The thesis is simple to state: AI demand keeps growing, and the companies that build the compute capture the upside. The harder part is expressing it well — because "AI" is not one stock, and how you structure the trade decides how much company-specific risk you carry. This post walks through the ways to get AI exposure on Legend, from a single chip leader to a diversified semis basket to a relative-value spread. It is one applied example of the thesis-trade framework.
The Thesis
AI infrastructure spending — chips, data centers, and the cloud capacity to run models — has been the dominant growth story in equities. If you believe that capex continues, the cleanest beneficiaries are the semiconductor designers and manufacturers, the companies renting out GPU compute, and the labs building the models themselves. Legend gives you equity perps on all three layers in one self-custody account, and because stock perps trade around the clock, you are not boxed in by the 9:30-to-4 session — here is why stock perps trade 24/7.
How to Express It on Legend
Go long a chip leader (outright)
The most direct expression. If you have a single high-conviction name, go long it:
- NVDA — the GPU bellwether
- AMD — the closest competitor in AI accelerators
- ARM — chip architecture and licensing
- AVGO — networking and custom silicon (Broadcom)
- ASML — the lithography monopoly upstream of every chip
- TSM — the foundry that actually fabricates them
- MRVL — data-center interconnect and custom silicon (Marvell)
One leg, full exposure to that company. Highest reward if you are right, highest concentration risk if that one name stumbles.
Go long the semis index (diversified)
If you like the theme but do not want to pick the winner, go long SMH, the semiconductor index perp. One position spreads your exposure across the sector, so a single bad earnings print hurts less. This is the basket expression in a single instrument.
Add AI-infrastructure and private-lab names
Beyond the chip designers, Legend lists names further along the AI supply chain:
- NBIS (Nebius) and CRWV (CoreWeave) — AI-infrastructure and GPU-cloud providers
- ZHIPU and MINIMAX — perps tracking private AI labs, exposure you cannot easily get in a normal brokerage
These let you express a view on the picks-and-shovels layer or on the model labs directly.
Relative value: long the leader, short the laggard
If your real view is who wins AI rather than does tech go up, express it as a relative-value spread: go long the name you think outperforms and short the one you think lags — for example long NVDA and short a competitor you expect to fall behind, sized so the two legs are roughly notional-balanced. If the whole sector rallies or sells off, the legs largely offset, and what is left is your call on relative performance. This strips out a lot of broad-market direction.
Start trading on Legend to put any of these on as real positions.
Sizing and Risk
- Decide your max loss before you size. AI names are volatile; size for the drawdown, not the dream. See how to manage risk.
- Use leverage deliberately. NVDA perps go up to 20x on Legend; available leverage is a ceiling, not a recommendation. Some equity perps are isolated-margin-only — use that to cap the loss on each leg.
- Diversify if you cannot pick. A single SMH long carries less single-name risk than concentrating in one ticker.
- Balance relative-value legs. Unequal notionals turn a clean spread trade into an accidental directional bet.
What Could Go Wrong
AI is a crowded, high-multiple theme, and that cuts both ways. The main risks:
- Concentration. A single-name long lives and dies on that company's execution and guidance.
- Earnings gaps. Equity perps can jump sharply on earnings; a leveraged position can be liquidated on a single report.
- Sentiment reversals. High-multiple names fall hardest when the market de-risks, even if the underlying business is fine.
- Wrong leg in a spread. A relative-value trade loses if the laggard you shorted outperforms the leader you bought.
This article is educational and is not financial advice.