The Person Who Exited at the Peak of NFTs Is Now the Most Hidden Winner Behind OpenClaw

Author: David, Deep Tide TechFlow

OpenClaw is hot, but among this wave of enthusiasm, the company quietly making money is one you might not have heard of:

OpenRouter.

Using OpenClaw requires connecting to various AI models to do work, with Claude, GPT, DeepSeek each having their own fees and interfaces. OpenRouter’s job is to bundle these models together, so you can use them through a unified platform, earning a margin from the difference.

The person doing this business is Alex Atallah. His company just raised $40 million led by a16z, now valued at $500 million.

What you might not know is that his previous company was OpenSea, the world’s largest NFT marketplace, which at its peak was valued over $13 billion.

However, he chose to exit at the height of the NFT craze, just a few months before the NFT market collapsed.

Now, he’s making money again in this AI wave.

From liquidity aggregation to large model aggregation

Alex Atallah graduated from Stanford’s Computer Science Department.

In 2018, he co-founded OpenSea with Devin Finzer. The business was simple: when others mint NFTs, they provide a place for buying and selling, taking a 2.5% fee per transaction.

OpenSea doesn’t produce NFTs or trade them; it just provides a marketplace, aggregating liquidity.

In 2021, during the NFT boom, popular NFTs like Bored Apes became cultural symbols. OpenSea’s monthly trading volume once exceeded $5 billion. Forbes estimated that he and Finzer’s combined wealth was $2.2 billion.

In July 2022, he resigned as CTO, saying he wanted to build something new.

What happened afterward is well known: the NFT market collapsed, the market entered a deep freeze, and OpenSea’s business suffered. But there are always those who buy the feast, and Alex exited before the music stopped.

In 2023, he started building something called OpenRouter. Simply put:

A large model aggregation routing platform that places hundreds of model APIs behind a single interface, allowing developers to call them with a 5% fee per use.

You might ask, why not just directly use OpenAI or Anthropic, call Claude or GPT?

Of course, you can.

But now, most people don’t use just one model—coding with Claude, searching with Gemini, saving money with DeepSeek, registering separately, recharging separately, and interfaces are all different…

Not to mention many users want to use both Claude and GPT, but can’t connect directly from China.

So, OpenRouter is the easiest route. One interface, over 500 models, unified format, automatic switching, all handled with one key.

You might not have noticed when using OpenClaw, but the default provider in the configuration file was previously OpenRouter.

Source: Zhihu user Feng Control Alchemist

When you call Claude or DeepSeek, the request first goes to OpenRouter, then forwarded to the model provider. Even OpenClaw’s documentation states:

If your API key format isn’t recognized, it defaults to OpenRouter.

How fast is this business growing?

By October 2024, the monthly revenue through OpenRouter is $800,000. By May 2025, this jumps to $8 million.

Tenfold in seven months.

In a year, the total money passing through is over $100 million. He takes a 5% cut, earning $5 million, with a team of fewer than ten people.

Source: sacra.com

a16z used his data to write an industry report titled “The Current State of AI with Trillions of Tokens”; Stripe even customized a billing system for him.

With the explosion of OpenClaw this year, more developers and enthusiasts are jumping in, trying all kinds of tokens, which inevitably increases the demand for calling various large models, further fueling OpenRouter’s business.

Moreover, with a16z leading the investment, the valuation is $500 million.

Once again, a seller of shovels becomes a seller of shovels.

Different hot topics, same model

Careful observation shows that Alex’s two businesses are structurally similar.

OpenSea’s business is not minting NFTs but providing a platform for others to mint NFTs, where buyers and sellers trade, and he takes a 2.5% fee. OpenRouter’s business is not training models but hosting models trained by others, where developers call them, and he takes a 5% fee.

This approach seems to be his comfort zone. Whether in NFTs or AI, the market structure is very similar:

Supply side is highly fragmented, demand side buyers don’t know where to find supply, and he stands in the middle as a marketplace.

