Cost to add internationalization to a web app
Cost to add internationalization to a web app
Adding internationalization (i18n) to a web app in 2026 typically costs $4,000 to $60,000 in engineering, plus $50 to $1,500 per month in translation tooling. A small greenfield app ships in 1-2 weeks. Retrofitting a mature SaaS takes 4-8 weeks before the first non-English string ever reaches a user. The biggest variables are string count, how much UI is hard-coded, and whether you actually need 3 languages or 25.
The honest version of this article that nobody else writes: most pre-PMF startups should not ship i18n yet. We will get to that. First, the real cost math.
What you are actually buying when you ship i18n
People say "i18n" like it is one feature. It is six.
- String extraction. Pulling every hard-coded string out of components and into a translation catalog (JSON, PO files, ICU messages).
- Locale routing. URL strategy (
/en/...,/fr/..., subdomains, or accept-language), middleware, and SEO tags. - Translation management. Where translators (human or AI) actually edit copy. A TMS like Lokalise, Crowdin, or Phrase, or a self-hosted Tolgee.
- The translations themselves. Either AI (Claude, GPT, DeepL) or human translators, or both in a tier.
- Format and locale correctness. Dates, currencies, numbers, pluralization rules (CLDR), bidirectional layouts for Arabic and Hebrew.
- SEO and legal overhead. Hreflang tags, locale-specific URLs in your sitemap, cookie banners that differ per jurisdiction, and review copy that is legally accurate in each market.
When founders quote a $5k i18n project, they usually mean line item 1. The other five are where budgets blow up.
Engineer-week cost: small app vs large retrofit
A clean greenfield app with maybe 200 strings and modern React Server Components takes a competent engineer 1-2 weeks to wire up next-intl, set up locale routing, extract strings, and ship 3 languages with AI-translated copy.
A 3-year-old SaaS with 8,000 strings, hard-coded English in 40 React components, fixed-width buttons, custom date pickers, and a marketing site that was built before the product? 4-8 weeks, and that is before a single translator touches the file. Retrofitting an existing app for i18n is one of the most painful refactors in web development because the work is everywhere and tedious, not deep.
The cost math at standard senior rates ($100-180/hr for US contractors) lands roughly here:
| App profile | Engineer weeks | Engineering cost |
|---|---|---|
| Greenfield, ≤500 strings | 1-2 | $4,000 to $12,000 |
| Mid SaaS, 1-3k strings | 3-5 | $12,000 to $30,000 |
| Mature app, 5-10k strings, retrofit | 4-8 | $25,000 to $60,000 |
| Enterprise app, 10k+ strings, multi-team | 8-16 | $60,000 to $150,000 |
The expensive variable is almost never library choice. It is how much existing UI was built without i18n in mind.
Library and routing choice (and what each costs you)
For a 2026 build, the practical shortlist for React/Next.js is small. Bundle size matters because translation catalogs ship to the browser; framework fit matters more.
| Library | Bundle | Best for | Tradeoff |
|---|---|---|---|
| next-intl | ~4 KB | New Next.js App Router projects | Locked to Next.js |
| react-i18next | ~6 KB | Cross-platform (Web + React Native) | Needs workarounds for Server Components |
| react-intl (FormatJS) | ~20 KB | Enterprise ICU messages, complex plurals | Largest bundle, steepest learning curve |
| Lingui | ~2-3 KB | Compile-time extraction, great DX | Smaller community |
| Paraglide-JS | ~1 KB / locale | Maximum bundle optimization | Newer ecosystem, build step |
| next-i18next | ~6 KB | Pages Router legacy projects | Being phased out for next-intl |
For most 2026 Next.js apps, the answer is next-intl unless you have a specific reason. For Vue, vue-i18n. For Remix, remix-i18next. For SvelteKit, paraglide-sveltekit. For React Native cross-platform, react-i18next is still the right call.
A senior engineer makes this decision in an hour. A team that argues about it for two sprints is the actual cost driver.
Translation management platforms: what they really cost
The TMS is where the per-month bill lives. Public pricing as of 2026:
| Platform | Pricing | Strength |
|---|---|---|
| Lokalise | $80 to $1,500+/mo (per project + seats) | Mature workflow, integrations, AI features |
| Crowdin | Free OSS / $50+/mo paid | Great for OSS projects, generous free tier |
| Phrase | $27+/user/mo | Enterprise governance, strong API |
| Tolgee | OSS self-hosted (free) / cloud from $25/mo | In-context editing, founder-friendly pricing |
| Locize | Usage-based, free up to 2k words | Built by the i18next team, tight integration |
| i18nexus | Free for small projects, paid as you scale | Fastest setup for indie projects |
If you are shipping fewer than 3 languages and 500 strings, JSON files in the repo work fine. Past that, a TMS earns its keep on translator workflow alone (no engineer should be merging Polish strings via PR).
The mid-bracket reality for a Series A SaaS with 5-7 languages is roughly $200-600/mo on Lokalise or Phrase. Crowdin and Tolgee are realistic free options if your team is comfortable owning a self-hosted instance.
