When a browser needs a really big machine
Standard sandstack sandboxes run on shared CPU/RAM sized for typical browsing. When your agent needs to hold 200 tabs open, chew through a 10-hour scraping run, or drive a canvas-heavy dashboard, spin it up on a memory-optimized or high-memory host.
Available on Pro and Team — request access and we'll provision within 1 business day.
Memory-optimized (r / x / z)
The realistic upgrade path. High RAM-to-vCPU ratio, best for most parallel-tab, long-agent, and heavy-SPA workloads.
r7i.4xlarge
r7i.24xlarge
x2iedn.16xlarge
z1d.6xlarge
High-memory (u-*)
Bare-metal, terabyte-scale. Reserved for the extreme end — SAP-HANA-adjacent workloads, whole-dataset in-memory analysis over automation output.
u-6tb1
u-9tb1
u-12tb1
u-18tb1
u-24tb1
What people actually use these for
Massively concurrent scraping
One customer, 200+ tabs open at once, each pulling a different product page. Standard sandboxes swap to disk and slow to a crawl — r7i.4xlarge keeps everything hot in RAM.
Long-horizon agent runs
Multi-hour agents accumulating context, DOM snapshots, screenshots, and extracted state in-process. The r/z families give you 4-16 GB per session with headroom for the Claude conversation history plus browser state.
fit → r-family / z-familyHeavy JS SPAs and dashboards
Figma files, trading terminals, Airtable, Notion databases, WebGL sites. These eat 2-6 GB per tab. z1d (fast clock + high RAM) is the pick when responsiveness matters as much as capacity.
Video / media-in-page pipelines
Running through TikTok, YouTube, Instagram Reels with buffering + capture. Media decode + full-frame screenshots at 1080p burn RAM fast — memory-optimized hosts absorb the spikes.
fit → r-familyIn-browser data / ML visualizations
Genomics browsers, molecular viewers, GIS map tiles at continental zoom, WebGPU inference demos. Web pages that were never meant to fit on a laptop.
fit → z-familyFleet automation for org accounts
Teams running 500+ parallel sandboxes for QA regression suites, price-monitoring at scale, or comprehensive competitor sweeps. x2iedn holds the whole fleet on one host.
Whole-warehouse in-memory analysis
Feed the scraped output straight into a duckdb/pandas process that lives on the same host — no round-trips to a warehouse. This is what the u-* line is for; niche but transformative when it fits.
fit → u-familyRegulated or sole-tenant compute
Bare-metal u-* instances give you dedicated hardware — no noisy neighbours, no shared kernel. Comes up in fintech, healthcare, and government contexts where SOC 2 / HIPAA controls demand isolation.
fit → u-familyCustom shape you have in mind
Have a specific instance type you want (i-family SSD-heavy, m7g Graviton, g6 GPU)? We can provision most EC2 shapes on request — just tell us the workload profile.
fit → anythingPricing
Custom, based on the instance you pick and how long it runs. Standard model is list AWS on-demand price + 25% platform fee, billed hourly. Reserved-instance discounts (up to 40%) available for month-long commitments. Storage, egress, and proxy bandwidth are pass-through, no markup. Full quote in the intro call.
Ready to size up?
Tell us the workload — tab count, session length, concurrency, region — we'll come back with an instance recommendation and a quote inside a business day.
Contact sales → See standard plans