Educational & illustrative only — not financial advice. This is a hypothetical model portfolio created to explain how the AI supply chain fits together. It is not a recommendation to buy or sell any security and is not tailored to anyone's personal circumstances.
Educational Investment Research · June 2026
Miss AI · Free
High-Growth AI
Portfolio.
A hypothetical US$1,000,000 model across the artificial-intelligence supply chain. 18 holdings. 3 risk tiers. Every layer mapped.
Keira Nesdale · @RealMissAI
↓ Download PDF
Section 1
The hypothetical
allocation.
The model is organised into three risk tiers. Roughly 64% sits in lower-volatility core anchors, 22% in core-growth/cyclical names, and 14% in higher-risk satellites. The weightings below are illustrative — they show one way a high-growth AI book could be structured, not a recommendation to anyone.
| Ticker | Company | Segment | Weight | USD |
|---|
| NVDA | NVIDIA | AI compute / GPUs | 13% | $130,000 |
| GOOGL | Alphabet | Cloud / AI platform | 9% | $90,000 |
| MSFT | Microsoft | Cloud / AI platform | 8% | $80,000 |
| AMZN | Amazon | Cloud / commerce | 7% | $70,000 |
| AVGO | Broadcom | Custom AI ASICs / networking | 6% | $60,000 |
| TSM | Taiwan Semiconductor | Foundry | 6% | $60,000 |
| ASML | ASML | EUV lithography tools | 5% | $50,000 |
| GEV | GE Vernova | Power generation / grid | 5% | $50,000 |
| CEG | Constellation Energy | Nuclear generation | 5% | $50,000 |
| Subtotal | 64% | $640,000 |
| AMD | Advanced Micro Devices | AI compute / GPUs (stretched) | 5% | $50,000 |
| VST | Vistra | Nuclear + gas power (value) | 5% | $50,000 |
| MU | Micron | HBM / memory (cyclical) | 4% | $40,000 |
| CRWD | CrowdStrike | AI-native security (stretched) | 4% | $40,000 |
| CCJ | Cameco | Uranium / nuclear fuel | 4% | $40,000 |
| Subtotal | 22% | $220,000 |
| VRT | Vertiv | Data-centre power / cooling | 4% | $40,000 |
| ORCL | Oracle | AI cloud infrastructure | 4% | $40,000 |
| TRGP | Targa Resources | Natural-gas midstream | 3% | $30,000 |
| IONQ | IonQ | Quantum computing (speculative) | 3% | $30,000 |
| Subtotal | 14% | $140,000 |
| Grand total — 18 holdings (hypothetical) | 100% | $1,000,000 |
Section 2
The AI supply chain.
AI is not one industry; it is a stack of dependent layers. Money flows from the cloud giants downward into chips and power, and value concentrates wherever a single company is hard to replace. The model deliberately includes a position in every layer.
Reading the chain: a dollar of AI spending starts at the top (a hyperscaler's budget), flows down through cloud → GPUs/ASICs → memory → foundry → tools, and sideways into the power and security needed to run it all. The strongest monopolies — ASML, NVIDIA's CUDA, TSMC's leading-edge node — are where pricing power is most durable, which is why they carry core weightings.
1
Design tools & lithography
ASML
ASML is a true monopoly in extreme-ultraviolet (EUV) lithography — no advanced AI chip below ~5nm can be made without its machines. Effectively single-source.
ASML makes the only 'printing presses' capable of etching the world's most advanced chips. If you want to build a cutting-edge AI chip, you must buy from ASML. That gives it huge pricing power — it sits at the very bottom of the chain, so everyone above depends on it.
2
Chip design — GPUs
NVDA, AMD
NVIDIA holds an estimated ~80–90% of the AI training-GPU market and owns the CUDA software moat. AMD is the credible #2 (MI300/MI450) and the main alternative hyperscalers fund to avoid sole dependence on NVIDIA.
GPUs are the 'brains' that do the actual AI maths. NVIDIA makes the best ones AND owns the software (CUDA) that developers have spent years learning — so switching is painful and costly. AMD is the only serious alternative, which big tech deliberately backs. Owning both means exposure whichever way that contest goes.
3
Custom silicon & networking
AVGO
Broadcom co-designs custom AI ASICs for hyperscalers (Google TPUs, Meta, OpenAI) and dominates high-end networking silicon — the connective tissue between GPU clusters. A near-duopoly with Marvell.
