Text | Kaori

Mo is the head of a transfer station for Claude. In less than two months, she has served a group of customers both domestically and internationally who have high demands for the performance of the Claude model, with 80% being domestic and the rest scattered across Europe, Latin America, South Asia, and the Third World, "but there are none in the United States."

She has also shared her observations about this business on her website and social media accounts, allowing some users who were previously unaware of the transfer station to understand how this gray area operates for the first time. The transfer station is a central role that is repeatedly calibrated between platform risk control, technical capabilities, and customer trust.

Mo describes herself as not a typical transfer station head; she does not engage in price wars and only provides high-quality service to B-end clients. She has a controversial opinion that the prerequisite for this business to exist long-term is the lasting dominance of American models. However, she is also not optimistic about AI overall; even with a 40% discount based on regional pricing, high-quality intelligent resources remain a heavy burden for many people in Third World countries.

Yesterday, the news came out that Claude would implement passport KYC verification. At first, Mo judged that this would not greatly affect the transfer station industry, but just one day later, the chill from the KYC spread. What she was more concerned about was that this new regulation might push more ordinary users toward transfer stations, while the cost for ordinary users to identify reliable channels is too high, which may not necessarily be a good thing.

Here are Mo's account:

Yesterday, when the news came out about Claude needing to do passport KYC verification, many thought that the transfer station would be completely out of business, but Claude's current KYC is randomly triggered. I initially believed that my business would not be significantly affected.

The next day, I still found that this had a negative impact on our business; the overall cost of the account pool would rise, and the stability of the channels would also be affected to some extent.

Whether it’s stockpiling accounts or finding KYC suppliers, there are many countermeasures for the entire transfer station business. Of course, this is just the current situation; it’s uncertain whether it will extend to all users later.

Claude conducting KYC is its own choice; the entire AI field is still in a growth period. Other companies do not yet face the same level of demand or pressure as Claude, so other models have not done KYC, but it is hard to say what the future holds.

The Transfer Station Business

The transfer station business is all about buying low and selling high, extracting account quotas, and selling out within a certain breakeven period to make profits.

The barrier to entry in this field is genuinely low; as long as you have one or two clients, you can start.

I once met someone who sold at a quite high price, and when I checked his website, he had been operating for half a year, and the site was still HTTP. The defense rate against attacks was very low, and customer data could easily be leaked.

Moreover, not everyone chooses to run this business cleanly.

One type would package user call logs; some transfer stations sell at ridiculously low prices to boost C-end volume by packaging backend databases, like the prompts you provide and the model's returned content, to sell to model training companies. He actually profits from this. In comparison, mixing models is considered minor.

The cost for ordinary users to discern these is too high; some transfer stations initially sell genuine products to you, but later on, they mix in some fakes.

Many friends ask me how to filter out reliable transfer stations. First, it's essential to clarify which upstream source the channel originates from. Generally, the highest quality service comes from Claude's service provided by Amazon AWS, followed by services rerouted from Claude's official sources, and then channels like Antigravity which has a good reputation in the industry. If they cannot clearly state where the service is extracted from, it should raise a red flag.

Different channels will affect output results; AWS and official rerouted channels will be very clean. Also, it’s crucial to pay attention to cache hit rates; for example, when you interact with the model three times, whether it can read previously stored information from earlier texts—higher hit rates save you money.

Further, you can test some typical features of official channels, such as native Web Search and multimodal image understanding capabilities. The native ones can do both tasks. Sending the same prompt to both the official model and the transfer station and comparing output quality, answer depth, and reasoning ability can help identify if the transfer station's responses are significantly shallower and faster, which indicates a downgraded version. You can also leverage third-party evaluation software.

In early April, Claude adjusted quotas, and the entire supply became extremely tight, which greatly impacted the transfer station. That period was tough; I could only ask my technicians to build different channels, and we also introduced some credible external channels to meet customer demand.

This experience also helped me understand a counterintuitive way of discernment: if the model's risk control is widespread across the internet, and your pool is dry, customers are generally sympathetic. However, if your pool isn't dry, customers may realize you're mixing fakes. So, being heavily restricted and running out of capacity may actually indicate you're a legitimate player.

In the C-end transfer station business, the competitive strategies to grab customers through price wars are intense. Sometimes they flood your customer group with their own advertisements or play a double act accusing each other, and in more extreme cases, they may attack your website to cause downtime.

The industry has only been around for two or three years at most; for the first two years, as far as I know, it was relatively easy, and there were those who got rich overnight. Now it has become more difficult; on one hand, there are more competitors, and on the other, risk controls have increased.

So many people this year just want to make a quick buck.