How fragmented was NFT in 2021? Dozens of chains, hundreds of project teams, tens of thousands of new series daily. If you want to buy a Bored Ape, you can’t visit each project’s official website. OpenSea aggregates them, so you browse and buy, and the counterparties sell to you.

How fragmented are large models in 2025? OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral, Zero One, and countless others… just the mainstream ten-plus companies, plus hundreds from open source communities.

Today, you might code with Claude, tomorrow Gemini releases a new version with better search, the day after DeepSeek drops its price by half. Every switch requires changing interfaces.

Atallah once said very clearly:

“OpenSea consolidates highly fragmented inventories into one place, and AI today looks very similar to that.”

He doesn’t need to know which NFT will rise or which model will win. He only needs to know one thing: the more fragmented the supply, the more valuable the middleman.

And interestingly, the timing.

When he left in July 2022, OpenSea’s valuation was still high. NFT monthly trading volume had fallen from its peak, but no one thought it was going to collapse. He said he wanted to “start from zero and build something new.” Six months later, ChatGPT was released, and the large model era began.

What did he see, or was it luck?

I don’t know. But one thing is certain:

When he registered OpenRouter in early 2023, there were almost no AI large model routing products on the market. By the time everyone realized the need for a unified interface, he was already there.

The last time, he did the same in the NFT space. When everyone rushed in, he was already the biggest platform.

Is it more important whether AI is hot?

In every wave of enthusiasm, most people ask: what will be popular?

In 2021, which NFT will rise; in 2024, which meme coin will multiply; in 2025, which AI app will emerge; in 2026, what can small lobsters do.

Atallah’s questions might be different. I think his thinking is: regardless of what gets hot, where is the money flowing from?

These two questions may seem similar, but they are fundamentally different bets.

Betting on “what will be hot” means guessing right once. Bored Apes will rise, PEPE will multiply, a certain AI product will be the next ChatGPT. If you guess right, you get rich; if wrong, you lose everything. Most people’s experience is the latter.

Betting on “where the money is flowing from” doesn’t require guessing any one thing. If NFTs rise, transactions happen on OpenSea, and he earns fees. If AI model wars intensify, developers need a unified interface to switch back and forth, and OpenRouter gets busier.

It’s not about who wins but about betting that this battle will last a long time.

Looking back, the platforms that make the most money in each cycle are usually those at the core of the ecosystem.

The prospectors come and go, but the water sellers keep collecting fees.

But I think simply saying “selling water” or “selling shovels” isn’t enough. Many shovel sellers have failed. Atallah succeeded because he did something more specific: he always focuses on the aggregation position.

You can’t just build a tool and collect tolls. You need to be the one who consolidates the fragmented supply. The more dispersed the supply, the higher the switching costs, and the more pricing power the aggregation layer has.

This also explains why he entered early both times. Because the aggregation business has a characteristic:

The first to arrive signs the supply, making it hard for later entrants to catch up.

So, Atallah’s key insight can be summarized in two sentences:

First, don’t guess who will win, just find the intersection everyone must pass through. Second, before others realize they need an intersection, you’ve already built the road.

Smart people never pick a table

Now, I notice two dominant voices around me.

One says AI Agents are toys; installing OpenClaw is just burning tokens and has no real use. The other says this is another wave of AI hype, and in three months, no one will remember.

Maybe both are right.

But for someone like Alex Atallah, it doesn’t matter.

Whether OpenClaw is useful or not, he’s collecting money. If you think Lobster is boring and stop using it, the tokens burned in the past two weeks are already in his hands.

Some think NFTs are dirty, Ponzi schemes, scams. He built a company valued at $13.3 billion on that. Others see AI Agents as bubbles, hype, with no clear business model. He built a company valued at $500 million on that…

Geniuses probably don’t need us to respect the track they’re on.

He made money on the NFT table. He’s making money on the AI table. What the next table will be, no one knows.

But I guess, when the time comes, he’ll still be collecting tickets at the door.