The Web3 + AI Daily: On Stripe, Bittensor, and Distributed AI Training
Stripe shifting from payments to multiproduct offering, Bittensor fighting the AI-fueled neo-fascism, and Europe embracing distributing AI training.
Hello, everyone, and welcome to all newly joined subscribers!
This is the 82nd edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI. Today’s edition covers topics that may seem diametrically opposed, but are actually deeply interconnected. AI is transforming businesses, industries, and the very fabric of our societies. We must act quickly to find viable ways to preserve our values and democracy, and not be afraid to try out unorthodox suggestions.
Thank you for being here! Let’s dive in.
What’s Hot in Web3 + AI?
Stripe Shifts From Payment Company to Multiproduct Platform
The tech world is still buzzing with excitement about Stripe‘s acquisition of OpenRouter, so today I want to add a bit more nuance. Here’s why this deal is so interesting.
In its essence, OpenRouter has the same business model as Stripe, namely acting as an intermediary and charging a percentage of every dollar flowing between the transacting parties as a fee. Yet, while the payments processing business is quite saturated, the AI model routing one has a serious growth potential, at least in short term (see below), which may explain the price of $7B Stripe is willing to pay.
As Aakash Gupta wrote on X:
Stripe just paid over $7 billion for a company doing an estimated $50 million a year in revenue. A 140x multiple sounds insane until you look at how OpenRouter actually makes money.
It passes model prices through untouched and earns a 5.5% fee when customers load credits to spend on inference. Stripe’s own core product charges 2.9% plus 30 cents to move money. OpenRouter collects nearly double Stripe’s rate on every dollar flowing to OpenAI, Anthropic, Google, and 400+ other models.
Stripe knew these numbers better than anyone. OpenRouter runs invoicing on Stripe, calculates taxes with Stripe Tax, and screens fraud with Stripe Radar. Stripe has been processing this company’s money for years, watching the volume curve inside its own dashboard.
That curve is steep. Inference spend through OpenRouter ran at $10 million annualized in October 2024, crossed $100 million by May 2025, and now implies close to $1 billion a year. The platform moves around 100 trillion tokens a month for 8 million users.The deal also completes a pattern.
Stripe closed its Metronome acquisition in January to get usage-based billing. In April it shipped per-token streaming payments and wallets for AI agents. Now it owns the marketplace where developers actually pay for models.
Stripe built the toll booth for internet commerce and takes 2.9%. It just spent $7 billion on the toll booth for AI compute, and that one takes 5.5%.
Just like Stripe doesn’t care about which retailer has the better product, OpenRouter doesn’t care about which AI model is more powerful or performant. The more models exist and the more customers are incentivized to switch between them, the better:
Moreover, Stripe’s business strategy and value proposition at large is changing, as William Gaybrick, President of Product & Business, clarified in a conversation with Andreessen Horowitz:
We began as a payments company with add-ons to now being a multiproduct platform built around financial infrastructure.
Everything we build is designed to help businesses grow by reducing friction and increasing their agency—to adapt their business models, operate in more countries, and move faster across everything that touches revenue and cash.
But what's the rationale for OpenRouter to sell? As I told you yesterday, the company's co-founder, Alex Atallah, was also the co-founder of OpenSea - another aggregation layer that made fragmented, hard-to-navigate supply easy to buy. Back then it was NFTs, and now it's AI models.
Neither of the two companies produces anything, or bears responsibility for the quality of the product. Rather, both offer discoverability, price comparison, and efficiency.
However, as Jesse commented, OpenRouter’s business model creates a potential paradox:
Low substitutability plus high buyability produced OpenSea: a market in assets that could not replace one another, with room for a 2.5 percent toll.
High substitutability plus high buyability produces something closer to a commodity exchange. Commodity exchanges are thin by nature.
This creates the central contradiction in OpenRouter’s model. The better it becomes at making models interchangeable, the less any individual model is worth defending. Buyer loyalty shifts from the underlying model toward price, performance, and availability.
But it does not necessarily shift toward the router.
The aggregator manufactures elasticity, then discovers that its own margin moves inversely to the elasticity it created.
The better OpenRouter succeeds at making models fungible, the harder it may be for the router itself to sustain high margins. The same structural forces that make aggregation powerful can eventually commoditize the aggregator’s economics. That said, Atallah may turn out to have a remarkable ability to leave the building right before it collapses.