The tech giants increasingly want their own bespoke AI chips. Broadcom is the partner that helps them design those, and also makes the high-speed 'plumbing' that lets thousands of chips talk to each other inside a data centre. Few companies can do either job.
4
High-bandwidth memory (HBM) is an oligopoly of three — Micron, SK Hynix, Samsung. Micron's HBM capacity has been reported sold out well into the future; memory has repriced as a strategic asset.
AI chips need ultra-fast memory sitting right beside them to feed them data. Only three companies on earth can make it. Demand is so high that Micron has effectively pre-sold its output for years — but memory is famously cyclical, so shares swing hard in both directions.
5
Manufacturing (foundry)
TSM
TSMC manufactures the leading-edge chips for NVIDIA, AMD, Apple and others — an estimated ~65%+ of the global foundry market and effectively the only volume source for the most advanced nodes.
NVIDIA, AMD and Apple design chips but don't actually build them — they hand blueprints to TSMC to manufacture. For the most advanced chips it's essentially the only option at scale. Almost every AI chip on earth passes through TSMC's factories. Those factories are in Taiwan — strength and risk.
6
Cloud platforms
MSFT, GOOGL, AMZN, ORCL
An oligopoly: AWS ~30%, Azure ~20%, Google Cloud ~13% of cloud infrastructure. Oracle is the fast-growing challenger via the OpenAI/Stargate build-out.
This is where AI is actually rented out and sold to businesses. These three control most of the market, and it's hard to dislodge them because customers' data and software already live there. Oracle is the fast-rising challenger. This is the layer that turns all AI spending into paying customers.
7
CrowdStrike is a leader in AI-native endpoint and cloud security — a fragmented but consolidating market where it is one of the clear platform winners.
More AI and more data in the cloud means a bigger target for hackers. CrowdStrike is one of the leaders that protects all of it, using AI to spot attacks in real time. Once installed across thousands of devices, ripping it out is disruptive — customers tend to stay and spend more each year.
8
Power generation & grid
GEV, CEG, VST
No monopoly, but scarce assets: GE Vernova is one of a handful of global suppliers of gas turbines and grid equipment; Constellation owns the largest US nuclear fleet; Vistra is a top independent nuclear/gas generator. AI data centres are power-constrained, making these chokepoints.
AI data centres are astonishingly hungry for electricity — a single campus can use as much power as a small city. There isn't enough to go around, so whoever can supply reliable power holds the leverage. Power is now the real bottleneck for AI growth, which is why this sleeve is large.
9
Fuel & power infrastructure
CCJ, TRGP, VRT
Cameco is a top-two Western uranium producer; Targa is a major US gas-midstream operator; Vertiv is a leading data-centre power & thermal-management supplier and NVIDIA partner.
Cameco digs up the uranium that fuels nuclear plants. Targa moves the natural gas that fires many power stations. Vertiv makes the specialised cooling and power gear that stops racks of red-hot AI chips from melting. These are the unglamorous 'pick-and-shovel' suppliers that get paid no matter which AI company ultimately wins.
10
Frontier compute (optionality)
IONQ
Quantum is pre-commercial. IonQ is a pure-play leader in trapped-ion systems — a lottery ticket on a market that barely exists yet but could be transformative.
Quantum computers are a completely different, experimental kind of machine that could one day crack problems today's computers can't touch. IonQ is one of the front-runners. This is a small, high-risk allocation: if quantum inflects this decade the payoff could be enormous; if it doesn't, the position is kept small enough not to hurt.
Section 3
Sector growth
forecasts.
Sources include IDC, McKinsey, Goldman Sachs, Gartner, IEA, Grand View Research, and MarketsandMarkets.
| Sector | Current → 2030 | Key data point |
|---|
| Semiconductors (total) | ~$680bn → ~$1.6–1.75tn | Data-centre chips alone ~$843bn by 2030 |
| AI data-centre capex | ~$405bn → ~$1.0–1.4tn/yr | Goldman models ~$7.6tn cumulative 2026–2031 |
| Cloud computing | ~$750bn–1.3tn → ~$1.6–2.4tn | ~12–20% CAGR |
| AI power / DC electricity | US DC ~33GW → ~95–120GW | Power demand +175% by 2030 |
| Nuclear / uranium | Demand +~28% by 2030 | WNA 5.3% CAGR to 2040 |
| AI in cybersecurity | ~$25bn → ~$94–134bn | ~24–28% CAGR |
| Quantum computing | ~$1.7–3.5bn → ~$4–20bn | ~20–42% CAGR but tiny base; pre-commercial |
Section 4
The modern
kingpins' lens.
