggressive portfolio.

Monolithic Power Systems (MPWR)

Power management ICs for AI servers. Direct beneficiary of every NVDA Blackwell rack deployment.

Q1 2026: revenue $665M (+39% YoY). Forward P/E ~38×.

Catalyst: AI server PMIC content per rack rising.

Risk: Customer concentration with NVDA platforms.

Verdict — Add: Picks-and-shovels for AI server power. Direct correlation with NVDA platform shipments. Quality, not cheap.

Astera Labs (ALAB)

Connectivity / retimers for AI accelerator interconnect.

Q1 2026: revenue $159M (+144% YoY). Forward P/E ~70×+. Sub-$10B market cap.

Catalyst: PCIe Gen 6 ramp; CXL (compute express link) adoption.

Risk: Speculative multiple. Customer concentration.

Verdict — Watch: Real catalyst, real growth, real moat. The multiple is the issue. Wait for a 30%+ pullback.

Credo Technology (CRDO)

Active electrical cables and SerDes for AI data center interconnect.

Q3 FY26: revenue $135M (+182% YoY). Forward P/E elevated. Small cap.

Catalyst: 800G/1.6T cable transition. Hyperscaler design wins.

Risk: Smaller scale than ALAB. Optical migration could shrink the AEC TAM mid-term.

Verdict — Add (speculative sleeve): Pure-play AI interconnect at the cable layer. Smaller position than ALAB. 1–2% of an aggressive portfolio.

Layer 5 top picks: TXN as the analog defensive; WDC for contrarian-cheap AI data exposure; MPWR for AI server PMIC; AMBA for the speculative robotics sleeve.

Layer 6 — Infrastructure and power

Every NVIDIA Blackwell rack needs power, cooling, switchgear, and the grid behind it. Multi-quarter visibility comes from the same hyperscaler capex that funds the chips themselves.

Vertiv (VRT)

Power and cooling for AI data centers.

Q1 2026: net sales $2.65B (+30% YoY, 23% organic). Backlog $15B (~80% YoY growth). Book-to-bill ~2.9×. Americas revenue $1.81B (~70% of total) growing 44% organic. Adjusted operating margin 20.8% (+430 bps).

2026 guide raised: organic sales +29–31%, adj EPS $6.30–$6.40, net sales $13.5–14.0B. Forward P/E ~54×.

Catalyst: Liquid cooling penetration accelerating (Blackwell systems require it).

Risk: 54× forward P/E. Margin compression at scale. Competition from Schneider, ABB, Eaton.

Verdict — Core buy: The cleanest "AI capex happens, this happens" play that isn't a chip. Don't pay chip-multiples for it — size like a quality industrial with AI optionality.

Equinix (EQIX)

The premium data center REIT. Hyperscaler colocation + interconnection.

Q1 2026: revenue $2.34B (+10% YoY). Forward P/AFFO ~24×. Dividend yield ~2.5%.

Catalyst: AI data center demand continues to outstrip retail-colocation supply.

Risk: REIT tax treatment for Indian residents (different from equities — most foreign REITs are taxed as dividends from foreign entity, not capital gains; check with your tax advisor).

Verdict — Add (REIT sleeve): Real beneficiary of the data center build, dividend income, structurally durable. Indian tax treatment is the asterisk — consult your CA before sizing.

Digital Realty (DLR)

Larger, slightly more wholesale data center REIT than EQIX.

Q1 2026: revenue $1.45B (+9% YoY). Forward P/AFFO ~22×. Dividend yield ~3.5%.

Catalyst: Same as EQIX.

Verdict — Hold: The slower-growing twin of EQIX. Pick one for REIT exposure; EQIX edges DLR on growth, DLR edges on dividend.

Eaton (ETN)

Electrical equipment for AI infrastructure — switchgear, UPS, transformers. Plus diversified industrial portfolio.

Q1 2026: revenue $6.4B (+8% YoY). Operating margin 24%. Forward P/E ~27×. Dividend yield ~1.4%.

Catalyst: Data center electrical content per build rising; grid interconnect demand.

Risk: Diversified industrial exposure dilutes the AI thesis. Cycle exposure outside AI.

Verdict — Add: AI infrastructure beneficiary with diversification cushion. Lower volatility than VRT for similar theme. Good for portfolios that want the AI infrastructure exposure with less single-theme risk.

Quanta Services (PWR)

Utility-grade electrical construction. Grid build-out and renewable connectivity.

Q1 2026: revenue $6.2B (+24% YoY). Backlog $36B+. Forward P/E ~30×.

Catalyst: US grid interconnect projects; data center transmission line construction.

Risk: Project-based revenue lumpiness. Permitting delays.

Verdict — Add: The grid behind the data center. Highest leverage to "data centers require grid" thesis. 2–4% sleeve.

Vistra Energy (VST)

Independent power producer with nuclear fleet (Comanche Peak, plus the recent acquisitions). Has signed direct hyperscaler PPAs (the OpenAI-Stargate adjacent deals).

Q1 2026: revenue $4.34B (+18% YoY). Forward P/E ~22×. Catalysts include nuclear PPA execution.

