Investment Thesis
$1.5T in Committed Capex
Microsoft, Google and Amazon have announced a combined $1.5T in AI infrastructure spending for 2026–2027 — the largest coordinated corporate investment in history. Every dollar flows through our universe of names.
Chips, Not Chatbots
We own the companies that make AI possible: NVDA, AVGO, MRVL, TSM, ASML. These have direct, measurable revenue tied to AI compute. Not AI hype — AI plumbing. The winners here are knowable regardless of which AI application wins.
Near-Monopoly Chokepoints
TSMC is the only foundry capable of manufacturing leading-edge AI chips. ASML sells the only machines that can make them. These structural positions take 10–15 years to replicate — pricing power is durable.
Memory Supercycle
Each NVIDIA H100 requires 80GB of high-bandwidth memory. Micron is the only US-listed pure HBM play. Every new GPU generation increases this memory requirement — the addressable market expands automatically.
Optical Interconnect
800G transceivers are the arteries between GPU clusters. Copper cable cannot carry enough data at AI scale — optical fibre is mandatory. Coherent, Lumentum and Fabrinet hold multi-year hyperscaler contracts already signed.
The Electricity Problem
A single AI data centre can consume 500MW. US data centres will take 8% of total grid capacity by 2030. GE Vernova, Eaton and Constellation Energy benefit from a power deficit that cannot be fixed in under five years.
GPU Cloud Operators
CoreWeave and Nebius rent GPU capacity to AI companies that cannot build their own data centres. This market did not exist three years ago. CoreWeave guided $12B revenue for 2026 — faster growth than any prior cloud platform at equivalent scale.
Racks, Switches & Real Estate
Every AI chip needs a server, a rack, a switch, and a data centre. Arista moves data between GPUs at 800Gbps; Vertiv handles thermal management; Equinix and Digital Realty own the physical infrastructure.
Frontier Positions
Small, high-conviction positions in emerging names: Rocket Lab (space connectivity), IREN (GPU cloud with $3.4B NVIDIA contract), Amprius (next-generation battery technology for data centres).
AI Infrastructure Value Chain
01Hyperscalers
MSFT · GOOGL · AMZN
Microsoft, Google and Amazon have committed $1.5T in AI infrastructure capex for 2026–2027. They are the demand engine — every dollar flows down into chips, servers, power and fibre. We hold them as anchor positions, not as the core thesis.
↓
02Semiconductors
TSM · ASML · AMAT · LRCX · NVDA · AMD · AVGO
The most critical layer. TSMC fabricates every leading AI processor; ASML sells the only machines that can print them. No chip, no AI. These positions carry near-monopoly pricing power and are the highest-conviction holdings in the portfolio.
↓
03Servers & Networking
ANET · VRT · CLS · APH · EQIX · CSCO · HPE
AI servers are assembled, racked and cooled inside data centres. Arista switches move data between GPUs at 800Gbps; Vertiv handles thermal management. Equinix and Digital Realty own the physical real estate this infrastructure sits in.
↓
04Optical Interconnect
COHR · CRDO · LITE · FN · CIEN · GLW
800G and 1.6T optical transceivers are the bandwidth arteries between GPU clusters. At AI scale, copper cable cannot carry enough data — optical fibre is the only solution. Coherent, Lumentum and Fabrinet hold signed multi-year contracts.
↓
05Memory & Storage
MU · PSTG · WDC · SNDK · STX
Training a large AI model requires petabytes of fast memory. HBM3 (Micron) sits directly on the GPU die; Pure Storage all-flash arrays hold training datasets at the speeds AI workloads demand. Memory content grows with every chip generation.
↓
06Power & Electricity
GEV · BE · ETN · CEG · VST · NVT
A single AI data centre consumes up to 500MW — as much as a small city. US data centres will use 8% of total grid capacity by 2030. GE Vernova, Eaton and Constellation Energy address a power constraint that physical infrastructure cannot solve in under five years.
↓
07Neocloud & GPU Compute
NBIS · CRWV · CORZ · SMCI · DELL
Independent GPU cloud operators — CoreWeave, Nebius — rent compute to AI companies that cannot build their own data centres. CoreWeave guided $12B revenue for 2026, growing faster than AWS did at equivalent scale.
