Boat, a Paris-based startup backed by Y Combinator, has entered the nascent market for AI agent computing infrastructure with a pricing strategy that undercuts established competitors by as much as eighteen-fold. The company now offers persistent Linux virtual machines for $0.036 per hour, according to its public documentation.
The move reflects a growing bet among infrastructure providers that artificial intelligence coding assistants will soon migrate from developers' laptops to always-on cloud environments, particularly as these tools graduate from experimental projects to production workloads. Where that inflection point arrives remains uncertain, but venture-backed companies are racing to stake out territory.
Boat's offering centers on what it calls "persistent Linux VM sandboxes for AI agents." Each instance runs a full Ubuntu environment with Docker pre-installed, SSH access, and dedicated IP addresses. These capabilities often come as premium add-ons in container-based alternatives, the company argues, though competitors dispute the direct comparisons.
Computing Power at Container Prices
The startup's default configuration allocates four virtual CPUs and eight gigabytes of RAM. Billing operates on a per-second basis while a virtual machine runs; stopped instances cost nothing and can resume from snapshots within seconds, according to the company's FAQ.
Four machine tiers are available, all running Ubuntu 24.04 LTS on x86_64 architecture. The range spans from a small configuration with two vCPUs up to an xlarge setup with 16 vCPUs and 32 GB of memory, allocated by request. Each VM includes up to two terabytes of monthly data egress and one snapshot retention. Documentation indicates the sandboxes operate on AMD Ryzen 9 9950X bare-metal hosts, falling back to Hetzner virtual machines when capacity constraints emerge.
The pricing puts Boat well below competitors tracked in the company's September comparison, which listed per-hour rates ranging from $0.33 for services like E2B and Daytona to $0.60 for Islo. Whether those rates reflect equivalent configurations remains a point of contention in a market where providers package features differently.
Developer Tooling and Agent Frameworks
Boat ships with pre-configured integrations for several coding agent frameworks, including Claude Code, Codex, OpenCode, and Prime Agent. Developers can execute tasks through a command-line interface, switch between models during active sessions, and preserve state across stop-resume cycles.
The company has released Python and TypeScript SDKs alongside a REST API. Registry data shows the JavaScript SDK reached version 1.2.0 around October 1, while the Python package first appeared on PyPI in mid-September. The relatively recent release dates suggest the product is still in early stages, though the company has not disclosed active user counts.
Benchmark Wars and VM Philosophy
In June, Boat commissioned what it described as an independent benchmark comparing five sandbox providers. The test involved building a Devin-style application three times on each platform. All 15 attempts produced functioning applications, but Boat claims its pause-resume-fork lifecycle succeeded in all trials where competitors faltered. The methodology and independence of the test have not been verified by outside parties.
A separate Node.js benchmark published by the company shows 1,000 build-test runs completing in 36 seconds on a four-vCPU Boat machine, compared to 54 seconds for Daytona and 89 seconds for E2B. Those measurements occurred on different dates per provider, with Boat's run dated August 24 and a Vercel Sandbox test conducted September 29.
The company's positioning leans heavily on the virtues of full virtual machines over containers. While rivals tout sub-second resume times for lightweight containers, Boat argues that persistent filesystems, running processes, and the ability to install arbitrary software matter more for complex agent workflows. A blog post from August frames seconds-level restore as an acceptable trade for those capabilities.
That stance puts the startup at odds with recent announcements from larger players. Docker introduced Cloud Sandboxes in late September, emphasizing microVM security and sub-second resume. DigitalOcean launched Managed Agents two days earlier, with investor materials arguing that agent environments must "start and resume in milliseconds, not the minutes a VM takes." Boat appears to be wagering that a small segment of developers will prioritize flexibility over speed.
Industry observers note that even OpenAI may be leaning toward VM-based infrastructure for agent work. Tom's Hardware reported in September that Geekbench results suggest the company's "dots" agent operates on nine-core AMD EPYC virtual machines, though OpenAI has not confirmed those details.
A Small Team with Big Ambitions
Boat emerged from Y Combinator's batch earlier this year. The three-person founding team includes Anicet Nougaret as CEO and an engineer identified only as Kirill in the CTO role, according to the accelerator's company listing. The Delaware-incorporated entity operates under the legal name Delade AI Corp.
The service currently runs sandboxes in Germany, Finland, and France, with additional regions planned but not yet scheduled. Boat offers 25 hours of free trial usage and supports between 100 and 2,000 concurrent virtual machines per account, expandable through direct agreement. SOC 2 certification is in progress, the FAQ notes, though no timeline has been provided.
According to an April report from Nordic9, the company raised a $500,000 pre-seed round from Entrepreneurs First, Founders Inc., and angel investor Guillaume Roux-Romestaing. That funding came under the brand ASCII, which Boat still uses in some contexts.
The aggressive pricing raises familiar questions about unit economics in infrastructure businesses. At $0.036 per hour for four vCPUs and eight gigabytes of RAM, gross margins depend heavily on efficient resource utilization and bargain procurement of underlying compute. Whether the startup can maintain those rates as it scales, or whether this represents an early land-grab strategy, will likely become clearer in the coming quarters.
For now, Boat is making a straightforward argument: that developers building serious AI agent workflows will eventually need the full power and flexibility of Linux virtual machines, and that someone should sell them cheaply. The market will decide if enough customers agree.