Web3 + AI Readings & Conversations
Can Bittensor Beat OpenAI?
The $7B acquisition of OpenRouter by Stripe has prompted a lot of questions about how a decentralized alternative to OpenRouter could be valued. Since Bittensor’s subnets are arguably that alternative, today I’m sharing an interesting interview with Jacob Robert Steeves (aka “Const”), co-founder of Bittensor / Opentensor Foundation and CEO of Affine (Subnet 120).
In it, Const recounts how he was inspired by Bitcoin and decide to apply its decentralized and permissionless computation mechanism to AI, and thus started Bittensor:
The most powerful computer in the world combined with the most important computational problem of the 21st century.
What’s more, he distills how $TAO’s tokenomics have evolved to accrue value, while also guaranteeing that no more rug pulls would happen:
How is $TAO designed to accrue value? The inflation of TAO incubates the projects on the chain; the projects pay back platform fees to the chain, and then these fees come out as a dividend to the $TAO holders.
When we distribute $TAO inflation over to these projects, they’re also paying us over a horizon with equity in those projects. Currently, it’s about a $100M per year in dividends to about $300M worth of infaltion. So we’re in an inflationary period.
The other aspect of the $TAO tokenomics is that in order to by any subnet token, to access the digital commodity they’re producing, you need $TAO. So the network captures the flow of those tokens through $TAO.
The Conviction update allowed subnet owners to choose to lock their funds and thus signal that they have long-term intentions and avoid rug pulling. It gives alpha token holders control over the subnets.
Most importantly, Const used the occasion to share his vision of the state of AI and where the industry is headed next. His thesis is:
AI is such a powerful and transformative force that we cannot allow for it to be built behind closed doors. It should be built from the ground up, in a way that anyone can join at any time and owns and controls part of it.
I think that we risk, probably for the first time in history, a form of totalitarianism we’ve never seen before. The reason for that is that [throughout history] humans have held a lot of weight against the emperor. If you abuse your population they’ll fight you, and they can win.
What we’re seeing right now is that war is becoming incredibly mechanized, and the weight of the workforce is dramatically changing. You can be a tyrant that loses 90% of your workforce potentially, and it doesn’t even affect your economy. That balance of power between the mass of individuals and the powerholder is drastically changing with AI.
What we can potentially suffer with that balance changing is a form of neo-AI-fascism. It would be some sort of an unholy union of nation states and their tech giants, which we’re already seeing with the Trump administration and the OpenAIs and Anthropics of the world. The administration is playing them against each other to get more control over the economic power of these megagiants, that are going to be trillion-dollar companies.
[We need to] come up with a way to reorganize ourselves into systems that have more transparent and distributed ownership and ground-up production, so that it can’t be yielded purely in the interest of small groups of people to exponentiate the concentration of power. Crypto is uniquely positioned to give us a solution to that problem, because of its ability to create transparency and ownership resistant to military cabal, and its ability to actually produce these digital commodities in a competitive way.
Fortunately, an alternative already exists. Bittensor embodies the essence of an open, permissionless, and transparent AI ecosystem, which is increasingly proving it can compete with the hyperscalers:
We’ve shown that we can beat them in more narrowband problems: mining compute, mining inference, mining storage. It’s only a matter of time before we have not only the largest decentralized models in the world, but the best models in different domains.
Can Europe Rely on Distributed Compute to Train Frontier AI?
Back in July, RAND Europe, a nonprofit, nonpartisan research organization, published a report exploring whether Europe can use distributed training methods to achieve frontier AI capabilities, and catch up the USA and China.
First, I have to say, I feel like yelling “Eureka!” What we’ve been preaching in the dAI space is finally breaking through. As the Covenant AI team commented:
Here’s what the report was all about:
The report concludes that distributed training can indeed offer relief on the power and political-coordination bottlenecks.
This technique cannot give Europe more chips, so it will not close the gap with the US on its own.
I sincerely hope the decision-makers at the European Commission take notice and start embracing innovative solutions that are already available, tested, and working.
Thank you for reading! I hope you found this article insightful.
If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.
And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.
The Web3 + AI Book Club is live on Fable. We’re currently reading Karen Hao’s ‘Empire of AI.’ Follow the link below and come join our literary community.
I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.
Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.
Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.