Portfolio Composition — v6, 54 names · 9 segments
Risk & Behaviour
Capital at risk. This strategy holds concentrated, high-growth technology equities. It will be significantly more volatile than MSCI World or the Nasdaq-100. Drawdowns of 40–60% during risk-off episodes or AI sentiment reversals are realistic — and should be expected before the full investment horizon plays out. This is a long-term, high-conviction position, not a short-term trade.
Concentration Limits
No single name exceeds 6% of the portfolio (currently: Micron at 4.5%). No single segment exceeds 16% (currently: Broad AI Anchor). Hyperscaler capex dependency means 4 companies — Microsoft, Google, Amazon, Meta — drive demand for most of our holdings simultaneously.
Valuation & Sentiment
Several holdings trade at meaningful premiums to historical multiples, priced for sustained AI capex growth. An earnings miss, a capex guidance cut from one major hyperscaler, or a shift in AI sentiment could trigger sharp multi-name de-ratings — with no warning. Short-term volatility is the price of long-term positioning.
Geopolitical & Regulatory
TSMC (Taiwan), ASML (Dutch export controls), and the broader semiconductor supply chain carry geopolitical risk. US-China trade policy is the single largest external variable. New US export control rounds could restrict sales of certain chip equipment overnight.
Realistic Drawdown — Risk-Off
−30 to −45%
Broad market sell-off or rate spike; AI fundamentals intact. We hold and may add.
Realistic Drawdown — AI Sentiment Reversal
−45 to −60%
Hyperscaler capex guidance cut or major chip shortfall. We reassess but do not automatically exit.
Realistic Drawdown — Structural Break
>−60%
Taiwan conflict, sweeping export ban, or genuine AI demand collapse. Thesis review triggered.
We Would Exit or Reduce If
Two or more hyperscalers materially cut AI capex guidance for 12+ months
A credible TSMC alternative emerges with a 2-year time-to-market
US legislation bans investment in Taiwan-domiciled semiconductor companies
AI compute demand plateaus for 18+ months across multiple end-markets
We Would Tolerate and Hold Through
Drawdowns of 40–60% driven by sentiment, rates, or macro — not fundamentals
Single-name blow-ups below 3% position size
Quarterly earnings misses on weather/supply disruption, not demand
12–18 months of underperformance vs Nasdaq during non-AI-driven rallies
AMC Terms & Portfolio Role
| Structure | AMC | Actively Managed Certificate — a listed, single-name instrument tracking a Blanc WM-defined basket. Held via standard brokerage account; no fund structure or lock-up. |
| Currency | USD | All holdings listed in USD. Non-USD clients bear USD/base currency risk in addition to equity risk. |
| Portfolio Role | Satellite | Intended as a satellite allocation of 5–15% of global equity exposure — not a core holding. It complements a diversified core (MSCI World, developed market fixed income) and increases the portfolio's overall technology and growth exposure meaningfully. |
| Min. Horizon | 3 yr | Minimum recommended holding period. The AI infrastructure buildout is a multi-year capex cycle; shorter horizons risk realising drawdowns at cycle troughs rather than participating in the recovery. |
Suitability Summary
Client-Facing Paragraph
The Blanc WM AI Infrastructure AMC is a concentrated, actively managed basket of 54 USD-listed equities providing exposure to the full physical infrastructure layer of artificial intelligence — semiconductors, data centres, cloud compute, optical interconnect, and power grid. Its role in a client portfolio is as a satellite allocation of 5–15% of total global equity exposure, complementing a diversified core and increasing structural exposure to one of the most significant technology investment cycles of the past three decades. The strategy carries substantially higher volatility and drawdown risk than MSCI World or the Nasdaq-100: investors should anticipate peak-to-trough drawdowns of 40–60% during risk-off episodes or periods of negative AI sentiment, without those drawdowns constituting a change in the fundamental investment case. The minimum recommended holding period is three years; clients unable to commit capital for this period or who cannot tolerate interim negative returns should not invest. Key risks include concentrated hyperscaler capex dependency, semiconductor geopolitical exposure (principally Taiwan and US-China export controls), valuation risk across multiple holdings, and the possibility of structural obsolescence of specific technologies within the basket.