Est. 2026 · AI-Native Long / Short
Hyperleap Capital is a hedge fund built around a single thesis: the buildout of artificial intelligence is the defining economic event of our lifetime, and public markets are still pricing it one bottleneck at a time. We invest across the entire chain — silicon, power, and the firms that AI remakes.
Flagship Fund
HLC Intelligence Master
Illustrative chart: HLC Intelligence composite versus MSCI World over the trailing 72 months. Figures are placeholders — see the performance table below for the full illustrative track record.
Ann. Return
—
Sharpe
—
Max DD
—
—B
Assets under management
—
Core positions
—%
Gross in the AI complex
—yrs
Average holding period
01 / Thesis
Most capital allocates to artificial intelligence the way it allocates to an industry. We underwrite it as a macro regime — one that reorders industrial capacity, energy demand, and corporate earnings power faster than consensus models can revise them.
01
Training compute continues to advance by orders of magnitude, and algorithmic efficiency compounds on top of it. Capability improvement remains a function of inputs we can count — wafers, memory, megawatts — long before it shows up in a revenue line.
02
Every eighteen months the binding constraint relocates: leading-edge logic, then high-bandwidth memory, then advanced packaging, then interconnect queues, then electricity itself. Return accrues to whoever owns the constraint before the market names it.
03
Markets discount transformative outcomes toward the mean because career risk punishes early conviction. That asymmetry is durable, and it is the reason a concentrated, thesis-driven book can outperform for years rather than quarters.
02 / Portfolio
The book is concentrated and long-biased, hedged with short exposure to the business models that AI erodes. Capital moves along the supply chain as the bottleneck migrates.
01
Leading-edge logic, high-bandwidth memory, advanced packaging, networking silicon, and the equipment makers with no substitute. The narrowest part of the chain and the hardest to replicate.
02
Generation, transmission, turbines, transformers, and thermal management. Frontier training runs are electricity problems long before they are software problems, and the grid cannot be scaled on software timelines.
03
Long the firms whose unit economics inflect once inference is cheap. Short the seat-based, labour-arbitrage, and attention-rent models whose moats dissolve at the same moment — the funding leg of the book.
04
Selective positions in frontier labs, neoclouds, and data-centre infrastructure where no public expression exists yet. Sized as option premium, not as core risk.
03 / Research
Our work is closer to industrial forecasting than to equity research. We build the physical picture first — how much silicon, how many megawatts, on what schedule — and only then ask what is in the price.
Fab starts, HBM allocation, packaging capacity, transformer and turbine lead times, interconnect queues. Bottom-up, unit by unit, from primary sources.
Only one input binds at a time. We identify it, verify who monetises it, and establish the position before the shortage becomes a headline.
Positions are sized for a distribution, not a point estimate — large enough to matter if the thesis is right, small enough to survive the drawdown before it is.
Model releases, capex guidance, power contracts, and export policy are scored against our written forecasts. When the physical picture changes, the book changes.
Illustrative · updated quarterly
Leading-edge logic — advanced node capacity
91% allocated · 2026
High-bandwidth memory — HBM4 supply
84% allocated · 2026
Advanced packaging — CoWoS-class capacity
76% allocated · 2026
Grid interconnection — queued data-centre load
68% of new load queued
Power generation — dispatchable capacity additions
52% of forecast buildout online
04 / Performance
Fund results, net of fees, against a global equity benchmark. Past performance is not indicative of future results.
| Year | Fund | MSCI World | Excess | Net exposure | Max drawdown |
|---|---|---|---|---|---|
| 2025 | +41.2% | +11.2% | +30.0% | 68% | −12.4% |
| 2024 | +36.8% | +9.4% | +27.4% | 74% | −9.8% |
| 2023 | +52.6% | +14.8% | +37.8% | 71% | −11.1% |
| 2022 | −14.3% | −18.1% | +3.8% | 52% | −26.7% |
| 2021 | +29.4% | +18.4% | +11.0% | 77% | −8.2% |
Figures shown are illustrative and provided for presentation purposes only. A concentrated, thesis-driven strategy carries materially higher volatility and drawdown risk than a diversified portfolio. Nothing on this page constitutes an offer to sell or a solicitation of an offer to buy any security.
05 / Insights
The research notes behind the book — bottom-up, dated, and kept up whether the call was right or wrong.
Why high-bandwidth memory allocation, not wafer starts, is setting the pace of the next training-cluster generation — and who captures the margin.
Read the noteA unit-by-unit accounting of queued data-centre load against dispatchable generation additions across three grid operators.
Read the noteMapping the software categories whose unit economics invert once inference cost falls below the labour it replaces.
Read the note06 / Firm
A thesis this concentrated is only investable with the right capital base and the right time horizon. Hyperleap is structured so that neither the firm nor its partners are forced to sell the idea at the wrong moment.
Multi-year lockups by design. We would rather take fewer dollars than take money that leaves at the first drawdown.
A material share of the firm's own capital sits in the same positions, on the same terms, with the same lockup.
Partners receive the full research record — the models, the forecasts, and the calls we got wrong.
Team roster illustrative — replace with the firm's actual partners before launch.
Sets portfolio construction and position sizing. Background spanning semiconductor equity research and macro rates.
Leads the bottom-up supply-chain work — fab starts, HBM allocation, packaging capacity, grid interconnection queues.
Owns risk sizing, hedging, and the short book that funds the long-duration compute and power sleeves.
First point of contact for allocators — due diligence, reporting, and the full research record.
07 / Contact
Hyperleap Capital accepts capital from qualified institutional investors and eligible professional clients. We are happy to share our research and our thesis in full.
Prefer email? Write to hello@hyperleapcapital.com