/ Kiz8/ Tech review/ Sandbox

Sandbox.

/ Where agents run code.

Compute sandboxes for AI agents — microVMs, containers, and snapshot architectures that let agents run code, browse the web, and persist state without crashing each other or the host.

46 vendors·Last update 2026-05-18

Trend · 04 of 07

$340M to $1.5M.
The revenue is bimodal too.

Estimated ARR for 10 funded sandbox vendors spans a 200× range. Vercel runs at $340M, Replit $265M — platform-extensions on mature products. The pure-plays sit two orders of magnitude lower: E2B at $1.5M, Daytona $3M, Northflank $2.1M. And then StackBlitz/Bolt went $0 → $40M in 5 months on a token-resale model. Funding shape predicts revenue shape.

Top ARR

$340M

Vercel · run-rate Feb 2026

Fastest growth

500×

StackBlitz · $80K → $40M, 5mo

Top revenue / FTE

$2.7M

StackBlitz · ~15 employees

Pure-play median

$2.1M

excl. platforms · n=4

Fig. 04 · Estimated ARR · 2026

The revenue dumbbell

Bars in √-scale · figures $M ARR · orange = platform play (sandbox bolted to mature product)

01VercelPlatform
$340
02ReplitPlatform
$265
03ModalGPU
$50
04ChainguardCompliance
$40
05StackBlitz / BoltToken resale
$40
06E2BPure-play
$1.5–18
07DaytonaPure-play
$3.0
08NorthflankBYOC
$2.1
09TensorlakeRL
$1.0
10Blaxel
$0.77

A · Hypergrowth

StackBlitz / Bolt

500×/ 5 mo

$80K$4M$20M$40M

In-browser WebContainers · zero cloud-compute COGS · sells Anthropic tokens at markup.

B · Hypergrowth

Replit

35×/ 6 mo

$2.8M$10M$100M$265M

Replit Agent · Firecracker microVMs · 1,556% YoY through 2025.

C · Enterprise

Chainguard

/ 24 mo

$5M$40M

640% YoY · compliance · custom registries · government networks.

D · Steady

Vercel

3.4×/ 24 mo

$100M$340M

84% YoY · operating with the discipline of a public entity · IPO-ready.

Fig. 05 · Revenue per employee

The AI-native efficiency gap

SaaS median · $167.5K / FTE · the top tier of AI-native vendors run 10–16× that. Ultra-lean teams, AI-leveraged automation.

Exceptional · $1M+

10–16×
median

StackBlitz~$2.7M · 15 FTE
Replit~$2.6M · 101 FTE

Strong · $300–500K

2–3×
median

Vercel~$378K · 900 FTE
Modal~$633K · 79 FTE

Medium · $150–250K

~1×
median

Chainguard~$200K · 200 FTE
Daytona~$167K · 18 FTE

Early · < $150K

0.4–0.7×
median

Blaxel~$110K · 7 FTE
E2B~$83K · 18 FTE
Northflank~$72K · 29 FTE

Fig. 06 · Capital efficiency

ARR / cumulative funding · the burn ratio

Best-in-class: 1.0×+ (every dollar raised has turned into a dollar of ARR). StackBlitz at 0.30× looks low but is exceptional given they were near-shutdown 18 months ago. Pure-plays cluster at 0.05–0.10× — still in investment mode.

Vendor
ARR / Funding ratio
ARR
Funding
Multiple
01Replit
1.18×
$265M
$224M
EXCELLENT
02Modal
0.45×
$50M
$111M
GOOD
03Vercel
0.39×
$340M
$863M
GOOD · @ scale
04StackBlitz
0.30×
$40M
$135M
HYPERGROWTH
05Daytona
0.10×
$3.0M
$31M
EARLY
06Northflank
0.07×
$2.1M
$30.8M
EARLY
07E2B
0.05×
$1.5M
$32.5M
EARLY
08Chainguard
0.04×
$40M
$892M
INVESTING

Bottom line for revenue

The revenue dumbbell mirrors the funding dumbbell — but with one wild card. Platform plays (Vercel, Replit) and token-resale models (StackBlitz/Bolt) are the only ways to scale past $40M ARR. Pure-play sandbox compute caps out under $5M ARR for now. The most efficient vendors run $1–2.7M revenue per FTE — that's the new bar for AI-native infra.

Compute is a cost line. The product is the surrounding ecosystem.

Active tab: Trend