On August 16, American payment giant Stripe announced the completion of its acquisition of the AI model aggregation platform OpenRouter, with the transaction amount exceeding $7 billion—just three months ago, the company's financing valuation was only $1.3 billion, a five-fold increase in just one quarter.
OpenRouter's founder Alex Atallah did something similar last time with the NFT trading platform OpenSea.
This is not his first time pinpointing so accurately. When he and Devin Finzer founded OpenSea in 2018, the term NFT had yet to enter the public consciousness; when he founded OpenRouter in 2023, the issue of "too many AI models and developers needing a unified interface” was also not widely recognized in the industry. Both times, he built the infrastructure before the market had formed a consensus.
Among those emerging from Web3, nearly everyone is repeating the same action.
Exactly what step was taken early
The problem OpenSea solved: the NFT trading was extremely fragmented, scattered across various independent platforms, with no unified entry point. This judgment was counter-consensus in 2018—at that time, the entire NFT market's trading volume could be less than tens of thousands of dollars a day.
The problem OpenRouter solves: an explosive growth of AI models, with inconsistent interface standards among vendors, locking developers into single suppliers. This judgment was also counter-consensus in 2023—most developers assumed that "picking one model supplier to use completely" was the normal state.
In both instances, Atallah was not solving a problem that had already been validated, but rather anticipating that "fragmentation will eventually be aggregated" would happen, then built the bridge first, so when others realized they needed to cross the river, the bridge was already there. Stripe's willingness to pay $7 billion this time is for not an API gateway, but for the foresight of that judgment which has already been validated once.
They excel at seizing entry points
Kris Marszalek, co-founder and CEO of cryptocurrency exchange Crypto.com, took a different path, but the core remains the same.
In April 2025, he spent $70 million to purchase the AI.com domain, settling in cryptocurrency, setting a public record for domain transactions. On the surface, this is just an expensive domain transaction, but when viewed within the history of Crypto.com, it fits Marszalek's consistent thinking—spending heavily on domain purchases, stadium naming rights, and Super Bowl ads has fundamentally been about seizing user entry points.
In the past, when users opened the exchange, it was to buy coins; in the future, when users open an AI Agent, it may be to book flights, handle emails, or complete payments. Marszalek is betting not just on a URL, but on who will become the new digital entry point once AI evolves from "answering questions" to "executing tasks for users"—this judgment was also ahead of the market at the time, while most people were still discussing the usability of chat boxes, he was already scrambling for the address of an emerging track.
From Web3 to AI, he did not change his competitive method; he merely shifted his past experience of competing for wallets, exchanges, and traffic entry points to the Agent entry.
This is not the first time doing this
If Atallah and Marszalek represent this round of migration, then Emad Mostaque, founder of open-source AI company Stability AI, actually walked this path much earlier.
Before entering AI, he had long focused on Bitcoin and Ethereum, and in 2019 he launched a project called Symmitree, aiming to use blockchain to reduce the barriers to accessing digital technology in impoverished areas, as hospitals, governments, and tech companies were unwilling to open data, leading to failure. This failure did not make him abandon his judgment; instead, it led him to conclude that AI models would inevitably move towards an open-source versus closed-source route.
When he released the open-source image generation model Stable Diffusion in 2022, the mainstream narrative in the industry was still that "top AI models must be controlled by a few giants," and this logic had not yet been truly broken. What Stable Diffusion did is highly similar to some of the early ideas of Web3: opening capabilities that were originally concentrated in a few institutions to developers and communities as much as possible. The models can be downloaded, developers can make secondary creations, and the community can continue to build around the models—he was the first to truly break the moat of commercial closed-source models.
This indicates that "getting there early" is not merely a coincidence that arose in this AI cycle; it is a behavioral pattern that this group has repeated at least twice. And this pattern quickly became not just the property of founders.
Two types of people emerging from the same vanished institution
Avital Balwit, current CEO office director at Anthropic, offers another version of the story.
