Master Vast.ai with Woven

Built by Hosts, for Hosts

What You Need to Know

Cloud GPU hosts focused on ROI and monthly KPIs need insights that help them spec and price servers to achieve the highest possible revenue and utilization at the lowest cost. Here is what you need to know:

• What GPU type and how many should I put in my server — and how should it be specified?
• How many minimum GPUs should I set in my offer?
• Is a 1-GPU offer better than a 4-GPU offer? Do larger offers take longer to rent but lock in longer utilization?
• How do I set my initial price for the highest utilization?
• How long should I hold a lower price before dropping it again?
• Is a higher price at lower utilization better than a lower price with higher utilization?

Woven was built to answer these questions. We deliver metric breakdowns by GPU count, price percentiles and distributions for rented offers, hardware-to-price correlations, and deep historical analysis — all in one analytics web application.

We ingest every offer and machine configuration from Vast.ai via data provider 500.farm. We use this to determine GPU cloud rental prices, utilization rates, contract durations, and how supporting hardware impacts pricing. That same data trains the AI behind the Woven Auto Listing Agent.

How Woven Analyzes the Vast.ai GPU Marketplace

Real-Time ETL from 500.farm

Every 2.5 minutes, Woven ingests the complete offers and machines dataset from 500.farm — Vast.ai's primary data provider. Each cycle processes 24,000 offers and 5,000 machines, deserializing raw JSON into our enterprise Java analytics engine built on the Spring framework.

One of the core challenges Woven solves: Vast.ai does not directly expose which offers are rented. We reconstruct the true rental picture by comparing utilization deltas and rentable-state changes between the current and previous sample, then apply Vast's price-lock rules to maintain accurate rented prices for the duration of each rental.

Six Parallel Analyzers, Every Cycle

Once we know what's rented, six analyzers run in parallel to produce the data behind every Woven dashboard:

• GPU Utilization — utilization rates, rented and vacant counts, and 10th/median/90th price percentiles for every GPU type and offer size, including regional breakdowns.
• Machine Utilization — the same metrics aggregated at the machine level, accounting for multi-offer configurations.
• Machine Deltas — tracks every machine from the moment it appears on Vast, recording time-to-verification and reliability progression.
• Price Efficiency — compares your machine's price against all similarly-configured rented peers within 10% of your key hardware metrics.
• Offer Price Correlations — 10th, 50th, and 90th price percentiles across every hardware metric range: CPU cores, RAM, disk space, disk bandwidth, and inet speeds.
• Listing Context — captures your auto-sort position and Vast score every cycle while vacant, and at every price-set event when the Listing Agent is active.

Summary Tables for Instant Historical Access

Querying 24,000 raw offer records per sample across two years of history would be impossibly slow. Instead, each cycle we write compact summary tables that let the app serve 90-day, one-year, and two-year historical views in seconds. Our database holds over 14 TB of offer history. For per-machine views — utilization timelines, price-correlation panels, and auto-sort history — data is indexed by machine ID for fast lookup without scanning the full offer set.

Planning Your Vast.ai GPU Build

Choosing the Right GPU

Choosing a GPU should be based on your budget and business plan's financial model. It should account for GPU cost, supporting hardware, electrical, data, HVAC, and co-location expenses.

The offer analytics dashboard lets you investigate offers on Vast.ai for specific GPU types and counts per offer. Review live and historical utilization, rented and vacant prices, offer flow, verification periods, machine specs, and utilization of rented and vacant servers.

Designing a Server Configuration

The Machine Utilization dashboard lets you investigate utilization and earnings for specific GPUs and quantities in a machine. Review historical rented and vacant prices alongside utilization data on Vast.ai.

Use machine configuration ranges and price correlations to see what the market is paying for the configuration you are planning. This prevents over- or under-specifying your machine so you achieve the highest earnings at the lowest build cost.

Setting the Right Price on Vast.ai

Price Concentrations, Not Just Percentiles

Percentile prices alone are misleading. If the 90th percentile for 4×RTX 4090 offers is 55¢ but the majority of offers cluster at 45¢ with a secondary group at 55¢, you need the price distribution — not just the headline number. Woven's offer analytics shows price concentration charts alongside percentiles so you know exactly where competition sits before you list.

