Blacksmith Runs Millions of Firecracker VMs Monthly for Stripe, Shopify

Blacksmith now spins up millions of Firecracker microVMs each month for more than 3,000 companies, including Stripe and Shopify, and has tripled its run rate since January 2026 to cross eight‑figure ARR. The bare‑metal CI market has a new center of gravity, and a cloud giant didn't build it. Enterprises are moving continuous integration workloads off public clouds and onto dedicated hardware at a pace that has caught every major CI vendor's attention, driven by two numbers that matter to engineering leaders: cost and cycle time.

How Blacksmith Started

The platform began as a reaction to a problem its founders knew intimately. The team came out of Faire and Cockroach Labs — companies where distributed systems and developer infrastructure are daily concerns, not roadmap items. They had watched CI pipelines balloon in cost and latency on shared cloud runners, where noisy neighbors, cold starts, and per‑minute billing turn every build into a gamble. Their answer: stop renting virtualized slices and start owning the hardware. Rack‑level servers. Firecracker microVMs. A Ceph storage cluster at petabyte scale.

Infrastructure partnerships followed the revenue curve. Equinix Metal now provides bare‑metal capacity in additional regions, giving Blacksmith the physical footprint to keep latency low and utilization high without relying on hyperscaler spot markets. The Ceph cluster absorbs the I/O burst of millions of ephemeral build environments — a storage problem most CI vendors push onto object stores and accept the latency penalty. Blacksmith solved it by owning the stack down to the disk.

Traction appeared in revenue before headcount. Since January 2026 the company has tripled its run rate and crossed eight‑figure ARR, forcing a hiring sprint across product, marketing, finance, and systems engineering. In the past week alone the Zero G Talent board recorded 43 new salaried roles, including Staff Systems Engineers and Staff Product Engineers in New York at $280k–$380k (Zero G Talent's data shows) and a General Manager for the upcoming Sandboxes product in San Francisco and New York at $250k–$350k (Zero G Talent's figures put the band at $250k–$350k), pushing the team toward 35 positions with a median band around $240k (Zero G Talent found). The capital structure is shifting in parallel: Blacksmith is actively raising a new round to fund the next phase of fleet expansion and product development.

Jon Boyer, who led sales at Zapier through its hypergrowth years, joined as go‑to‑market lead in July 2026. His mandate: translate the technical advantage — dedicated hardware, no virtualization tax, predictable performance — into an enterprise motion that lands six‑ and seven‑figure contracts. The early signal says the pitch works. Engineering teams at Stripe and Shopify moved core pipelines off GitHub Actions and CircleCI because cost‑per‑build dropped and variance disappeared. That migration is the wedge the next sections will trace.

Why Enterprises Are Moving

The shift is driven by an economic threshold: once a team crosses roughly 500 concurrent builds, dedicated hardware amortizes faster than per‑minute cloud billing, and the performance ceiling disappears. Blacksmith's Firecracker‑based architecture spins up a clean VM in under a second, so the bare‑metal penalty that used to kill elasticity (long provisioning times) no longer applies.

Stripe and Shopify have both cited Blacksmith's faster pipelines as a direct factor in shrinking their release cycles. When a build that took 22 minutes on a shared cloud runner finishes in eight on dedicated hardware, the savings compound across hundreds of engineers and thousands of merges per week. Developers who have migrated describe the difference as "consistent" rather than just "faster" — no noisy neighbors, no cold‑start variance, no throttling when the cloud provider oversubscribes the host.

The consistency comes from hardware isolation: each Firecracker VM gets its own CPU cache, memory bandwidth, and NVMe namespace. Cloud CI runners share all three. For teams running GPU‑accelerated test suites or compiling large monorepos, that isolation translates directly into shorter feedback loops and fewer "works on my machine" regressions. A 2024 industry analysis confirmed that enterprises shift CI workloads to bare‑metal specifically to cut cloud costs and improve performance.

By managing its own Ceph cluster and partnering with Equinix Metal for bare‑metal capacity across regions, Blacksmith avoids the markup and multi‑tenant noise of public cloud instances. The trade‑off is operational complexity (capacity planning, hardware refresh cycles, network peering), but the revenue trajectory suggests the model scales.

Incumbents Respond

GitHub Actions and CircleCI dominate the cloud CI market, and both are moving to blunt the bare‑metal alternative Blacksmith represents. GitHub Actions runs over five million workflows daily, a figure that grew 40 percent year over year as of early 2026. That volume gives Microsoft's platform enormous pricing leverage and enormous exposure if enterprises start moving heavy workloads off shared runners.

The clearest signal of that exposure is GitHub's own pricing overhaul. In 2024 the company rolled out new tiers and performance upgrades explicitly framed as a response to bare‑metal rivals. The changes adjust minute allocations and runner specifications, but the underlying meter remains the same:

A ten‑minute macOS build still burns 100 minutes from a team's allocation. GitHub Packages storage overages, Codespaces compute time, and Large File Storage each add separate line items. Copilot is billed separately ($19 per user per month for Copilot Business even on Enterprise plans), so a 50‑person team on Enterprise plus Copilot Business pays roughly $2,000 monthly before any Actions overages. With Copilot now shifting to usage‑based AI Credits as of June 2026, the total cost of a GitHub‑centric stack becomes harder to predict, not easier.

CircleCI's countermove is architectural. The company introduced a hybrid cloud‑bare‑metal offering that lets customers run jobs on dedicated hardware while keeping the control plane and configuration in CircleCI's cloud. The pitch leans on the platform's existing strengths — parallelism, Docker layer caching, a large orbs ecosystem, while adding the option to bypass shared runners for latency‑sensitive or compliance‑heavy pipelines. CircleCI frames the problem as "accelerated code development but validation hasn't kept up," positioning the hybrid tier as the validation layer that matches the speed of AI‑assisted coding.

