Own the bottlenecks.
Not just the narrative.
A concentrated, 30-stock global portfolio spanning the complete AI value chain—from power and silicon to models, applications and the scarce inputs between them.
Physical scarcity led the result.
Chips and infrastructure contributed +24.32 INR percentage points together. The strongest returns came from HBM, security and datacentre power-and-cooling exposure.
of the portfolio’s INR return came from Chips & Compute plus Physical Infrastructure—the two most direct scarcity sleeves.
Five layers. Six ways to win.
Sixty-eight percent sits in the physical stack—energy, chips and infrastructure—where long lead times and engineering complexity create the strongest bottlenecks.
Global by merit.
Not by quota.
Nine markets supply distinct pieces of the AI stack. Each non-US position displaced a credible incumbent; none was added simply to make the map look broader.
Platforms over prototypes.
Tencent, Alibaba and Baidu provide models, cloud, distribution and cash flow. SMIC and newly listed labs remain candidates, not compulsory holdings.
Two quiet chokepoints.
Sumitomo Electric spans AI fiber and power cable; Lasertec owns a highly specialized EUV inspection niche.
Industrial scarcity.
ASML, Siemens Energy, Schneider and RELX contribute lithography, grid hardware, electrical systems and proprietary data.
Exposure through copper.
Freeport's global mine base captures copper intensity without adding a lower-conviction regional name merely for domicile.
Every position has a job.
Filter the portfolio, rank the evidence, and open any row for the full selection rationale and principal risk.
| Company / strategic role | Listing | Target | Now | INR return | INR contribution | |
|---|---|---|---|---|---|---|
10 000660 SK hynix ↗HBM memory leader | KRXSouth Korea | 4.5% | 6.2% | +85.3% | +3.84 pts | |
23 PANW Palo Alto Networks ↗AI-era security platform | NasdaqUnited States | 2.0% | 3.9% | +162.8% | +3.26 pts | |
12 VRT Vertiv ↗Power density + liquid cooling | NYSEUnited States | 4.5% | 5.3% | +60.6% | +2.73 pts | |
06 NVDA NVIDIA ↗System-level AI compute | NasdaqUnited States | 8.0% | 7.9% | +33.9% | +2.71 pts | |
13 2308 Delta Electronics ↗Efficient power + thermal | TWSETaiwan | 4.0% | 4.9% | +64.6% | +2.58 pts | |
11 ANET Arista Networks ↗Ethernet AI fabrics | NYSEUnited States | 4.5% | 5.2% | +57.2% | +2.58 pts | |
09 AVGO Broadcom ↗Custom silicon + networking | NasdaqUnited States | 5.5% | 5.9% | +44.2% | +2.43 pts | |
07 TSM TSMC ↗Foundry + advanced packaging | NYSE ADRTaiwan | 6.5% | 6.5% | +34.4% | +2.23 pts | |
22 NET Cloudflare ↗Edge AI + security fabric | NYSEUnited States | 2.0% | 3.0% | +101.2% | +2.02 pts | |
08 ASML ASML ↗EUV lithography monopoly | Nasdaq ADRNetherlands | 5.5% | 5.5% | +36.3% | +2.00 pts | |
14 FIX Comfort Systems USA ↗Skilled datacentre deployment | NYSEUnited States | 3.5% | 4.0% | +54.3% | +1.90 pts | |
01 GEV GE Vernova ↗Turbines + grid equipment | NYSEUnited States | 4.5% | 4.7% | +42.1% | +1.89 pts | |
15 EQIX Equinix ↗Neutral interconnection | NasdaqUnited States | 3.5% | 3.6% | +37.9% | +1.33 pts | |
03 SU Schneider Electric ↗Power from grid to rack | Euronext ParisFrance | 3.5% | 3.3% | +27.9% | +0.98 pts | |
18 AMZN Amazon ↗Open cloud compute platform | NasdaqUnited States | 3.0% | 2.9% | +31.9% | +0.96 pts | |
27 LITE Lumentum ↗Photonics bottleneck | NasdaqUnited States | 1.0% | 1.3% | +70.4% | +0.70 pts | |
