Frontier models for the half of the world without a credit card

There is a quiet line that cuts the AI world in half. It is not drawn by GPU supply, not by language, not even by skill. It is drawn by whether the developer has a Visa or Mastercard that the issuing model provider will accept, and it removes roughly half the planet from the front-line of frontier AI.

You can confirm this in five minutes. Open the signup page of any major US-headquartered model provider and pretend to live in Karachi, or Moscow, or Jakarta. The card form will accept your card number, then quietly fail at the address verification step, because your bank issued a card that doesn't have a US-routable AVS endpoint. Or it will silently 3DS-decline because the provider's risk model treats your IP as adversarial. Or — most often — your bank simply refuses the international micropayment because it sees a transaction it doesn't understand and assumes fraud.

You did not make a mistake. You are doing nothing wrong. The product just isn't for you.

Who this actually affects

Five markets we hear from constantly:

• China. A billion-plus people, an enormous developer talent pool, and a payment system that does not interoperate with international card schemes for AI services. Alipay and WeChat Pay solve the domestic problem; they do not solve the "I want to call Anthropic from a notebook in Shenzhen" problem.
• Russia. Cards from Russian banks have been functionally cut off from US-headquartered AI providers since 2022. The technical work happening inside Russian universities and companies hasn't stopped. The access has.
• Indonesia. A vibrant, English-fluent developer scene. Local debit cards rarely clear international AI provider checkout. Pre-paid international cards are a workaround that fails about 40% of the time.
• Pakistan. Similar shape to Indonesia, with the additional friction that international remittance and FX volatility eat the top off any developer's budget before the model even runs.
• Saudi Arabia and the wider Gulf. The cards work more often. The corporate-procurement requirements (W-9 forms, US tax IDs, USD invoicing) increasingly do not.

These are not edge cases. Together they represent more than two billion people, a large fraction of the developer talent on Earth, and a meaningful share of the people who will build the next decade of AI products.

What "no card needed" actually means

We did not set out to build the AI access layer for underbanked markets. We set out to build a wallet for autonomous agents. But the choice of a USDC settlement rail — forced by the unit economics of fractional-cent inference — turned out to also solve a different, larger, more politically inconvenient problem:

You don't need a credit card. You don't need a bank that the model provider's risk system trusts. You don't need a US billing address. You don't need a working W-9. You need a USDC balance, which you can get on any phone, in any country, in under five minutes.

Once you have that balance, Franklin gives you:

• 55+ frontier models — including the ones you cannot pay for directly from where you live, settled per-call from your wallet.
• Image and video generation, trading data, web search — same wallet, same rail.
• One install. No US-routable card. No SMS verification to a +1 number you don't have. No corporate procurement form.

No terms of service are being bypassed. The BlockRun gateway that Franklin runs on top of is an authorized customer of the underlying providers, paying them in USD via standard rails. What Franklin lets you do is trade USDC for that already-paid-for capacity, in cents, on-chain, from anywhere a phone can run a Node.js binary. The international payment friction collapses into a wallet you already have.

The asymmetry that no one names

There is a polite version of this conversation in which everyone agrees that of course frontier AI should be globally accessible, and then nothing changes for years.

Here is the impolite version:

The provider risk model that declines a Pakistani Visa is not "broken." It is operating exactly as designed. From the provider's perspective, the marginal cost of accepting that card is non-zero (chargeback risk, fraud risk, support load) and the marginal revenue is small (a developer doing $10/month of API). So the math says: decline. The math is right, locally. The math is also catastrophic, globally — because the long-run revenue from those developers, once they are building products, is enormous, and every year they cannot pay the provider is a year they go to a competing local stack instead.

A USDC rail breaks the math. Chargeback risk: zero. Fraud risk: bounded by the wallet balance. Support load: handled by us, not the provider. Marginal cost: cents. The provider says yes through Franklin where they have to say no through Visa. Everybody wins, including the provider, who now has a developer in Karachi paying them $30/month instead of zero.

What we see in our analytics

We launched without picking a target geography. The user base self-selected. Today, our 5M+ API users are spread across more than 50 countries, and the growth distribution is not what a US-marketed product looks like. The strongest organic signals are exactly the markets above plus a long tail of others — Vietnam, Turkey, Iran-adjacent diasporas, Brazil, Argentina, Nigeria, Egypt.

The pattern is the same in every one of those markets:

1. A developer hits a card-decline wall on a frontier provider.
2. They go searching for "how to use [model] without a credit card."
3. They find Franklin (often via a forum post in their own language).
4. They install, fund $5 in USDC, and run the call that didn't work yesterday.
5. They tell three friends.

That is not a marketing strategy. That is a distribution strategy that happens for free as long as we keep the install pleasant and the wallet honest.

What this means for the next year

If you live in any of the markets above and you are reading this in English, you already know how to get a USDC balance and you already know how to install a CLI tool. We do not have to teach you. We have to show up in your local-language search results when you hit the card decline.

So we are doing two things:

1. Translating Franklin's documentation, blog, and onboarding into the languages of the developers who are actually finding us. This post is the first in English. There will be Mandarin, Russian, Bahasa Indonesia, Urdu, and Arabic versions, and they are not machine-translated drop-ins — they are written by people who know the local pain.

2. Making the wallet experience native. A developer in Karachi should not have to learn what "USDC on Base" means to use Franklin. They should have to learn what "load $5 onto your wallet" means, and the wallet should handle the rest. We are most of the way there. We will be all the way there by the end of the quarter.

If you live somewhere where your card just doesn't work for AI, you are the user we built this for. Install Franklin, set up your wallet, and try the call your card declined yesterday. The geography of inference is changing. There is no reason it has to map to the geography of Visa.

If you are a developer in one of the markets named in this post and you'd like to help translate Franklin's docs into your language, we pay in USDC. Email care@blockrun.ai with a sample of your work.