Working at Vast AI in 2026: 30 People, 20,000 GPUs, and the Cheapest Compute on Earth

Most AI infrastructure companies follow a familiar playbook: raise hundreds of millions, build or lease data centers, hire aggressively, and compete on scale. Vast AI threw that playbook out. Founded in 2018 by brothers Jake and Travis Cannell, Vast AI operates the world's largest decentralized GPU cloud marketplace — connecting over 1,400 GPU providers across 500+ locations with researchers and companies who need cheap compute. They did it with roughly 30 people and $30M in total funding. The company is profitable.

That last detail matters more than any Glassdoor score. In a sector where CoreWeave raised $7.5B and Lambda raised $800M+ to compete, Vast AI's bootstrapped profitability is genuinely unusual. It means the equity employees hold is backed by real revenue, not future fundraising hopes. It means the company doesn't answer to growth-stage VCs pushing for headcount expansion or geographic land-grabs. And it means working there feels fundamentally different from the other GPU cloud companies in our directory.

Vast AI at a Glance

A caveat on the Glassdoor data: a perfect 5.0 from a 30-person company means a tiny sample size. Take it as directional — employees who have reviewed are happy — but don't weight it the same as a 4.2 from a company with 2,000 reviews. The CEO approval and WLB scores are similarly limited but consistently positive.

What Vast AI Actually Does

The simplest analogy is Airbnb for GPUs. Data centers, mining operations, research labs, and individual GPU owners list their spare compute on Vast's marketplace. Researchers and AI companies rent that compute at prices significantly below AWS, GCP, or Azure — often 3–5x cheaper for equivalent hardware. Vast takes a commission on each transaction.

The platform supports everything from RTX 4090s for hobbyist fine-tuning to clusters of A100s and H100s for production training runs. Users search by GPU model, VRAM, PCIe bandwidth, geographic location, and host reliability scores. It's a remarkably transparent marketplace — you can see exactly what hardware you're renting, from whom, and with what track record.

The trade-off is reliability. Because hardware is owned by third parties, uptime and performance can vary. A long training run might get interrupted if a host goes offline. This makes Vast ideal for cost-sensitive, interruptible workloads — research experiments, fine-tuning, inference at scale — but less suited for production-critical training where a 4-hour interruption costs $50,000 in lost progress. For those workloads, companies like Lambda or CoreWeave with their own data centers offer higher reliability at higher prices.

The Culture: Flat, Fast, and Intensely Technical

Working at a 30-person company managing 20,000+ GPUs creates a culture that's impossible to replicate at scale. Every engineer has direct access to Jake Cannell, the CEO and technical co-founder. There are no middle managers, no skip-levels, no quarterly OKR ceremonies. The organizational chart is essentially a spoke-and-wheel with Jake at the center.

This is not "flat" in the way a 500-person company with three management layers calls itself flat. This is genuinely flat — the kind of flat where the person reviewing your PR also decides company strategy, and where your feature ships to 25,000 customers without passing through a product committee.

What employees love

The recurring theme is ownership. At 30 people, there's nowhere to hide — but there's also no ambiguity about your impact. Engineers describe shipping features that directly affect revenue the same week. The tight feedback loop between building something and seeing it used by thousands of customers is something you'd normally only get at a pre-seed startup, but Vast has the revenue and customer base of a much larger company.

What could be better

The cons are predictable for a company this size: process gaps, broad job scopes, and the cognitive overhead of context-switching between multiple domains. The many hats reality means a backend engineer might also be debugging provider-side networking issues or writing customer-facing documentation. If you want clearly scoped sprints and a well-defined career ladder, this isn't it. If you want to learn the entire stack of a real business in two years, it might be exactly right.

Compensation & Equity

Systems and GPU Research Engineer roles at Vast AI pay between $160,000 and $320,000 in total compensation. For a 30-person startup, that range is competitive — the upper end rivals what Modal and mid-stage AI infrastructure companies pay. The equity component is where Vast gets interesting.

Because the company is bootstrapped and profitable, employee equity isn't contingent on closing a Series C or hitting a revenue milestone that unlocks the next tranche. The business generates real cash flow from marketplace commissions. That makes equity grants feel qualitatively different from the lottery-ticket equity at pre-revenue startups. It's not "we might be worth something someday" — it's "we're already making money and the AI compute market is growing 40%+ annually."

