1. What problem does it solve?
An AI coding agent needs a computer to work on — a real one, not a chat window. Give Claude Code or Codex a task and it wants files that persist, Docker, a browser, a terminal, and hours of uninterrupted runtime. Your laptop can't run fifty agents in parallel. The existing options were the wrong shape: short-lived microVMs built for quick function calls, or enterprise sandbox platforms priced for companies, not developers.
boat gives each agent a full Ubuntu virtual machine in the cloud — SSH, Docker, systemd, a 60fps virtual desktop, persistent files — that you can fork from a snapshot in seconds and run by the thousands. It feels like a true dev machine, except it's rented by the second and stops billing the moment you stop it.
The problem is old (developers have rented servers forever). The new part is the customer: not a human developer, but their agents — and the volume. One developer now runs ten, fifty, a hundred agents at once. The machine is the same; the buyer multiplied.
2. Who pays, how much, and in what scenario?
The buyer: a developer running coding agents (Claude Code, Codex, OpenCode) in parallel, or a company building an agent product that needs one sandbox per user. The customer is technical, price-sensitive, and allergic to sales calls — which is why boat sells self-serve, with no sales team at all.
| Size | vCPU | RAM | Disk | $/hour | $/month |
|---|---|---|---|---|---|
| small | 2 | 4 GB | 51 GB | $0.018 | $13 |
| default | 4 | 8 GB | 51 GB | $0.036 | $26 |
| large | 8 | 16 GB | 111 GB | $0.072 | $53 |
| xlarge | 16 | 32 GB | 251 GB | $0.200 | $146 |
Plans start at $20 a month after a 7-day free trial. The $20 plan is $20 of VM time for that month — it does not carry over. VM time bills per second, only while the machine runs. Overages come out of a prepaid wallet ($0.036/h for the default size), saved snapshots cost $0.056 a day each, and plans allow 100 to 2,000 machines at once. There is also an organization tier: in a support email the founder sent in October, he quoted a $2,000-a-month organization plan, each seat adding 2,400 included starts a day.
The scenario that makes someone pay: the first night you want ten agents working while you sleep. A subscription alone doesn't cover a heavy user — and that's the point. The founder says subscriptions are "just a fraction" of total revenue, because heavy users pay for extra machine time on top of their plan. The subscription is the floor; metered usage is the ceiling. The more agents a customer runs, the more boat earns — no renegotiation, no upgrade call.
Sanity check (our estimate, not a reported figure): $1M ARR is roughly $85,000 a month. At the default machine's $0.036 an hour, that is about 2.4 million VM-hours a month — the equivalent of roughly 3,300 machines running around the clock. That is the scale of compute boat is actually selling, and it explains why the business is a fleet operation, not just a website.
3. How did it find these people?
Almost entirely through the founder's own posts on X. From the founder, on October 27: "no paid ads, no events, no big contracts, no sales motion".
Build in public, with receipts. Founder Anicet Nougaret posts the numbers — the pivots, the flat months, the vertical ones. The November 7 post announcing $1M ARR came with a dashboard screenshot showing $1.02M, and a breakdown of the whole 22-month journey. For a developer audience that distrusts marketing, a revenue chart is the most credible ad there is.
Win the comparison. boat publishes head-to-head comparison pages against E2B, Daytona, Modal, Vercel Sandbox and other sandbox providers — and a pricing-comparison site, Battleships, that ranks providers on the same workload. On price, boat usually wins, because it usually is the cheapest. A developer googling "cheapest sandbox for agents" lands on a page boat wrote.
Let the product argue. The homepage leads with a third-party benchmark (StarSling's V8 Web Tooling Benchmark, October 3, 2026): boat at 28.06 runs per second, second of 15 providers — ahead of Daytona, Modal, Vercel Sandbox and E2B. Speed is the other half of the pitch, and it's measured by someone else.
YC helped too: boat went through Y Combinator's Fall 2026 batch, which put it in front of exactly the AI builders who became its first customers. But the batch was the introduction, not the engine — the engine is the founder posting, and the numbers doing the talking.
