From OpenSea to OpenRouter: A Serial Entrepreneur's "Aggregation" Business Strategy

Author: David, Deep Tide TechFlow

Original Title: The person who exited at the NFT peak is now the most secretive winner behind OpenClaw

OpenClaw has become popular, but in this wave of hype, the quiet money-maker is a company you might not have heard of:

OpenRouter.

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

The person behind 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, and a few months later, the NFT market collapsed.

Now, he’s making money again in the AI boom.

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 platform for buying and selling, taking a 2.5% fee per transaction.

OpenSea doesn’t produce NFTs or speculate on 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 crashed, entering a deep freeze, and OpenSea’s core business suffered. But someone always pays the bill, and Alex exited before the party ended.

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

A routing platform for large models, integrating hundreds of model APIs behind a single interface, with developers calling it and paying a 5% fee each time.

You might ask, why not just directly contact OpenAI or Anthropic to access Claude or GPT?

Of course, you can.

But now, most users don’t stick to just one model—coding with Claude, researching with Gemini, saving money with DeepSeek—registering separately, recharging separately, with different interface formats…

And 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 managed with a single key.

When you use OpenClaw, you might not notice that the default provider in the configuration file was previously OpenRouter.

Image source: Zhihu user Feng Control Alchemist

When you call Claude or DeepSeek, the request first goes to OpenRouter, then is 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 passing through OpenRouter is $800,000. By May 2025, it hits $8 million.

Seven months, tenfold increase.

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.

Image source: sacra.com

a16z used his data to write an industry report titled “The Current State of AI with $100 Trillion Tokens”; Stripe even customized a billing system for him.

With OpenClaw’s explosion this year, more developers and enthusiasts are jumping in, trying all kinds of ways to burn tokens, which has inevitably driven demand for calling various large models—completely fueling OpenRouter’s business.

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

Once again, a tool seller has become a major player.

Different Hotspots, 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:

Extremely fragmented supply, 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 every project’s official site. OpenSea aggregates them all, letting you browse and buy, and sellers list their prices.

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

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

Atallah once explained this logic 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 matters.

In July 2022, when he left, OpenSea’s valuation was still high. NFT monthly trading volume had fallen from its peak, but no one thought it would 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.

Did he see something, or was it luck?

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

When he registered OpenRouter in early 2023, AI large model routing products were almost nonexistent. 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 about AI being hot, or is it important?

In every hype cycle, most people ask: what will be the next big thing?

In 2021, which NFT will rise; in 2024, which meme coin will multiply a hundred times; in 2025, which AI app will emerge; in 2026, what can Xiaolongxia do?

Atallah’s questions might be different. I think his thinking is: no matter what gets hot, where will the money flow from?

These two questions seem similar but are actually completely 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 flows from” doesn’t require guessing any one thing. When NFTs rise, transactions happen on OpenSea, and he earns fees. When 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.

Prospectors come and go, 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 by doing something more specific: he always focuses on the aggregation layer.

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

This also explains why he entered early twice. Because in aggregation business, the first to arrive signs the supply, making it hard for later entrants to catch up.

So, I summarize Atallah’s genius in two sentences:

First, don’t guess who will win—just find the intersection everyone must pass through. Second, build the road before others realize they need it.

Genius never chooses the table

Now, I hear two dominant voices around me.

One says AI Agents are toys; with OpenClaw, they only burn tokens and have no real use. The other says it’s just another wave of hype, and in three months, no one will remember.

Maybe both are right.

But for someone like Alex Atallah, he doesn’t care.

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

Some think NFTs are dirty, a Ponzi scheme, a scam. 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.

Genius may not need us to respect the track he’s on.

He made money in the NFT space. He’s making money in AI now. What’s next? Nobody knows.

But I guess he’ll still be collecting tickets at the door.