16 GOOGL Alphabet ↗Full-stack model platform | NasdaqUnited States | 4.0% | 3.4% | +14.2% | +0.57 pts | |
04 CEG Constellation Energy ↗24/7 operating power | NasdaqUnited States | 3.5% | 3.0% | +16.0% | +0.56 pts | |
25 RELX RELX ↗Proprietary vertical data | NYSE ADRUnited Kingdom | 1.5% | 1.4% | +28.5% | +0.43 pts | |
02 ENR Siemens Energy ↗Global grid second source | XetraGermany | 4.0% | 3.3% | +10.1% | +0.40 pts | |
30 6920 Lasertec ↗EUV mask inspection | TokyoJapan | 0.8% | 0.9% | +44.4% | +0.36 pts | |
29 FCX Freeport-McMoRan ↗Copper intensity | NYSEUnited States | 1.0% | 0.9% | +26.4% | +0.26 pts | |
28 5802 Sumitomo Electric ↗Photonics + grid cabling | TokyoJapan | 1.0% | 0.9% | +23.9% | +0.24 pts | |
26 0981 SMIC ↗China strategic foundry option | Hong KongChina | 1.2% | 0.9% | +3.2% | +0.04 pts | |
05 267260 HD Hyundai Electric ↗High-voltage transformers | KRXSouth Korea | 2.5% | 1.8% | -3.5% | -0.09 pts | |
19 0700 Tencent Holdings ↗China models + distribution | Hong KongChina | 3.5% | 2.4% | -8.3% | -0.29 pts | |
21 TSLA Tesla ↗Physical-AI option | NasdaqUnited States | 2.5% | 1.6% | -12.0% | -0.30 pts | |
17 SPCX SpaceX ↗Frontier model + hyperscale platform | NasdaqUnited States | 3.5% | 2.4% | -8.9% | -0.31 pts | |
24 9888 Baidu ↗China AI cloud + autonomy | Hong KongChina | 2.0% | 1.2% | -19.9% | -0.40 pts | |
20 9988 Alibaba Group ↗China cloud + open models | Hong KongChina | 3.0% | 1.9% | -16.4% | -0.49 pts |
The portfolio thesis, company by company.
The core barbell owns hard-to-replicate bottlenecks and scaled distribution. Application and tactical positions remain deliberately smaller.
GE Vernova
Turbines + grid equipmentThe cleanest public combination of gas turbines, grid hardware and long-duration service revenue. A $176bn backlog and more than $5bn of 2026 data-centre orders turn the AI power thesis into contracted demand.
Siemens Energy
Global grid second sourceA global alternative to GE Vernova in turbines and a major transformer supplier. Record orders show the grid constraint is worldwide, while advance payments and service revenue improve visibility.
Schneider Electric
Power from grid to rackMonetizes breakers, UPS, power distribution, automation and cooling integration across architectures. It is less binary than a generator and data centres remain the leading Energy Management growth driver.
Constellation Energy
24/7 operating powerOperating nuclear and gas assets supply firm power today rather than promising future capacity. Its enlarged fleet and signed data-centre arrangements create direct leverage to scarce round-the-clock electricity.
HD Hyundai Electric
High-voltage transformersA focused beneficiary of transformer and grid-equipment shortages. Its 2026 targets explicitly cite AI data centres and global grid expansion, reinforced by participation in the tight US 765kV market.
The portfolio is concentrated.
The process cannot be.
This is an aggressive thematic model. Position caps, evidence gates and geopolitical limits matter more than false precision in a long-term target price.
Capex / ROI air pocket
MOST EXPOSEDGEV · VRT · FIX · ANET · AVGO · LITE
MONITOR / RESPONSECloud backlog, supplier book-to-bill, cancellations and hyperscaler capex guidance.
Taiwan / China rupture
MOST EXPOSEDTSM · 2308 · NVDA · AVGO · 0700 · 9988 · 9888 · 0981
MONITOR / RESPONSEKeep any one China name below 4%, TSM below 7%, and review export controls quarterly.