Previously, she worked at SBF-funded FTX Future Fund, responsible for screening and assessing long-term projects that had not yet been mainstream recognized but could impact the future of humanity. AI safety was one of the few focal directions at that time (FTX had previously invested $580 million in Anthropic). The fund ultimately disappeared with the collapse of FTX, but the training of "predicting which areas would become the next key variables" did not disappear. Balwit later joined Anthropic, becoming an important decision-making member alongside CEO Dario Amodei.
Leopold Aschenbrenner, who also came from FTX Future Fund, took a different path.
He joined OpenAI's "super alignment" team, and later became one of the most followed young researchers in the AI field with a 165-page essay titled "Situational Awareness," and because he bet on the AGI trend early, the investment circle nicknamed him "the new stock god of AI."
One entered the highest levels of an AI company, one entered the AI investment market, and a previously bankrupt institution has dispatched completely different types of people to the AI industry. The significance of this is larger than "FTX once invested in Anthropic": it indicates that what the Web3 bull market left behind was not just protocols, exchanges, and tokens, but a group of talents who have experienced rapid growth, capital frenzy, and industry collapse—they are familiar with an environment where technologies are not yet mature, rules are not yet defined, but capital has already begun betting. AI has just entered such a phase.
Judgments were correct in direction, but...
Aschenbrenner used the fame of "Situational Awareness" to fundraise and establish a hedge fund of the same name; from July 2024 to June this year, net returns exceeded 439%, and the fund scale even surged to $45 billion, proving that his judgment on the AGI direction was indeed early and correct.
However, in July this year, the Situational Awareness fund suffered a decline of over 35% in a single month, heavily invested in semiconductor and AI infrastructure stocks with 400% leverage, as Goldman Sachs, JPMorgan Chase, and Bank of America issued simultaneous margin call notices, forcing him to sell about $16 billion of publicly held positions to hedge fund giant Citadel (founded by Ken Griffin) at about a 10% discount; the $45 billion dropped to only about $10 billion within a month.
A person who can accurately judge the long-term trends of AI can still encounter serious failures in the short-term capital market—this is not contradictory. Recognizing the correct technical direction and making money from this direction have never been the same thing. Judgment is responsible for answering "what will happen," while leverage and position management are responsible for answering "how much to bet, can it hold until the day of verification"; these are two completely independent abilities, the former is a shared talent of this group, while the latter is the real variable that determines success or failure. The experience of Stability AI also illustrates the same principle: those who saw the potential of open-source AI early are not necessarily the final business winners—Mostaque himself also left the CEO position due to internal disputes in early 2024.
So it seems they are not better at predicting answers, but rather more accustomed to placing bets before the answers have emerged.
What is truly being exported
The real abilities of this group transferred from Web3 to AI can be broken down into three specific sensitivities:
The first sensitivity is to infrastructure gaps—when NFTs exploded, Atallah saw the trading infrastructure; after the explosion of large models, he saw the routing and aggregation between models.
The second sensitivity is to new entry points—Crypto.com is seizing trading and user entry points, while AI.com is seizing the entry into the Agent era.
The third sensitivity is to opportunities before consensus is formed—when FTX Future Fund was researching AI safety, AI was far from becoming today's mainstream capital narrative; when Stable Diffusion was launched, open-source models were still not receiving the attention they have today.
These three sensitivities point to the same thing: they are not predicting a specific hot trend, but identifying what the next round of the market will lack. Web3 kept this group long in an environment where rules were not yet mature, business models were constantly rewritten, and technology and finance were highly mixed, pushing them to repeatedly experience a complete cycle—concept emergence, capital influx, infrastructure explosion, business model competition, bubble burst, and remaining talents searching for the next poker table. When AI enters a similar phase, they already know where to look in advance.
Aschenbrenner's liquidation precisely confirms the boundaries of this ability: it can tell you where the direction is, but it will not decide how much effort you should put into betting on that direction.
AI has yet to complete its first round of reconstruction, and the real opportunities often do not appear until everyone has seen them. The next batch of industry changers may have already begun searching for the next unformed blank space.
*The content of this article is for reference only and does not constitute any investment advice. The market carries risks, and investments should be made cautiously.
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。