Catalyst: New nuclear/hyperscaler PPAs; grid power pricing tailwinds.

Risk: Regulatory uncertainty around behind-the-meter nuclear power for data centers. Commodity power pricing volatility.

Verdict — Add: Power has become the binding constraint on AI buildout. VST is a leveraged play on power prices + nuclear monetization. Real catalyst, real risk — size 2–4%.

Constellation Energy (CEG)

Largest US nuclear fleet. Microsoft Three Mile Island restart deal.

Q1 2026: revenue $6.47B (+10% YoY). Forward P/E ~24×.

Catalyst: New hyperscaler nuclear PPAs (Three Mile Island restart by 2028).

Risk: Regulatory delays on nuclear restarts. Premium multiple for power.

Verdict — Add: Less leveraged than VST but cleaner exposure to nuclear baseload power for data centers. 2–4% sleeve.

GE Vernova (GEV)

Spun-off GE power generation. Gas turbines, wind, grid.

Q1 2026: revenue $8.0B (+12% YoY). Forward P/E ~45×. Backlog $116B.

Catalyst: Gas turbine demand for AI data center peaking power; grid orders.

Risk: 45× forward P/E. Wind segment exposure (loss-making).

Verdict — Hold: Real backlog story but the multiple has re-rated meaningfully. Don't chase at these prices. Existing holders stay.

Ciena (CIEN), Coherent (COHR), Lumentum (LITE)

The optical networking trio. Pluggables, transceivers, optical components for AI data center interconnect.

All three trade in the ~20–35× forward P/E range. CIEN is the systems integrator; COHR makes optical components and EUV lasers; LITE focuses on datacom transceivers.

Verdict — Watch (all three): The optical theme is real but executive volatility has hit each name. Wait for cleaner inflection points. If picking one, COHR has the broadest exposure (EUV laser business + datacom).

Layer 6 top picks: VRT (core), EQIX (REIT), VST + CEG (power), ETN + PWR (electrical/grid).

Three model portfolios

These are sketches, not advice. Position weights are within the semiconductor allocation — not your full portfolio. Sizing semi exposure inside your overall asset mix is the question above this one (rule of thumb: thematic sector exposure capped at 25–30% of an equity sleeve).

Defensive: capital preservation, structural moats only

For investors who want AI infrastructure exposure but cannot tolerate a 50% drawdown in a single year. Expected max drawdown 25–35% in a serious correction.

Layer	Name	Weight
Equipment	ASML	18%
Equipment	KLAC	8%
EDA	CDNS or SNPS	12%
Foundry	TSM	12%
AI compute	NVDA	12%
AI compute	AVGO	10%
Infrastructure	VRT	8%
Infrastructure	EQIX	5%
Analog defensive	TXN	5%
AI power	CEG	5%
Cash equivalent	—	5%

Logic: equipment + EDA = 38% of the sleeve. These are the most insulated tiers. TSM sized down for Taiwan risk. NVDA + AVGO as the barbell. Power and infrastructure for diversification.

On Rs 25 lakh deployed: roughly Rs 4.5L ASML, Rs 3L TSM, Rs 3L NVDA, Rs 2.5L AVGO, Rs 2L KLAC, etc. LRS budget: well within the ₹10L investment threshold for the FY in any 25–50 lakh annual deployment cadence.

Balanced: mix of core compounders and opportunistic exposure

For investors who want broader participation and can tolerate ~35–45% drawdown.

Layer	Name	Weight
Equipment	ASML	12%
Equipment	KLAC	5%
Equipment	AMAT	5%
EDA	CDNS	8%
EDA	SNPS	5%
Foundry	TSM	10%
AI compute	NVDA	10%
AI compute	AVGO	10%
AI compute	AMD	5%
AI compute	MRVL	4%
Networking	ANET	5%
Memory	MU	4%
Memory	WDC (contrarian)	3%
Infrastructure	VRT	6%
Infrastructure	EQIX	4%
AI power	VST	2%
AI power	CEG	2%

Logic: still ~30% equipment + EDA, but the AI compute layer broadens to include AMD + MRVL (hyperscaler diversification) and ANET (networking). Memory exposure capped at 7% combined. Infrastructure and power for cycle diversification.

Aggressive: maximum AI capex leverage

For investors with conviction in continued AI infrastructure build through 2028 and tolerance for 50%+ drawdowns.

Layer	Name	Weight
AI compute	NVDA	18%
AI compute	AVGO	14%
AI compute	AMD	8%
AI compute	MRVL	6%
Networking	ANET	8%
Equipment	ASML	10%
EDA	CDNS	5%
Foundry	TSM	8%
Memory	MU	5%
Infrastructure	VRT	8%
AI power	VST	4%
AI power	CEG	3%
Speculative	AMBA / ALAB / CRDO	3% combined

Logic: 54% in AI compute layer. Direct leverage to the buildout. ASML + TSM + CDNS = 23% in the most insulated tiers. Speculative sleeve for asymmetric upside.