Eight investors evaluated. Based on Q1 2026 13F filings (positions as of 31 March 2026) and public commentary.
Buffett / Abel
Berkshire Hathaway
Approves
Approves the anchors, avoids the speculative tail. Berkshire tripled its Alphabet stake into a top-5 holding — their signature 2026 bet.
Druckenmiller
Duquesne
Approves
Approves the infrastructure spread, picky on price. One of the closest matches to this model's thesis.
Laffont
Coatue
Approves
"This is his neighbourhood." Most comfortable in the AI infrastructure space of any major fund.
Tepper
Appaloosa
Approves
Tepper's book is the closest match to this model's thesis. Amazon is his largest position (~15%). He doubled Vistra — his clearest AI-power conviction add.
Ackman
Pershing Square
Selective
Added Amazon (+19%), new buy on Microsoft. Sold out of Alphabet entirely. Selective rather than broad approval.
Klarman
Baupost
Wait / Avoid
At forward multiples of 30–170x he would find little safety here. Would mostly wait for lower prices.
Howard Marks
Oaktree
Caution
Proceed with caution, respect the cycle. Would not deploy aggressively at current valuations.
Michael Burry
Scion
Short It
Betting the bubble bursts. Has roughly a US$1.1bn bet against the AI complex. Views current valuations as a repeat of the late-1990s dot-com build-out.
The growth-oriented kingpins (Laffont, Druckenmiller, Tepper) are comfortable in this neighbourhood. The value-and-cycle voices (Klarman, Marks) would wait for lower prices. Burry is betting against the trade outright. The most-owned name across funds is Amazon. Alphabet splits the room: Berkshire tripled it while Ackman, Druckenmiller and Altimeter sold out.
Section 5
Contrarian views
& risks.
Bottom line: a model like this is built for a high-risk, high-growth profile and could lose a large portion of its value in a downturn.
Risk 01
The circular financing / bubble critique
NVIDIA invests $100bn in OpenAI, which buys NVIDIA chips; AMD grants OpenAI warrants for ~10% of itself in exchange for GPU orders. Critics compare this to the late-1990s dot-com build-out. Michael Burry has ~US$1.1bn bet against the AI complex.
Risk 02
Three holdings already trade above consensus
MU, CRWD, and AMD have run past their average analyst targets. Average target implies downside from current prices for these three positions.
Risk 03
Extreme thematic concentration
~100% in one theme — AI infrastructure. If the AI capex cycle disappoints, or if regulatory action targets the sector, there is nowhere to hide in this model.
Risk 04
Valuations are stretched across the board
Forward multiples of 30–170x leave little room for disappointment. Any miss on earnings or guidance could produce large drawdowns quickly.
Risk 05
Hyperscaler capex could disappoint
The thesis assumes ~$700bn+ of 2026 spending keeps rising. Any slowdown or reallocation in hyperscaler capex (Microsoft, Google, Amazon, Meta) would ripple through the whole chain.
Risk 06
Single-name execution risk
Oracle's heavy debt load, nuclear-restart timelines, IONQ's cash burn, TSMC's geopolitical exposure (Taiwan). Each position carries idiosyncratic risk that could materialise independently.
Miss AI
More AI research
& build guides.
For more AI workflows, prompt packs, build guides and educational content for founders and creators, come find me.
← Back to vault
Legal disclaimer: this document is educational and illustrative only and does not constitute financial advice. The hypothetical portfolio is not a recommendation to buy, sell, or hold any security. It is not tailored to any individual's objectives, time horizon, risk tolerance, or financial situation. All figures are in US dollars and are illustrative; they do not account for transaction costs, taxes, currency risk, or individual circumstances. Past performance and analyst price targets are not guarantees of future results. Do your own research and consult a licensed financial adviser before making any investment decisions. Market data sourced from TipRanks, MarketBeat, Yahoo Finance, Investing.com, GuruFocus, Barchart, Zacks (early June 2026); 13F filings via SEC EDGAR; sector forecasts from IDC, McKinsey, Goldman Sachs, Gartner, IEA, Grand View Research, MarketsandMarkets, BCC Research; company releases and SEC filings via public sources.