Customer Segments

Segmenting Vast's demand

The honest segmentation isn't by company size or industry — it's by workload tolerance for variability and price sensitivity. Cross those two and you get:

Vast wins the high-tolerance/high-sensitivity quadrant outright. Vast competes for some of the low-sensitivity/high-tolerance quadrant. Vast loses the low-tolerance/low-sensitivity quadrant to enterprise neoclouds. The high-sensitivity/low-tolerance corner is structurally hard for anyone.

The core segments Vast wins

Academic researchers

Grad students, postdocs, faculty. They have small grants and big workloads. Tolerance for variability is high (they understand how compute behaves). Price sensitivity is high (the grant dollar matters).

• Why Vast: 60-80% off hyperscaler list. Fits their grant constraints.
• Why not Vast: Rare — typically only when they're at an institution with onerous procurement.
• Switching risk: Free academic tiers at hyperscalers (AWS Academic Hub etc.) compete, but Vast's price advantage holds in most cases.

Indie AI builders

Solo founders or small teams building AI products on bootstrapped or small-seed budgets.

• Why Vast: Burn rate is the constraint. Cheap GPU compute extends runway.
• Why not Vast: As they raise and scale, they hit the segment ceiling (compliance, multi-node, reliability).
• Switching risk: Vast loses them when they scale. The retention curve at this segment is the leakiest bucket in Vast's customer base.

ML hobbyists and enthusiasts

Self-funded people exploring ML for fun or learning.

• Why Vast: Cheapest way to access serious GPU. Doesn't require a billing relationship with a hyperscaler.
• Switching risk: Low. This segment doesn't have alternatives at Vast's prices.

ML consultancies / boutique service shops

Firms doing project ML work for clients.

• Why Vast: Marginal cost of compute is critical to their gross margin.
• Switching risk: Moderate. As they grow, they may negotiate enterprise neocloud reservations.

Contested segments

Series-A/B AI startups

Funded enough to afford enterprise neoclouds but still cost-conscious.

• Vast use: Often as a secondary cloud for research, ablations, non-production workloads.
• Primary cloud: Usually a hyperscaler or an enterprise neocloud for production.
• The opportunity: Vast could capture more of this segment's research/dev spend with better orchestration tooling, persistent storage, and clearer support tiers.

Research-focused enterprises

Big-company research labs that want to experiment quickly without going through internal IT for new infrastructure.

• Vast use: Some adoption for experimental workloads; rarely for production.
• The risk: Enterprise IT often blocks Vast on compliance grounds. The lab uses Vast on personal credit cards as shadow IT, which has its own ceiling.

Inference service providers

Small companies offering inference APIs. They could run on Together.AI / Fireworks for a managed experience or on Vast for raw compute and roll their own.

• Vast use: Some operators run inference on Vast. The economics work when they have enough scale to absorb operational complexity.
• The trend: Managed inference is winning at most scales; Vast loses ground here as Together / Fireworks / others mature.

The segment ceiling

The honest statement of Vast's strategic ceiling:

Marketplace economics work for workloads where variability is tolerable. They structurally don't work for workloads where it isn't.

Specific blocks Vast can't reasonably penetrate:

• Regulated industries (HIPAA, PCI, FedRAMP). The supply-side population can't be uniformly certified.
• Production inference at scale with five-nines SLAs. Interruption profile is wrong.
• Frontier model training requiring multi-thousand-GPU clusters with InfiniBand. Vast doesn't have the fabric.
• Long-term reserved-capacity deals (multi-year). Marketplace structure doesn't naturally support this.
• Customers requiring dedicated account managers, custom contracts, or service-level guarantees.

These segments collectively represent the bulk of enterprise spend on AI compute. Vast can't take them with the current model. That's not a defect of the company; it's a feature of the strategy. Marketplaces aren't designed for these segments.

The enterprise question

Could Vast launch an enterprise tier and capture some of the segments above? In principle yes; in practice the obstacles are real:

• Going up-market dilutes brand and culture. Vast's appeal is partly that it's the indie / hacker option. An enterprise tier risks alienating the core base.
• Sales motion is expensive. Enterprise sales requires AEs, SEs, customer success — a sales org. Vast's lean culture is anti-this.
• Compliance is expensive. SOC 2, HIPAA, FedRAMP each take many quarters and ongoing operational cost. Marketplace heterogeneity makes this harder than at a unified cloud.
• Pricing structure clashes. Enterprises want predictable monthly bills and committed pricing. Marketplaces don't naturally provide either.

The realistic options for Vast at the enterprise boundary:

1. Stay focused on the segments they win. Don't chase enterprise. Maximize share and depth in indie / research / hobbyist.
2. Launch a separate brand or product for enterprise. Curated supply, vetted hosts, simulated SLAs. Build it as a separate vertical with separate ops.
3. Partner with enterprise platforms. Become the back-end for a managed offering somewhere else.

Public signals so far suggest Vast is mostly choosing option 1 — staying in the lane where the marketplace is structurally advantaged. That's defensible but caps long-term TAM.

Secondary segments

Beyond ML / AI, Vast has secondary segments:

• Rendering (3D, video). GPU rendering benefits from marketplace pricing. Niche but real.
• Scientific computing (molecular dynamics, climate, simulation). Academic users overlap with this category.
• Crypto mining. Vastly smaller than it once was. Some pockets remain for GPU-mineable altcoins.
• Distributed compute experiments (BOINC, Folding@home offshoots, etc.). Niche.

These secondary segments don't change Vast's strategic positioning materially but add resilience — if AI demand softens, other use cases provide a partial floor.

Geographic segments

Vast's geographic distribution differs from the enterprise neoclouds. North America and Europe are big but not dominant; significant volume comes from East Asia, Eastern Europe, Latin America. The factors:

• Self-service onboarding (no region-specific sales coverage required).
• Crypto payment options (some markets are friction-laden for SaaS billing).
• Price sensitivity in markets where hyperscaler GPU rates are unaffordable.

This geographic spread is a real strategic asset. It diversifies Vast's revenue base and makes the company less exposed to North American AI funding cycles. Few competitors have this profile.

Takeaway

Vast wins the segments where price beats consistency. It loses where consistency wins. The boundary is sharp and the company seems to have chosen to live on its side of it rather than fight battles it would have to fundamentally re-architect to win.

The next chapter places Vast in the competitive landscape — who they're competing against and how the strategic picture looks for the next 2-3 years.