4. Why does it work?
boat's wedge is not the virtual machine — VMs are a commodity, and every cloud sells them. The wedge is a full machine, at a commodity price, at the exact moment agent usage exploded: a real Ubuntu box with Docker and a kernel, not a stripped-down microVM, for $0.018 an hour — roughly 13x cheaper per hour than Modal's equivalent rate in the same benchmark run — billed by the second, with no sales process between the developer and the machine. The incumbents were built for a world where a human runs one sandbox at a time; boat was built for a world where one human runs a hundred agents. And the product that finally worked was the infrastructure under an earlier one: the team spent 16 months and five pivots selling products on top of their own cloud VM layer, then realized the layer itself was the business — the founder says the ARR of the product they sold from February to May is now what they add every two hours selling the same VM infrastructure.
5. Could you build it?
The product, yes. The business, not by copying the product.
Replicable: the shape — a sandbox API that spawns Linux VMs, bills per second, and forks from snapshots — can be assembled from existing virtualization and billing pieces by a small team. The comparison-page and build-in-public playbook is open to anyone.
Not replicable by cloning: the fleet economics. boat runs its own servers (AMD Ryzen 9 9950X hosts) in EU data centers, and its price is its moat. Undercutting $0.018 an hour requires the same scale, the same utilization, and the same tolerance for thin margins — you can't rent someone else's cloud and resell it cheaper than boat sells its own. And the trust is earned the slow way: 22 months, six pivots, and a public revenue chart.
The opening it leaves: boat runs only in the EU (Germany, Finland, France). A US or Asia fleet, GPU-equipped sandboxes, or sandboxes pre-loaded for one specific agent are all open ground. So is the segment boat doesn't serve: the non-developer who wants agents but will never touch a CLI. And the founder still answers support emails himself — service is a strength today and a bottleneck tomorrow.
The numbers
| Revenue | $1M ARR (dashboard screenshot shows $1.02M), founder-reported Nov 7, 2026; ~$85,000/mo (our math) |
|---|---|
| Revenue growth | $0→$10K in 16 months (5 pivots); $10K→$100K in 4 months (1 pivot); $100K→$200K in 1 month; $200K→$1M in 1 month; adding $40–50K of ARR a day at announcement |
| Total revenue | Higher than $1M ARR — founder says subscriptions are "just a fraction" and the full run rate is "way more"; no total disclosed |
| Team | 4 people (founder, Oct 2026; YC's directory lists 3) · founded 2025 in Paris · YC Fall 2026 batch |
| Founders | Anicet Nougaret (CEO) and Kirill Makarov (CTO); met five years earlier, failed a first startup together, came to sandboxes as frustrated users |
| Pricing | From $20/mo after a 7-day free trial; $0.018–$0.20/hour by size, per-second billing, wallet overage; $2,000/mo organization tier |
| Product | Full Ubuntu VMs (SSH, Docker, browser, 60fps desktop), snapshot forking, 100–2,000 machines at once, EU regions only |
What we don't know
- Total revenue run rate — subscriptions are only a fraction, and the founder gave no total, so we don't estimate one.
- Paying customer count — not disclosed. (Founderland cites 350+ paying users from mid-October; we could not confirm that figure at its source.)
- Gross margin — the fleet is boat's own hardware, and what share of $0.036 an hour is margin is unknowable from outside.
- Churn, and how much of the $1M ARR sits with a few heavy users versus a broad base of $20-a-month customers.
- Whether adding $40–50K of ARR a day is a durable rate or a YC-batch spike — the November 7 post was six days ago.
Sources
The AI Receipts, "How Much Does boat.dev Make? $1M ARR After 6 Pivots (Self-Reported, 2026)" — theaireceipts.com
boat.dev homepage — boat.dev (product, pricing table, benchmark; checked Nov 13, 2026)
YC Fall 2026 batch listing (Boat, formerly Ascii) — techstartupslist.com
Sparse Media, Instagram case study on Boat ($1M ARR, Nov 7, 2026) — instagram.com
Threads post sharing the founder's Oct 31, 2026 support email (organization plan pricing) — threads.com
MilestoneWins listing: milestonewins.com/launches/boat/ ($1M ARR, founders Anicet Nougaret, Kirill Makarov)