How Vast.ai Auto-Sort Actually Works

When a client opens Vast.ai to rent a GPU, machines are ranked by a hidden score. That score is driven by three factors in descending order of influence: price, reliability, and DL performance. Changing your price by a single cent can move you three spots in auto-sort. The difference between 99.8% and 99.9% reliability carries a massive score multiplier. DL performance is fixed by hardware and cannot be tuned day-to-day. The practical takeaway: price is the only lever you can pull to meaningfully improve your rank — and Woven gives you the market data to pull it correctly.

Finding Your Target Price Range

Woven calculates five suggested price levels for your machine, derived from rented offers within 10% of your CPU cores, RAM, disk speed, disk space, and inet speeds. The goal is to price low enough to enter the top 5 in auto-sort — the positions visible in Vast's default rental view without clients needing to scroll.

Once in the top 5, check the score gap against the machine directly behind you. If that gap exceeds 15%, your price is too low — you can raise it without losing your position, recovering revenue you would otherwise leave behind.

Manually Setting Prices

Use the offer price distribution panel and machine-to-market price efficiencies to see what other hosts are listing at and where you should list. Monitor rented and vacant offer volumes historically to determine whether to adjust price or hold steady. If utilization is low across the board, changing minimum GPU count per offer may be more effective than dropping price.

Woven Listing Agent: Automated Pricing End-to-End

Automatic Auto-Sort Optimization

Auto-sort placement on Vast.ai is key to getting rented quickly. The Woven Listing Agent adjusts your offer price every minute using suggested price levels derived from rented offers on rigs with similar GPU type, count, CPU cores, RAM, disk, and inet specs.

The agent's logic is straightforward: check your current auto-sort position against a configurable target (default: top 5). If you're outside the target, step the price down one level and check again next minute. If you're inside the target, check whether the score gap to the machine directly behind you exceeds 15% — if it does, raise your price to recover margin while holding position.

In a live demonstration, a 2×RTX 5090 rig moved from 29th place to 3rd–5th in Vast.ai's global auto-sort within minutes of enabling automatic mode — with no manual intervention.

Woven's Strategy

Woven's revenue and utilization strategy is focused on optimizing a machine's offer prices to maximize auto-sort position in the Vast Rental UI. Auto-sort is most impacted by price, internet speed, reliability, and DLPerf — in that order. Once online, a machine's DLPerf will not change significantly and reliability should only rise, so price is the only variable you can tune to impact auto-sort placement.

Machines on Vast have a lifecycle that traverses verification state and climbs a reliability scale. These are important because verification states have completely different market prices and utilization. Offers with reliability under 90% do not appear in the default Vast Rental UI view regardless of Vast score. Verification takes approximately 2–4 weeks, and reliability climbs fastest when listed and rented.

Configurable Controls & Alerts

You set the floor: a minimum price the agent will never go below. Configure minimum GPU count, inet up/down pricing, storage pricing, discount rate, and contract end date — rolling 30-day window or a fixed date. For hosts who want visibility without full automation, advisor mode shows recommended price-level changes without applying them.

The agent fires real-time notifications the moment something changes: offer rented, offer vacant, Vast API bandwidth errors, timeout errors, and verification state changes. Receive alerts via email or a Discord webhook.

Deploy the Woven Listing Agent on any machine that has the Vast CLI app installed and your API key configured. We never need or store your Vast API key. Contact us via Discord or email for more info.

Woven Pricing

Listing Agent Client

The Listing Agent Client requires an Inconel subscription.

The Inconel subscription includes Vast RTL & Events:

• Capture, analyze, and persist vast show machines into Listings.
• Receive Vast events via Listings, Discord, and Email.

Listing Agent — Price Automation ($USD/GPU/month):

GPU listing automation licenses can be purchased in-app once you are an Inconel Analytics subscriber.

• Tier 1 — GPUs avg $/hr < $0.29 = $1.99/GPU/month
• Tier 2 — GPUs avg $/hr $0.29–$0.99 = $3.99/GPU/month
• Tier 3 — GPUs avg $/hr > $0.99 = $9.99/GPU/month

Example GPU Servers:

• Four × RTX 3080 = 4 × $1.99 (Tier 1) = $7.96/month
• Eight × RTX 4090 = 8 × $3.99 (Tier 2) = $31.92/month
• Four × H100 = 4 × $9.99 (Tier 3) = $39.96/month

Contact us via Discord or email for more info.