Forex on AI subscriptions: what Indian startups actually lose in 2026

Every Indian startup with an AI workload is paying a quiet 3.5% tax on every USD subscription. Anthropic, OpenAI, Cursor, Vercel, Pinecone — all of them bill in USD. Most teams swipe a default retail card and never look at the forex line on the statement.

The math, when you finally add it up, is not subtle.

The default tax

A typical 5-person Indian AI startup in 2026 burns roughly $1,200/month on USD-billed SaaS:

| Tool | Typical monthly spend | |---|---| | Anthropic (Claude API) | $400–600 | | OpenAI (GPT API + ChatGPT Plus seats) | $200–300 | | Cursor Pro (5 seats) | $100 | | Vercel Pro | $20–60 | | Pinecone Standard | $70–150 | | Sentry Team | $26 | | Resend Business | $20 | | GitHub Team | $20 | | Linear Standard | $40 | | Total monthly | ~$1,200 |

That's $14,400/year ≈ ₹12.2L at a ₹85/USD rate.

At the default 3.5% forex markup, the bank quietly skims:

₹12.2L × 3.5% = ₹42,700/year

Forty-two thousand seven hundred rupees, every year, for the privilege of paying with rupees. That's a single great hire's signing bonus. That's three months of a junior engineer's compute budget.

Where the 3.5% comes from

The 3.5% isn't a single line item — it's two stacked fees most cards combine:

1. Bank cross-currency markup (~3% to 3.5%, depending on the issuer)
2. Visa/Mastercard network fee (~0.6%, usually absorbed into the headline 3.5%)

What the bank doesn't tell you: this is negotiated. Premium card tiers explicitly carve a lower forex rate as a feature. The standard premium rate is 2%. Specialised forex cards run 0%.

The math on switching

Three switching paths Indian founders actually have:

Path 1 — Premium Indian card (2% forex)

Move to an HDFC Diners Black, Axis Magnus, or SBI Aurum and your forex drops from 3.5% to 2%. On the same ₹12.2L stack:

Saved: 1.5% × ₹12.2L = ₹18,300/year

Most premium cards have a ₹2,500–₹10,000 annual fee. Net of that, you pocket ₹8,000–₹15,000/year just from the forex delta. Then you also collect 1.5–3% rewards on the spend, which adds another ₹15,000–₹35,000.

The full breakdown is on our best-for-SaaS card guide.

Path 2 — Specialised forex card (0% forex)

Cards like Niyo Global, RBL Insignia Preferred, or some HDFC Marriott variants carry 0% forex markup. On the same ₹12.2L:

Saved: 3.5% × ₹12.2L = ₹42,700/year

The catch: these cards usually have weaker rewards on Indian spend (or no rewards at all on forex). They're a SECOND card you carry just for USD subscriptions. For an AI startup with high USD burn and low INR domestic spend, this is the strict optimal path.

Path 3 — US LLC + US card

If you have US revenue routing through a US LLC + Mercury / Wise account, get a US business card (Chase Ink Business Preferred, Amex Business Gold). Pay USD with USD — zero forex friction. You also pick up category bonuses on the US card's "software" reward category.

This is overkill until your US ARR crosses $100K/year. Below that, the LLC overhead (Delaware franchise tax, registered agent, BOI filings) eats the savings.

What 3.5% × N years compounds to

The compounding effect, at typical AI startup growth, is alarming:

| Year | Monthly USD burn | Annual forex tax (3.5%) | |---|---|---| | Year 1 | $1,200 | ₹42,700 | | Year 2 | $3,000 | ₹107,000 | | Year 3 | $7,500 | ₹268,000 | | Year 4 | $20,000 | ₹714,000 | | 4-year cumulative | — | ~₹11.3L |

Eleven lakhs lost to forex in 4 years. By Year 4, the forex tax alone is enough to fund a junior engineer.

What to do this week

Pragmatic action list, ordered by leverage:

1. Pull last 6 months of card statements. Add up the international transaction fees. (Most card apps show this filtered.)
2. Apply for a 2%-forex premium card if your spend justifies the ₹2.5K–₹10K annual fee. The best-for-SaaS card page ranks them.
3. Run the math on a Niyo Global / RBL Insignia for 0% forex — useful if your forex spend exceeds ₹3L/year.
4. Set up notifications on USD purchases. Most premium cards send realtime SMS with the forex line broken out — surfaces the tax visually.
5. For new tools, default to checking the billing currency. Razorpay and Cashfree bill in INR. Stripe bills in USD. Pick the one whose currency aligns with your card if features are equal.

The card-side answer

We built a smart matcher that takes a SaaS tool and surfaces the right card to pay it with. Three tiers:

• For INR-billed SaaS (Razorpay, Zoho, Tally) — business cashback cards
• For USD-billed SaaS (Anthropic, OpenAI, Vercel) — low-forex Indian premium cards
• For US-issued cards — for founders with a US LLC

It runs the same scoring math we just walked through, on every tool we track. Every tool detail page shows the recommended card to pay it with, with annualised savings.

The honest take

The 3.5% retail forex markup is a default that costs Indian businesses real money once they have any meaningful USD subscription burn. AI startups hit that threshold in month 3 — Anthropic + OpenAI + Cursor alone routinely cross $400/mo by week 6.

The fix is a 30-minute card application and a single statement-rebill. The math is one-sided. The only reason most teams don't do it is that they haven't bothered to add up what 3.5% becomes when compounded across 60+ tools and 12 months.

Now you've added it up. Go switch.

---

Related reading:

• The 2026 AI startup stack — what to build on, with monthly burn modelled
• Best credit cards for SaaS bills — INR + USD + US tiers
• SaaS FX cost guide for Indian businesses — the deeper dive
• Forex calculator — model your specific stack