Hyperliquid Market Data Moved to Dedicated Fiber and One Feed Is Pre-Trade

DoubleZero switched on a set of Hyperliquid market data feeds over dedicated fiber on September 24, and the product worth reading twice is not the order book. Four of the five feeds carry sequenced market data. The fifth carries order intents pulled from mempool transactions, meaning orders that have been submitted but not yet sequenced into a block. Hyperliquid market data just split into two tiers, and only one of those tiers is written in the past tense.

What DoubleZero Edge Ships for Hyperliquid

According to the September 24 announcement on PR Newswire, DoubleZero Edge now distributes five products for the venue: four machine-readable order book feeds covering Hyperliquid native perpetuals and trade[XYZ] HIP-3 real-world-asset markets including oil, gold and silver, plus the order intents feed built from mempool transactions. Each subscription includes two IPs per region, with infrastructure positioned in Tokyo.

The pitch is delivery rather than content. Instead of assembling venue state from public endpoints, a subscribing firm receives a sequenced stream over dedicated fiber. Hyperion DeFi chief executive Hyunsu Jung described the model to Cointelegraph as publish once, distribute simultaneously over dedicated fiber, which is how a traditional futures exchange has always run its colocated data feed.

Hyperliquid is the third venue on Edge after Solana and the prediction market Kalshi, per the same Cointelegraph report. Validator operators Hyperion DeFi, MAVAN and Kinetiq worked on the build. The DoubleZero release calls it the fastest commercially available Hyperliquid feed on the market, a claim Hyperion DeFi repeated in its own September 24 statement on GlobeNewswire.

For scale, DoubleZero puts Hyperliquid second-quarter 2026 volume at $662 billion. That is enough flow to make a few milliseconds worth paying a cross-connect bill for.

Sequencing Is the Product, Not Speed Alone

A public websocket gives you messages. A sequenced feed gives you messages in a known, gap-checked order with a single publication point. Those are different guarantees, and the second one is what quantitative desks actually buy.

When you reconstruct book state from a public endpoint, you are the one stitching updates together, and every reconnect, dropped frame or out-of-order snapshot becomes silent error inside your own data. Two firms reading the same public feed can hold two different pictures of the same instant. A single published sequence removes that class of disagreement.

This is why the fiber build matters even to traders who will never subscribe. The reference version of Hyperliquid market data now lives somewhere other than the documented public API, and the public path becomes the derived copy.

The Order Intents Feed Is Pre-Trade Information

Every orderflow metric most traders use is post-trade. Cumulative volume delta is built from fills. Order book imbalance is built from resting size that has already been posted. Both describe what happened.

An intents feed built from mempool transactions describes what is about to happen. A subscriber observes submitted orders before they are sequenced, which is a short window given Hyperliquid block cadence, but a window that exists and that nobody reading a fills-based feed has access to. That is a structurally different dataset, not a faster copy of the same one.

Sit with the consequence. The gap between a fiber subscriber and an API user is no longer only latency. It is the difference between reading fills and reading the queue that produces them.

Public API Depth Was Already Thinning

DoubleZero own framing supplies the reason the product exists. Its release states that public API changes have reduced the update frequency and depth available over those endpoints, pushing traders toward node-based access.

That should land hard for anyone computing Hyperliquid perp analytics from the documented websocket. If depth snapshots arrive less often, or arrive truncated, then book-derived measurements drift. Order book imbalance computed on a throttled five-level view is not the same quantity as OBI computed on full depth arriving continuously, even though your code prints both to the same column.

The practical read: a thinner public feed does not only slow you down, it quietly changes the number your indicator produces. Anyone backtesting a depth-ratio or OBI rule across a period when endpoint behavior changed is comparing two different instruments and calling the difference alpha.

Which Orderflow Metrics Survive a Tiered Feed

When a venue fastest participants run on a private distribution path, the visible tape becomes a lagging summary of their activity rather than a live record of it. Passive size that disappears before your snapshot refreshes reads as absorption that was never there. Aggressive prints that arrive batched read as one large sweep instead of the sequence that actually traded.

None of that makes orderflow analysis worthless. It makes the choice of metric decisive. Measurements built on completed trades hold up better than measurements built on instantaneous book state, because a fill is a fact that cannot be pulled after the event. CVD, trade-size distribution and flow toxicity estimates such as VPIN all read realized aggression rather than quoted intention.

Position structure is the other class that survives. A liquidation price on Hyperliquid is a function of margin and entry, it is published on chain, and it does not degrade when a websocket throttles. Neither does whale wallet attribution. Those are records, not quotes.

Trading Hyperliquid When Speed Is Not Your Edge

Nobody reading this is about to buy a Tokyo cross-connect. The useful response is to stop competing on the axis where fiber wins and start weighting the data that a subscriber speed advantage cannot erase: realized flow, position structure, and the exact prices where forced sellers sit.

Buildix computes CVD, VPIN, order book imbalance and whale attribution across Hyperliquid perps including HIP-3 markets, and serves the same series over its API from buildix.trade/screener. The goal is not to be first into the queue. It is to be correct about who is positioned where when the queue clears.

The fiber build is itself a signal. Dedicated infrastructure gets constructed for venues that professional firms intend to trade at size for years, and that is a more durable statement about Hyperliquid than any single volume record it prints this quarter.