The $30M in total funding also means minimal dilution compared to companies that have raised $500M+. Early employees at Vast hold meaningful ownership stakes that haven't been diluted through five rounds of fundraising. For engineers who understand cap tables, this matters enormously. For a deeper look at how startup equity actually works, see our equity guide.

Engineering & Tech Stack

Vast AI's engineering challenges are genuinely unusual. The team builds and maintains a real-time marketplace that matches GPU supply with demand across a heterogeneous network of hardware they don't own or operate. Think of the complexity: different GPU models, different host configurations, different network topologies, different pricing structures, different reliability profiles — all needing to be searchable, rentable, and monitorable in real time.

Tech Stack

The company maintains open-source tools including vast-cli (a Python CLI and SDK for managing GPU resources) and vast-sdk (a Python SDK for programmatic access). They also build base images and serverless workers for their inference platform. The GitHub presence is lean but functional — reflecting the engineering-driven culture of shipping what works rather than polishing what impresses.

Key engineering domains

• Marketplace matching & pricing. Real-time supply/demand optimization across 20,000+ GPUs with variable pricing, reliability scores, and hardware configurations. This is a genuine systems design challenge — think exchange-like matching engines but for heterogeneous compute resources.
• Provider infrastructure. Building the software that runs on provider machines to manage GPU allocation, monitor health, and report metrics back to the marketplace. This requires deep systems-level expertise — CUDA, drivers, networking, containerization.
• Customer-facing platform. The search interface, billing system, API, CLI, and SDK that 25,000+ monthly customers interact with. Despite the small team, the developer experience is remarkably polished — a 4.4 Trustpilot rating from ~200 reviews confirms this.
• Neural network optimization. Vast is also developing software to accelerate training and deployment of complex neural networks on their decentralized infrastructure — essentially building the tooling layer that makes heterogeneous GPU clusters feel like a unified compute fabric.

The Competitive Landscape

Understanding Vast AI's position requires understanding who it's not. The GPU cloud market has stratified into distinct tiers, and Vast occupies a unique niche.

Vast AI's competitive moat is network effects. More providers listing GPUs means lower prices, which attracts more customers, which attracts more providers. At 20,000+ GPUs and 25,000+ monthly customers, Vast has a liquidity advantage that new marketplace entrants can't easily replicate. The 8 years of operations data also feed into reliability scoring that helps customers avoid unreliable hosts — a data asset that compounds over time.

Who Thrives at Vast AI

Based on employee reviews, the company's structure, and the nature of the work, here's who tends to do well at Vast AI:

• Systems engineers who want breadth. If you want to touch CUDA kernels, marketplace infrastructure, customer-facing APIs, and provider-side networking in the same quarter, Vast is one of the few places where that's not just possible but expected. The many-hats culture is genuine.
• People who prefer small teams over large organizations. Thirty people means everyone knows everyone, decisions happen fast, and bureaucracy is near-zero. If you've ever felt suffocated by process at a large company, the relief of working at Vast's scale is immediate.
• Engineers motivated by the AI compute thesis. If you believe — as many do — that GPU compute will be the most valuable commodity of the next decade, building the marketplace where it's traded is a compelling position.
• People comfortable with ambiguity. Processes are still evolving. Roles are loosely defined. Priorities shift based on what the marketplace needs. If you need a clear career ladder and quarterly performance reviews, look at Databricks or Stripe instead.
• Equity-motivated engineers. The bootstrapped, profitable model combined with minimal dilution means early employees hold significant ownership. If you understand that equity in a profitable, growing company is worth more than options in a cash-burning one, the math is compelling.

Vast AI is not a fit for engineers who want to specialize deeply in one domain, who need mentorship from senior leadership layers, or who prioritize brand-name recognition on their resume. The company has low public visibility compared to competitors like CoreWeave or Lambda — you won't get the "oh, cool" reaction at a conference. What you'll get is ownership, speed, and a front-row seat to one of the most important infrastructure markets in tech.

The Bottom Line

Vast AI is one of the most unusual companies in our directory. A 30-person, bootstrapped, profitable team managing more GPUs than most well-funded competitors have employees. The culture is genuinely flat — not corporate-flat, but startup-flat, where every person matters and every engineer ships to production. The compensation is competitive, the equity is meaningful, and the market they're building in is growing explosively.

The risks are real: startup-stage process gaps, broad responsibilities that border on overwhelming, and a marketplace model whose reliability depends on third parties the company doesn't control. But for the right engineer — someone who values ownership over process, breadth over depth, and real equity over prestigious logos — Vast AI is a compelling bet on the future of AI infrastructure.