I’ve seen the most outrageous case where someone sold at the original price of $1.80, discounted to $0.80, still claiming it was authentic. Ultimately, users found discrepancies by the evening, tried to argue with him, and then there was no choice but to run away. I also know friends who serve a lot of individual customers; they’re increasingly finding it troublesome and are thinking of pulling out in a couple of months.

"But There Are None in the United States"

I have been following this industry for almost a year now and have joined different circles, but I've actually only been running a transfer station for less than two months.

Clients are from both domestic and international markets, with about 80% being domestic. I primarily work with Claude, also providing GPT versions 5.4 and 5.3. Clients coming in based on these two models have higher demands for performance and functionality, with many being purely development teams.

Initially, due to OpenClaw’s explosion in popularity, a wave of inexperienced users interested in Agent came in. However, due to various costs and usage habits associated with Claude, this group of users gradually disappeared, leaving behind developers, research teams, and some B-end companies.

I also do SEO and GEO, so I have numerous overseas developers reaching out for orders from Europe, Third World countries, Latin America, South Asia, but there are none from the United States.

There was even an owner of a surf shop in Bali who was curious and asked how we sell this. He is a white man and wondered why Claude was so expensive. The whole world knows that Claude is very useful, and the whole world is puzzled by Claude's pricing. Claude is an excellent model for people globally, but aside from enterprises or firms that can provide stable token supplies, not everyone can afford it.

What touches me the most are clients from some Third World countries, like India and Iran.

Based on their monthly salary levels, Claude's pricing is quite a burden for them, forcing them to look for cheaper channels.

They directly ask if there’s Sonnet. The price difference between Sonnet and Opus might be two to three times; they naturally focus on more cost-effective channels and even look for relatively cheaper models on top of that.

Not long ago, there was an Iranian client; they were blocked due to war, requiring intelligent AI to assist with many tasks, even risking arrest by the government. During our communication, I felt that the war seemed not so far away from me.

Previously, we imagined conducting overseas business as primarily dealing with relatively developed countries, but in fact, many Third World clients are facing such difficulties and also want to use AI, so they are forced to turn to transfer stations.

No one knows who Anthropic truly aims to prevent with this KYC measure—transfer stations, distillation attack laboratories, or users in restricted areas; I feel it encompasses all.

However, my feelings about this matter are complex. On one hand, it blocks the way for many ordinary users, who genuinely need Claude and can only turn to transfer stations, which is somewhat beneficial for the business. On the other hand, ordinary users find it very hard to find reliable channels, and the market may be disrupted by transfer stations that purely sell fakes.

To be honest, this is not a positive development. Users with the ability to maintain Claude subscriptions have high demands for stability. Spending time discerning whether there's fake content may lead to disappointment regarding transfer stations.

Reverse Transfer Station

I don’t think of myself as a typical transfer station head.

The existence of transfer stations is primarily due to information asymmetry; as long as there are customers who consistently demand cheap APIs, whether domestically or internationally, this business will continue to exist.

I have a controversial viewpoint: if American models maintain their strength, transfer stations will continue to exist because they will always be expensive.

The USD exchange rate is there; training the same model costs $10 million in the United States but only ¥10 million in China—this price differential will always exist.

However, if Chinese models truly rise to a level strong enough to surpass American models, then the transfer stations will really be out of business since the acceptance of Chinese models will undoubtedly be higher than that of American ones.

This means that the long-term existence of this business presupposes the enduring dominance of American models.

From a business perspective, it has low entry barriers that anyone can enter, and the ceiling is high; one could even say it "mutually enables" the platform because the more locked down the platform becomes, the less likely the transfer station will disappear.

But I'm not optimistic about AI as a whole.

I previously imagined that with the arrival of the AI era, many tedious physical, basic mental labor, and dangerous jobs would be resolved by machine intelligence, leaving humans time to focus on the spiritual world and on themselves.

But then I found out that even though many services are already priced regionally, such as Claude models costing only 60% in Africa, 60% is still a significant expense for local people.

In these regions, those who use AI are like people with wings, and those who do not use AI are just walking on the ground. This disparity only serves to widen the gap between the rich and poor, concentrating more wealth in the hands of the wealthy.

Although open-source models are advancing, very few can be deployed on personal devices to serve daily life.

The future of large models may be an era that is both open-source and locked down. My goal is to help everyone equally access better intelligence; this is a serious issue.

What I aim to do long-term is to take some open-source models or relatively high-cost-performance models in China overseas, directly providing them to some Latin American countries and underdeveloped European countries, and then moving on to South and Southeast Asia.

In principle, AI should be accessible to everyone. From a business perspective, I will continue to run the transfer station, but I will also be inclined to pursue some genuinely open-source initiatives outside of this business. This is what I can see and what I can do.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。