Build a Marketplace for $150/mo

Hosts the Next.js app and the background worker for image processing, listing approval, and payout reconciliation.

Vercel and Cloudflare are great for stateless requests, but a marketplace has long-running work — image resize jobs, scheduled payouts, fraud-check pipelines — that doesn't fit serverless cleanly. Render's Starter at $19 gives you a web service plus a worker, both with persistent connections to your database and queue. The deploy story is git-push the same as Vercel.

• Cold starts on the Starter tier are slower than Vercel's edge — first request after idle can spike to 3+ seconds. Use Render's keep-alive or upgrade to Standard if it bites.
• Background workers don't have a free tier; the moment you add the second service, you're at $19 + $19. Keep the worker count down by batching jobs.
• Region selection matters. Pick the region closest to your database (Neon's region is the constraint) — cross-region calls add 50-100ms.

Serverless Postgres with autoscaling compute and storage, plus database-branching for preview environments.

Marketplaces have joins — buyer ↔ listing ↔ seller ↔ payment ↔ review — and Postgres handles them well. Neon Scale at $19 includes 10 compute hours and connection pooling that survives serverless fan-out. The branching feature means every Render PR gets its own isolated database, which catches schema bugs before main.

• Cold-start compute is real. After 5 minutes of idle, the next query takes ~2 seconds to wake. Set a heartbeat ping or upgrade if user-facing latency matters.
• The 10 compute-hour cap on the Scale tier sounds large but isn't if you have constant traffic. Budget for the next tier ($69/mo for Pro) if usage trends up.
• Neon's branching is fast but not free — every active branch consumes storage. Auto-expire PR branches in CI to avoid surprise bills.

Typo-tolerant, sub-50ms search across listings, with faceted filters and ranking rules.

Postgres full-text search is serviceable for blog content; for a marketplace it's not. Buyers expect Algolia-quality search — typos forgiven, filters stacked (price, location, condition), ranking that puts good listings first. Meilisearch's pricing at $30 covers up to 100k documents with sub-50ms responses, and the SDK reads like REST.

• Index sync is your job. Either dual-write from app code (risky, can drift) or use a CDC pipe (Debezium, custom worker). The worker on Render is the right place for it.
• Synonym tuning takes time. Out of the box, 'jeans' won't match 'denim' — you set those mappings explicitly. Plan a few hours of search relevance work in week 2.
• 100k documents on the $30 tier is plenty for a niche marketplace; broad horizontal marketplaces (everything-for-everyone) outgrow it fast.

Accepts payment from buyers, splits the platform fee, pays sellers, and handles seller identity verification (KYC) for tax compliance.

Connect is the only path that's both legal and not custom-built. Sellers on your platform are paid through Stripe-managed accounts — Stripe handles the tax forms (1099-K), identity verification, and disputes. Your platform takes its cut on every transaction. Without Connect you'd be running money for unverified sellers, which is illegal in the US and most other jurisdictions. The $25 is the platform-fixed Connect fee — transactional fees are on top (2.9% + 30¢ per charge, plus 0.25% + 25¢ per payout).

• Onboarding sellers through Stripe-hosted Connect Onboarding is slower than your homepage suggests. Some sellers (especially international) get stuck on identity docs for days. Build a clear status page so they can self-serve.
• 1099-K tax forms are filed by Stripe but you're responsible for sellers actually receiving them. Build a year-end communication flow.
• Disputes are routed to your platform email by default. Without a process, they'll pile up and you'll lose them by default — Stripe rules in the customer's favor on no response.

Sends every transactional email — listing alerts, message notifications, payment receipts, dispute notices.

Marketplaces have higher transactional volume than SaaS — every listing change, message, and transaction triggers an email. Postmark's $15/mo covers 10k emails and has the best transactional deliverability of any provider in this price range. Their server architecture separates transactional from marketing, so a misconfigured marketing send doesn't tank deliverability for receipts.

• Postmark blocks marketing email types by policy. Anything that looks like a digest, broadcast, or promotional send needs a separate ESP (Beehiiv or MailerLite). Don't try to game it — they suspend accounts.
• Bounce and complaint handling is automatic but the dashboard is sparse. If you want detailed deliverability analytics, the $50 tier or a service like SparkPost is more visible.
• Templates use a Markdown-like syntax (Mustache + Markdown). It's fine but doesn't accept React-email components, unlike Resend.

Stores listing images with global CDN delivery and zero egress fees.

S3-compatible storage at $0.015/GB/month, with zero egress charges — the differentiator. A marketplace serves the same images many times; on AWS S3 you'd pay $90/TB egress, on R2 it's free. $10 covers ~600GB of stored images. Cloudflare Images on top adds resize-on-the-fly if you don't want to bake an image pipeline.

• R2's API is S3-compatible but not exact. Some SDKs assume AWS-only behaviors (region in URL, presigned URL formats); test the SDK against R2 before assuming it works.
• No native image transforms in R2 itself — that's Cloudflare Images, a separate $5/mo service. Bake transforms with sharp on the worker, or pay for Images.
• Backup strategy is your job. Cloudflare won't lose your data, but versioning isn't on by default. Turn it on if listing-image deletion needs to be reversible.

Captures errors and performance traces from app + worker, with source maps and release tracking.

The Developer plan at $14 covers 50k errors and 100k performance events — enough for an early marketplace. The reason to pay even at this scale: marketplace bugs are often payment bugs, and a 5-minute outage on the checkout flow is direct lost revenue. Sentry's release tracking and breadcrumbs cut triage time from 'an afternoon' to 'fifteen minutes'.

• Sample rates apply per-environment. tracesSampleRate at 1.0 in production blows the quota in a week — set it to 0.1 from day one.
• Worker-side errors need separate Sentry SDK setup. Don't assume the web SDK is also catching background-job exceptions.
• The $14 Developer plan caps team members; Team plan ($26) adds unlimited members. With one founder you're fine; with a co-founder you're already over.