The Supply Side

Provider archetypes

Vast's supply is heterogeneous. Five archetypes account for most of it:

1. The hobbyist

Single individual with one or a few GPU machines at home. Often a gamer or hobbyist ML enthusiast. Lists when they're not using the machine. Provides RTX 30/40/50-series consumer cards.

• Scale: 1-4 GPUs typically.
• Reliability: Variable; depends on residential power, internet, family use of the machine.
• Economics: Pays back the GPU purchase if utilization is decent. Hobbyist isn't doing it as a primary income source.

2. The small-scale fleet operator

Solo entrepreneur or small team running 10-100 GPUs in a basement, a small colocation rack, or a converted warehouse. Often a former or current crypto-mining operator who retooled for AI compute.

• Scale: 10-100 GPUs typically.
• Reliability: Better than hobbyist (often has UPS, redundant internet) but not datacenter-grade.
• Economics: Primary income for the operator. Optimizes hard for utilization.

3. The professional datacenter operator

Small or mid-sized datacenter that has built capacity for compute rental. Has proper power, cooling, redundant networking. Lists on Vast as one of several sales channels.

• Scale: 100-1000+ GPUs.
• Reliability: Datacenter-grade or close.
• Economics: Vast is one channel; they also might sell directly to enterprises or through other markets.

4. The research lab

University or corporate research lab with idle GPU capacity. Lists to recoup costs. Less common than the others; institutional friction limits this archetype.

• Scale: 4-50 GPUs typically.
• Reliability: High when up; downtime for maintenance.
• Economics: Subsidizes the lab's own GPU costs.

5. The repurposed crypto operator

Specific subset of the fleet operator archetype, but worth calling out separately because of how influential it's been. After the Ethereum proof-of-stake transition in 2022, large fleets of GPU mining hardware became economically marginal for crypto and pivoted to AI rental. Vast was a major beneficiary — those fleets needed a sales channel, and Vast was the obvious one.

Why they list

From the provider's perspective, why pick Vast over alternatives like running their own sales, listing on competitors (RunPod, TensorDock), or selling capacity to a single enterprise customer?

• Liquidity. Vast has the biggest demand-side user base of any GPU marketplace. A provider listing on Vast will see utilization faster than on smaller competitors.
• Self-service. No sales overhead. List the machine; it gets rented. No contracts to negotiate; no customer success required.
• Diversified renter base. A given GPU machine on Vast might serve dozens of different customers in a month. No customer-concentration risk.
• Payouts are reliable. Vast pays providers promptly through standard channels.

The tradeoff: the provider doesn't own the customer relationship. They can't upsell, can't capture lifetime value, can't build their own brand. Vast does that.

Onboarding

The provider onboarding flow is light:

1. Create a Vast.AI host account.
2. Install the Vast host software on the machine. Linux, typically Ubuntu. Docker required.
3. The software registers the machine — detects GPUs, runs initial benchmarks, reports back to Vast.
4. Set pricing — either accept Vast's pricing guidance or set custom rates.
5. Decide which payment methods to accept; configure payout details.
6. The machine appears in the marketplace.

The whole process is hours, not weeks. There's no application, no interview, no contract negotiation. A teenager with a gaming PC can list it as easily as a small datacenter operator with a hundred-machine fleet.

This openness is the supply moat. Competitors that gate providers — vetting, contracts, minimum-fleet sizes — sacrifice supply growth for quality. Vast bet on openness and lets the ranking system handle quality.

Ranking & reputation

Vast's ranking system is what makes the open-supply model work. Without it, low-quality providers would crowd out good ones and the marketplace would degrade.

Several signals feed the ranking:

• DLPerf benchmark. Vast runs a deep-learning benchmark on each provider's hardware periodically. The score is a normalized performance metric — same model card from two different providers can have different DLPerf scores due to thermal throttling, PCIe configuration, software stack, etc. Surfaces in search.
• Reliability score. Tracks uptime; how often the machine goes offline; how often it crashes during a rental.
• Bandwidth measurements. Internet upload/download benchmarks. Critical for clients who need to move datasets.
• Customer ratings. Renters can rate providers. Aggregated into a reputation score.
• Region and latency. Surfaced as filters but also factors into client search.

The ranking system is the closest thing to a sales-and-quality apparatus that Vast has. It's algorithmic, not human, and it does most of the work that an enterprise cloud's account managers do — match a buyer with appropriate supply.

Provider economics

What does a provider actually earn? The math, roughly:

• Gross rental revenue: Hourly rate × utilization × hours per month.
• Vast's take: ~15-30% of the gross.
• Provider's net: The remainder.
• Provider's costs: Power (often the biggest), bandwidth, hardware amortization, space, cooling, labor for maintenance.

Worked example for an H100 in a small fleet operator's setting (illustrative, not Vast-published):

• Hourly rate on Vast: $2-3/hour (varies widely with market conditions).
• Utilization: 60-80% if the listing is competitive.
• Vast's take at 20%: providers see roughly $1.6-2.4/hour net.
• Operating cost: power at ~$0.10/kWh × ~0.7 kW = $0.07/hour. Bandwidth, space, cooling add maybe another $0.10/hour.
• Gross margin: roughly $1.4-2.2/hour.
• Capital recovery: An H100 cost $25-30k new. At those margins, full payback is 12-18 months at high utilization.

That's a reasonable rate of return for the small operator. It's much worse than the cap-ex math at CoreWeave (who has multi-year reserved deals with enterprises at higher revenue per GPU-hour) but the small operator doesn't have CoreWeave's sales motion.

For an enterprise neocloud operator listing on Vast, the economics don't make sense — they can charge more by selling directly. So Vast's supply is structurally biased toward small operators and hobbyists.

Payouts

Vast pays providers on a regular schedule. Crypto payouts have historically been a notable channel — Vast supports payment in crypto for both clients and providers, which is unusual for the broader cloud industry. The reasoning: many of Vast's small operators came from crypto mining and prefer crypto rails; many of Vast's international users found USD-denominated SaaS payment friction-laden.

This is one of the small ways Vast's "indie-first" culture shows through: features that wouldn't make sense at AWS but make perfect sense in Vast's customer base.

Strategic tradeoffs for providers

From the provider perspective, the choice between Vast and the alternatives is real. The honest comparison:

• Vast vs RunPod (marketplace side): Vast generally has higher liquidity; RunPod has a slightly more vetted supply pool. Some providers list on both.
• Vast vs TensorDock: Similar tradeoff. TensorDock targets slightly more reliability; Vast targets scale.
• Vast vs direct sales: Direct sales can get higher margins per GPU-hour but require sales effort and customer concentration risk. Smaller providers can't realistically do direct sales.
• Vast vs being acquired into a larger neocloud: Sometimes happens — a small operator gets bought into a bigger one or becomes a contractor to one. The exit math depends on the timing of the GPU generation cycle.

Takeaway

The supply side of Vast is a long-tail distribution of small operators with a heavy skew toward post-crypto and hobbyist providers. The flexibility and self-service onboarding are the structural advantages; the variable quality is the structural challenge.

The next chapter does the same exercise on the demand side: who actually rents GPUs on Vast, and what for.