AI translation vs human translators
This is the line item that swung hardest from 2023 to 2026.
AI translation via Claude, GPT, or DeepL Pro now runs about $0.001 to $0.01 per string, depending on length and model. A 5,000-string app translated to 10 languages is roughly $50 to $500 in API spend, one-time, with marginal cost on updates.
Human translation still costs $0.10 to $0.30 per word, plus 20-30% for review. The same 5,000 strings (assume 6 words each, so 30,000 words) at 10 languages runs $30,000 to $90,000.
The realistic 2026 stack:
- AI for v1 of every string. Cheap, fast, good enough for 90% of UI copy.
- Human review for legal, billing, marketing, and brand voice. A native speaker for Spanish and German on your top 5 pages is $500-2,000 and pays for itself in not getting roasted on Twitter.
- Translation memory (every TMS has this) so you only pay to translate the changed string, not the whole file, on each release.
AI fails predictably on: anything legally regulated (terms of service, GDPR notices), brand voice (your marketing copy in German will sound like a manual unless you review), and idioms (the AI will get them right grammatically and wrong culturally). Budget human review there.
If you are migrating data layers as part of a broader rebuild, our cost to migrate from Firebase to Supabase breakdown shows similar patterns: the engineering work is bounded, but the surrounding ecosystem (auth, RLS, observability) is where time disappears.
The hidden costs nobody quotes
These are the line items that turn a "$10k i18n project" into a $40k project halfway through.
- RTL languages. Arabic and Hebrew flip your UI. Margins, padding, icons, animations, even table layouts. Plan 30-50% extra UI engineering if you need RTL. Tailwind's
rtl:modifier helps; it does not eliminate the work. - Pluralization rules. Polish has 4 plural forms. Arabic has 6. English has 2. ICU MessageFormat handles this; you have to actually write the rules. CLDR plural categories are the underlying spec.
- Date, currency, and number formatting. Use the standard
IntlAPI (built into every modern browser/runtime), plusdate-fnsorLuxonfor relative time. HardcodedMM/DD/YYYYis a bug in every market that is not the United States. - Hreflang and locale URLs. Every page needs
<link rel="alternate" hreflang="...">tags pointing to its translations. Get this wrong and Google serves your French page to English searchers. Sitemaps must list every locale variant. This is 1-2 days of SEO engineering nobody scopes. - Cookie banners and legal review. GDPR, ePrivacy, the UK's separate post-Brexit rules, California's CCPA, Brazil's LGPD. Each market has nuances. Use a vendor (OneTrust, Cookiebot, Termly) and budget legal review for the ToS in each language at $1,000-5,000 per market.
Add these together: an honest i18n bill at 5 languages with RTL support and proper SEO and legal review lands at $25,000 to $50,000 all-in for the first year, not the $5,000 the engineering estimate suggests.
Real budget math: 3, 10, and 25 languages
Here is the actual P&L for a mid-sized SaaS with 2,500 strings.
| Cost item | 3 languages | 10 languages | 25 languages |
|---|---|---|---|
| Engineering (string extraction, routing, library) | $12,000 | $14,000 | $18,000 |
| TMS (annualized) | $1,200 | $4,800 | $12,000 |
| AI translation (one-time, with updates) | $50 | $200 | $500 |
| Human review (top markets) | $1,500 | $8,000 | $25,000 |
| RTL UI work (if Arabic or Hebrew in scope) | $0 | $4,000 | $4,000 |
| SEO (hreflang, sitemap, locale routing) | $1,500 | $2,500 | $4,000 |
| Legal review per market | $3,000 | $15,000 | $40,000 |
| Year-1 total | $19,250 | $48,500 | $103,500 |
Two things jump out. First, the engineering cost is roughly flat after the initial investment; you build the pipeline once. Second, legal review and human translation scale linearly with markets, and they dominate the bill past 10 languages.
The 25-language plan is not "5x the 3-language plan." It is a different business decision.
Cost by approach: full-time, agency, freelancer, Cadence
If you are a founder who needs i18n shipped and is choosing how to staff it, the realistic options:
| Approach | Cost | Timeline | Pros | Cons |
|---|---|---|---|---|
| US full-time hire | $140-220k/yr | 8-12 weeks (hire + ramp) | Owns it long-term | Massive overkill for a one-off i18n project |
| Dev agency (US/EU) | $25-80k fixed bid | 6-12 weeks | Predictable scope, project managers included | Slow kickoff, expensive change orders |
| Freelancer (Upwork) | $3-15k | 3-8 weeks | Cheapest line item | Vetting overhead, quality variance, rarely AI-native |
| Toptal | $60-110/hr ($10-30k project) | 2-4 weeks to start | Vetted talent | Premium hourly creates friction |
| Cadence | $500-$2,000/wk | 48-hour trial then ship | AI-native by default, weekly billing, replace any week | Less suited to enterprise procurement |
A few honest notes. Toptal is genuinely good if your procurement department needs a recognizable vendor and you are billing under an enterprise contract. Agencies are the right call when scope is fuzzy and you want a project manager between you and the engineer. Freelancers from Upwork win on price; vetting is on you.