Valuation compression
MOST EXPOSEDSPCX · NET · TSLA · LITE · VRT · GEV · SK hynix
MONITOR / RESPONSEBuild in tranches; do not average down after a thesis break.
Efficiency shock
MOST EXPOSEDNVDA · TSM · SK hynix · power names
MONITOR / RESPONSETrack aggregate tokens, cloud demand and electricity—not benchmark efficiency alone.
Power / permitting delays
MOST EXPOSEDCEG · EQIX · FIX · VRT · GEV
MONITOR / RESPONSEWatch energization, interconnection queues and customer prepayments.
Model commoditization
MOST EXPOSEDGOOGL · SPCX · AMZN · 0700 · 9988 · 9888
MONITOR / RESPONSERequire cloud, distribution and cash flow; avoid pure model laboratories at venture-like valuations.
Evidence before enthusiasm.
These rules turn the thesis into an investable process.
- Enter in three tranches: 40% initially, 30% after the next earnings cycle and 30% on a drawdown or thesis confirmation.
- Review quarterly and conduct a full thesis audit every six months; rebalance for risk or evidence, not merely because a stock rose.
- Trim any single position above 12%, any layer above 35%, any one China holding above 4%, and keep tactical names at or below 1.5% each.
- For physical names, two quarters of book-to-bill below one, cancellations or declining backlog margins force a review.
- For platforms, capex growing materially faster than AI revenue without better backlog or utilization forces a weight reduction.
- For applications, AI ACV, RPO or adoption must outgrow the company baseline to prove the exposure is incremental.
China's lithography champion is not publicly listed; no direct security can enter the portfolio.
Compelling robotics leader, but IPO subscription only began on 10 August; wait for trading, liquidity and public results.
Frontier-model exposure is real, but post-listing valuations and losses do not yet beat Tencent or Alibaba.
Excellent global electrification names; the five selected energy incumbents have stronger direct AI order evidence.
High-quality Japan beneficiaries, but current multiples leave less margin of safety than Sumitomo Electric.
Important magnet supply optionality, but AI demand is more indirect than copper, optics or power cabling.
Direct AI capacity growth with financing, customer-concentration and residual-value risk.
The investable universe was screened across North America, Europe, China, Japan, Korea, Taiwan, India, Southeast Asia, Australia, Africa and Latin America. A name enters only when its evidence and risk/reward beat the incumbent slot.
How to read the number.
This is a retrospective hypothetical backtest, not a live track record. The final 30-stock portfolio was selected using information available through 11 August 2026 and is simulated from 5 February, so the result contains hindsight and selection bias. Holdings received target weights at the adjusted close on that start date and then remained buy-and-hold with fractional shares and no rebalancing. The 3.5% SpaceX model-and-platform allocation is held as USD cash until the stock's first trading close on 12 June, then invested. NAV is shown for every weekday through 10 Aug 2026.
US ADRs are used for TSMC, ASML and RELX. Tencent, Alibaba, Baidu and SMIC use their Hong Kong listings; Siemens Energy and Schneider are priced in euros; HD Hyundai Electric and SK hynix in won; Delta in Taiwan dollars; Sumitomo Electric and Lasertec in yen. Local adjusted closes are converted first to USD using contemporaneous Yahoo Finance FX series. When a local market is closed, its last available close is carried forward.
Every return figure is INR-denominated: the portfolio NAV and daily chart, holding returns, layer returns, contribution, attribution, drawdown, volatility, best day and worst day. Each USD total-return series is translated daily through USD/INR before the INR return is calculated. Every monetary value shown—portfolio value, gain, allocations and holding values—remains in USD. No rupee monetary values are displayed.
Adjusted closes incorporate splits and cash distributions. Use the refresh control in the top-right corner to fetch and recalculate the latest completed US market session. A successful refresh becomes the shared verified snapshot every visitor sees; if any required price or FX input fails validation, the prior snapshot remains live. The model excludes brokerage, bid/ask spreads, withholding tax, investor-level tax and custody constraints. A $100,000 notional is included only to make the NAV intuitive; this is not a brokerage statement.