Is Hyperliquid Market Making Still Profitable in 2026?

Short answer: yes, but the easy money is gone. Passive quoting on the majors like BTC and ETH now gets your resting orders stepped in front of by dozens of colocated bots before you fill, and the 2026 fee-tier reshuffle trimmed the maker rebate that used to paper over sloppy inventory management. What's left is a real business with thinner margins, where the winners are the ones who model inventory cost correctly and quote where competition is weaker.

Let me show you the actual math, because "is Hyperliquid market making profitable" only has a useful answer once you put numbers on the spread you capture and the inventory risk you carry.

Where the edge comes from

A market maker on Hyperliquid earns money three ways, and it helps to keep them separate in your head:

• Spread capture — you buy at your bid, sell at your ask, pocket the difference. This is the headline number and the one people overestimate.
• Maker rebates / fee tiers — Hyperliquid's fee schedule rewards volume. At the top tiers, taker fees drop and maker economics improve, but the 2026 changes flattened the curve for mid-volume makers specifically.
• Funding and directional drift — usually a cost, occasionally a subsidy. If you're systematically long the perp while quoting, funding either pays you or bleeds you depending on the sign.

Spread capture is gross revenue. Inventory cost and adverse selection are what eat it. Most people who conclude market making "doesn't work anymore" are only measuring the first one.

The spread-capture math, done honestly

Take HYPE-PERP as an example because it's liquid but not as brutally contested as BTC. Say the top-of-book spread is 4 bps and you quote one tick inside on both sides, so you're offering a 3 bps round-trip if you get both legs.

Assume:

• You fill $2M notional per side per day (a modest, realistic target for a single mid-tier maker).
• Your effective captured spread is 1.2 bps, not 3. You never fill both legs symmetrically; adverse selection means the side that fills is disproportionately the wrong one.
• Maker fee at your tier is roughly flat to slightly positive after the 2026 change — call it a net 0.1 bps rebate.

Gross on $2M/side/day:

daily_volume   = 2_000_000 * 2      # both sides = $4M
captured_spread = 0.00012           # 1.2 bps effective
rebate         = 0.00001            # 0.1 bps net maker credit

gross_spread   = 2_000_000 * captured_spread   # $240/day
rebate_pnl     = daily_volume * rebate         # $40/day
gross_pnl      = gross_spread + rebate_pnl      # $280/day

That's about $280/day gross, or roughly $100k/year gross on that flow. Now the deductions.

Inventory cost is the line item people skip

When you fill, you hold a position until it mean-reverts or you flatten. That inventory has a carrying cost with two parts: the variance risk (price moves against you before you offload) and the funding you pay or receive while holding.

A quick way to bound the variance cost: if your average inventory sits at $150k notional with a holding half-life of ~2 minutes, and HYPE's 1-minute vol is ~15 bps, your inventory is exposed to a 1-sigma move of roughly $225 per turn of the book. Over a day with dozens of inventory cycles, the realized inventory PnL is a random walk that averages near zero only if your skew logic is honest. Get the skew wrong — quote symmetrically while trending — and this line goes deeply negative fast.

Concretely, on a trending day I'd expect inventory to shave 30–50% off gross for a naively-skewed book. So that $280/day becomes $150–190/day net on a good day, and negative on a day you get run over. This is exactly why the inventory model, not the quoting loop, is where the engineering effort belongs. If you're building the position-management layer from scratch, the tradeoffs in the Python vs Rust guide for the Hyperliquid API matter more than they look — order/cancel latency directly caps how tight you can quote before adverse selection eats you.

What changed in 2026, specifically

Two things moved the profitability line:

1. Fee-tier compression. The old schedule gave a meaningful maker rebate at moderate volume. The 2026 revision pushed the good rebates further up the volume ladder, so a maker doing $50–150M/month lost a chunk of the subsidy that used to cover their inventory misses. If your strategy was only profitable because of the rebate, it's probably underwater now.

2. Maker density. More colocated bots quoting the majors means tighter spreads and worse queue position. On BTC-PERP you are competing against participants who cancel-replace in single-digit milliseconds. Your resting order is adversely selected — it fills when informed flow is picking you off, and sits idle when it isn't.

The takeaway isn't "quit." It's "stop quoting the crowded book." The alts, the newer listings, and the perps with lumpier flow still have spreads wide enough to make the inventory math work.

Where it's still genuinely profitable

• Second-tier perps and new listings. Wider spreads, fewer competitors, more inventory risk — but that's a risk you can price. This is where a purpose-built Hyperliquid market-making bot with per-asset skew and inventory caps earns its keep.
• Funding-aware quoting. If you're going to hold inventory anyway, quote with a skew that leans into favorable funding. When perp funding is paying you to be long, widen your ask and tighten your bid. The mechanics overlap heavily with a standalone funding-rate arbitrage strategy, and running both off one position book is more capital-efficient than treating them separately.
• HLP-adjacent strategies. Providing liquidity through the vault has a different risk profile than quoting directly; the HLP vault strategy breakdown is worth reading before you assume direct quoting is the only path.

A rule of thumb for go/no-go

Before committing capital to an asset, I check that:

edge_check = captured_spread + net_rebate - expected_inventory_cost
# quote only if edge_check > 0 across your realistic fill distribution,
# not just the happy path where both legs fill.

If that number isn't comfortably positive after you model adverse selection — not the quoted spread, the captured one — don't quote that book. Most of the "market making is dead" posts come from people who never subtracted the inventory term.

The honest bottom line

Profitable, yes — for a maker with real inventory management, per-asset quoting logic, and the discipline to skip the contested majors. A single sloppy bot spraying symmetric quotes on BTC will lose money after the 2026 fee changes, and it'll feel like the whole strategy is broken when really it's just mis-specified. The infrastructure bar has risen: you want tight cancel-replace latency, a live inventory-risk view, and funding baked into your skew. If you're pairing quoting with directional signals, the same position layer feeds a Hyperliquid perps bot cleanly, and a proper trading dashboard is what turns "I think we're profitable" into a number you can trust.

If you want that inventory-and-skew layer built correctly the first time, our Hyperliquid market-maker development is where the real edge gets engineered.