For most startup-shaped i18n projects, the right shape is a senior contractor for 3-5 weeks: long enough to ship the pipeline, short enough that you are not paying full-time benefits for a one-off. Every engineer on Cadence is AI-native by default, vetted on Cursor, Claude Code, and Copilot fluency before they unlock bookings, which matters here because the AI-translation pipeline plus string-extraction work is exactly the kind of repetitive code transformation an AI-native engineer ships in days, not weeks. Median time to first commit on the platform is 27 hours.
When you should not ship i18n yet
The unwelcome answer: most pre-PMF startups should not ship i18n.
Symptoms that you are too early:
- You do not have measurable traffic or signups from a non-English market.
- Nobody has asked. (Search your support tickets for "Spanish" or "français." Zero hits = wait.)
- You are doing it because a competitor did, or because it sounds good in a pitch deck.
- You have not nailed your English copy yet. (Translating bad copy 10x produces 10x bad copy.)
Symptoms that it is time:
- 15%+ of your inbound traffic comes from non-English markets and bounces fast.
- Customers are paying you in euros and asking for invoices in their language.
- You are running paid ads in a market where your landing page does not match the ad language.
- Your sales team is losing deals because the product UI confuses non-English buyers.
i18n is rarely the constraint on early product-market fit. It is often the constraint on expansion past it. Ship English-only until you have hard signal that your buyers want something else. Similar tradeoffs show up when you add AI features (see our cost to add RAG to a SaaS app breakdown): the right sequence is product-market fit first, scale-up infrastructure second.
How to reduce costs without cutting corners
Five things that consistently work:
- Build with i18n in mind from day one. A
t('home.hero.title')from the start adds 1-2 days; retrofitting it adds weeks. Even if you only ship English, set up the pipeline. - Start with 3 languages, not 10. English plus the next two largest markets. Validate the workflow on 3 before you commit to a 25-language plan that doubles ops cost.
- Use AI for v1, humans for review on the top 20% of pages. The pricing page, ToS, and homepage hero get human review. The settings panel does not.
- Pick a TMS that lets you self-host or grow into the bill. Tolgee or Crowdin's free tier are real options; you do not need to start on Phrase Enterprise.
- Hire for the spike, not the role. A 3-week senior contractor who has shipped i18n before is faster and cheaper than a full-time hire who learns on your dime. Our breakdown of a related project pattern, the cost to integrate Stripe payments, follows the same logic: bounded scope plus an AI-native contractor outperforms hiring full-time.
The fastest path from English-only to multilingual
The 3-step playbook for a founder who decided this week:
- Pick the library and route in one afternoon. next-intl for Next.js, react-i18next for cross-platform, vue-i18n for Vue. Stop debating.
- Pick a TMS and set up the AI translation pipeline. Tolgee (free) or Crowdin (free OSS) for week 1. Migrate to Lokalise or Phrase later if you outgrow them.
- Ship 3 languages first. English, plus your top two non-English markets by traffic. Measure conversion lift over 4 weeks. If the lift is real, expand. If not, you saved $40,000.
If you do not already have an engineer on this, the fastest path is a senior contractor on a 3-5 week engagement. Booking a senior engineer on Cadence takes about 2 minutes from spec to shortlist, with a 48-hour free trial before you pay for the first week.
FAQ
How long does it take to add i18n to a web app?
1-2 weeks for a small greenfield app, 4-8 weeks to retrofit a mature SaaS. The string-extraction pass dominates the timeline. If your UI was built with hard-coded English everywhere, expect the longer end of that range.
Should I use AI translation or hire human translators?
Use AI for the first pass at $0.001 to $0.01 per string. Add human review for legal copy, marketing pages, and brand-voice content at $0.10 to $0.30 per word. Most teams ship AI-only for v1 and add human review for their top 1-3 markets once revenue justifies it.
Which i18n library should I pick for Next.js in 2026?
next-intl for App Router projects (4 KB, full RSC support, type safety). react-i18next if you also need React Native. Paraglide-JS if bundle size is critical and you have time to learn the compiler-based approach. Avoid next-i18next on new projects; it is being phased out.
Do I need a TMS like Lokalise or can I just commit JSON files?
Up to 3 languages and ~500 strings, JSON files in the repo are fine. Past that, a TMS pays for itself in translator workflow alone. No engineer should be merging Polish or Arabic copy via pull request.
Is i18n worth the cost for a pre-PMF startup?
Almost never. English-first ships product-market fit faster, and you can always add languages later. Add i18n once you have measurable inbound from a non-English market, paying customers asking for it, or a sales team losing deals over language fit. Otherwise, ship the next product feature instead.
What is the cheapest credible way to ship i18n in 2026?
next-intl plus Tolgee (self-hosted, free) plus AI translation via Claude or DeepL plus a senior contractor for 3-5 weeks. Expect $8,000 to $20,000 all-in for 3 languages on a mid-sized app, with no recurring TMS bill.