What not to chase at these levels

The five categories that look like AI plays but carry more risk than the multiples compensate for:

Pure-play AI software without semi exposure — multiples expanded with no structural silicon moat. Different risk profile entirely.
China-listed semis ADRs (SMIC, etc.) — SMIC's ADRs were withdrawn from OTCQX on January 31, 2026 following a November 2025 executive order. Listing risk is real.
Intel at 100×+ forward — the turnaround is fully priced.
Micron at peak-cycle 81% gross margin — memory cycles invert fast.
Anything trading on a momentum chart without earnings to back the move — the rally compressed lots of names. The ones without contracted backlogs will be the first to compress.

The risk scenarios you should price in

Four scenarios that materially change the picture. Each one has different exposure consequences.

Taiwan crisis (semi-supply disruption). Highest tail risk in the stack. Direct semiconductor revenue concentrated at TSM (>70% global, >90% leading-edge). De-risking moves: under-weight TSM, over-weight ASML (sells to all foundries), over-weight equipment makers (post-event, capex will accelerate in non-Taiwan capacity). Add domestic-fab plays (GFS, INTC). Power and infrastructure layers largely unaffected by Taiwan supply (they are downstream demand stories).

China escalation (export controls tighten). Direct ACMR + SMIC exposure. Indirect ASML China shipment compression (recent rule rounds have already removed ~5–8% of ASML revenue). De-risking: avoid ACMR, prefer US-domiciled equipment and AI-compute names with diversified customer bases.

AI capex pause (hyperscaler ROI questions). The scenario where Microsoft, Google, or Meta pulls back 2027 capex guidance. Signals to watch: ROI-on-AI language in hyperscaler calls; "supply-balanced" language replacing "demand exceeds supply" framing; hyperscaler net-add slowdown. De-risking: trim NVDA + AVGO; rotate to power / data center REITs / equipment (already-contracted multi-year deliveries). The chips are most cyclical at the moment of pause; the picks-and-shovels keep shipping through it.

Memory cycle inversion. When DRAM/NAND supply catches demand. Direct exposure: MU. Indirect: LRCX (memory equipment). Signals to watch: HBM allocation language in hyperscaler calls; DRAM spot pricing rolling; capex announcements from SK Hynix/Samsung exceeding visible demand. De-risking: trim MU + LRCX; rotate to KLAC + foundry-focused equipment.

How an Indian-resident investor actually executes this

Practical layer:

Access. All names listed on NYSE or NASDAQ are accessible through Vested, INDmoney, IBKR, and Rovia. The structural cost differences are covered in Vested vs INDmoney vs IBKR vs Rovia — for a portfolio deploying Rs 25 lakh+ per year, IBKR's FX cost edge becomes material; for under Rs 10 lakh per year the Indian-aware platforms cost less in absolute terms.

LRS limits and TCS. Total LRS limit: USD 250,000 per FY. TCS on LRS for investment purposes: 20% above ₹10 lakh per FY (the ₹7 lakh figure applies to education / medical / tour purposes, not investment). For a Rs 25 lakh annual deployment, plan for roughly Rs 3 lakh of TCS that you'll reconcile against your tax liability at year-end.

Tax on gains. Foreign equity LTCG = 12.5% under Section 112 with a 24-month holding period. STCG = slab. Indexation removed for foreign equity post-Budget 2024. The 24-month holding period is real — selling at day 720 vs day 740 can cost ₹50k–₹2 lakh on a meaningful gain.

Schedule FA. Any of these names held at any point during the calendar year (January–December — Schedule FA is the one ITR schedule that doesn't use the financial year) triggers a Schedule FA disclosure obligation. Peak value, closing value (December 31), dividends, gross sale proceeds all need to be reported. The penalty for non-disclosure is ₹10 lakh per year of default under the Black Money Act. See the Schedule FA disclosure guide and the Schedule FA helper tool for the field-by-field walkthrough.

Dividend treatment. Most names here pay token dividends or none. AVGO, TXN, ADI, QCOM, EQIX, DLR are the meaningful payers. US dividend withholding is 25% under the India-US DTAA with a valid W-8BEN; the credit flows through Form 67 in your Indian ITR (renumbered Form 44 from Tax Year 2026-27).

Currency exposure. When you buy any of these you're long USD assets against the rupee. The 24-month holding period for LTCG also locks in USD-INR exposure for that window. Build that into position sizing — USD-INR is not a free option.

The closing

The big four hyperscalers — Microsoft, Amazon, Google, Meta — combined are tracking roughly $650–700 billion of capex in 2026. Google Cloud alone reported a $460 billion backlog. Broadcom's AI backlog is $73 billion. Vertiv's is $15 billion at +80% year-over-year. ASML's order book is €38.8 billion. NVIDIA shipped $75.2 billion of data center revenue in a single quarter.

These aren't forecasts. They're contracts.

After the rally is the worst time to chase. It is the best time to own the businesses whose next ten quarters are already in the order book — not in the model. Pick your layer. Size your conviction. Hold long enough to outlast the next chart.

The forums see the chart. The order books are already in 2028.