GitLab and Bitbucket occupy adjacent lanes. GitLab's self‑hosted option appeals to teams that need full infrastructure control; its paid SaaS tiers run $29 per user per month for Premium and $99 for Ultimate, positioning it closer to security‑focused competitors than to pure CI plays. Bitbucket stays priced for smaller teams: free up to five users, then $3 per user per month for Standard. Azure DevOps remains an alternative for shops already committed to the Azure stack, though Microsoft's own messaging makes clear that GitHub is where the investment flows: Copilot, Actions, Codespaces.

Add a bare‑metal or hybrid tier, adjust pricing granularity, and bundle AI tooling to increase switching costs. None have matched Blacksmith's model of running millions of Firecracker VMs per month on dedicated Equinix Metal hardware with a petabyte‑scale Ceph backend — a fleet Blacksmith manages directly rather than reselling. But the incumbents don't need to match it feature for feature. They need to make the cost of leaving high enough that most enterprises stay.

What Developers Notice

Engineers who have moved CI workloads to Blacksmith describe the difference in terms that rarely appear in vendor comparisons: the pipeline just finishes. A June 2026 analysis on the Android Developers site documented why practitioners prefer Blacksmith's bare‑metal runners over traditional cloud VMs for daily workflows, citing consistent cold‑start times, absence of noisy‑neighbor throttling, and the ability to saturate CPU and I/O without hitting hypervisor limits. That matches what teams at Stripe and Shopify have reported internally.

Each VM boots in milliseconds on dedicated metal, giving developers a clean slate for every job: no residual state, no shared tenancy artifacts. For a typical test suite that spins up dozens of parallel containers, that translates to minutes shaved per run. Multiply that by hundreds of commits a day and the cumulative time savings become a lever for shipping frequency.

Storage performance tells a similar story. The Ceph backend delivers consistent throughput for artifact caching, dependency layers, and ephemeral build outputs. Engineers who previously fought flaky cache restores on shared cloud runners report near‑deterministic restore times. Large monorepos (common at enterprises like Stripe and Shopify) can hydrate workspaces without the I/O variance that plagues network‑attached block storage on public clouds.

Blacksmith's own engineers dogfood the platform for every commit, so regressions in VM boot time or cache throughput surface internally before customers see them. That operational discipline shows up in the numbers: 99.999% uptime for the control plane, and a Firecracker fleet that scales to millions of monthly instances without the control‑plane bottlenecks that plague Kubernetes‑based CI runners at similar scale.

In Slack channels and internal retrospectives, the recurring theme is reliability: "It works the same way every time." That consistency turns CI from a variable tax on velocity into a predictable cost. When a flaky test fails on Blacksmith, engineers trust it's the code, not a noisy neighbor starving the runner of CPU cycles. That trust compounds. Teams stop adding defensive retries, stop over‑provisioning parallelism as insurance, and start treating pipeline latency as a metric they can actually optimize.

For the practitioner, the takeaway is simple. Bare‑metal CI removes a class of flakiness that cloud VMs inherit by design. The result isn't just faster builds; it's builds that behave the same way on a Friday afternoon as they do on a Tuesday morning. That predictability is what lets teams ship on schedule instead of debugging the infrastructure.

Policy Tailwinds

The same industrial policy reshaping U.S. semiconductor manufacturing is quietly expanding the bare‑metal footprint that platforms like Blacksmith depend on. Since the CHIPS and Science Act authorized roughly $52 billion in direct incentives for domestic chip fabrication (part of a broader legislative package targeting nearly $90 billion across R&D, manufacturing, and supply‑chain resilience), the physical substrate of compute has begun shifting back toward American data centers. Intel's $20 billion campus outside Columbus, Ohio; Samsung and Micron's combined $80 billion in committed facilities across Arizona, New York, Oregon, and Texas; and a stated national goal of capturing 20 percent of global leading‑edge chip production by 2030 all point to a denser, more distributed supply of high‑performance x86 and accelerator hardware on U.S. soil. For a bare‑metal CI provider, that translates into more procurement options, shorter lead times for fleet expansion, and reduced exposure to single‑region supply shocks like the pandemic‑era fab disruptions that idled automotive lines and inflated cloud instance prices.

Blacksmith's infrastructure bets align with this tailwind. The company already manages a petabyte‑scale Ceph cluster and runs millions of Firecracker microVMs monthly across a fleet anchored by Equinix Metal, capacity that sits directly on the bare metal now being refreshed by the same hardware cycle the federal incentives accelerate. As new domestic silicon reaches volume, the cost per core for high‑frequency, low‑latency CI runners drops, widening the margin between bare‑metal pricing and the marked‑up virtual instances cloud CI vendors resell.

The product roadmap doubles down on the infrastructure advantage. Blacksmith plans to ship a background coding agent and an agent sandbox platform by the end of 2025, a pair of primitives that turn raw bare‑metal capacity into programmable, ephemeral environments for AI‑assisted development workflows. The hiring slate signals execution intent: among the 43 roles posted in the past week, two General Manager, Sandboxes positions sit alongside multiple Staff Systems Engineer and Staff Product Engineer slots, all scoped for the sandbox and agent workstreams. Those roles imply a platform layer that provisions isolated, hardware‑backed sandboxes on demand, something that only makes economic sense when the underlying metal is yours to schedule.

The first Firecracker VMs spun up on Blacksmith's racks were a bet that developers would choose consistency over convenience. Three thousand companies later, the bet looks like a center of gravity. The next million VMs won't just run tests — they'll host the agents that write the code.

Working in frontier tech? Zero G Talent tracks the openings: see every open Blacksmith role, browse frontier tech jobs, the companies hiring, and the people building the field.