5 min läsning“The OpenAI governance drama highlighted just how centralised AI power has become”
Summer hadn’t quite kicked in yet in Paris early June, but the heatwave was already bearing down on us near the Louvre, where the Musée des Arts Décoratifs, chosen to host the blockchain conference Proof of Talk, is located. As tourists wandered through the gardens, with too little shade to escape the sun, we headed to meet Ala Shaabana at Hôtel Château Voltaire, a cosy, air-conditioned refuge.
Inside its bar, full of mirrors and famously photographed with Kate Moss, we explored why blockchain might help solve some of the systemic issues AI start-ups are running into, from monetisation to data verification. A neural-like network of more than a hundred subnets, Bittensor aims to become the base layer of tomorrow’s AI. Thanks to its token, TAO, the platform rewards participants for the resources they contribute. A gimmick? Its co-founder is betting on the opposite. And he’s not the only one: despite multiple highs and lows, the token was still well above its initial market capitalisation at the time of writing.
The Node: Ala, before diving into Bittensor, can you tell us a bit about your background?
Ala Shaabana: I finished my PhD in applied computing in 2017, and my background is much more academic than crypto-native. A lot of my early work focused on shrinking AI models so they could run on very small circuit boards, often applied to the human body.
After that, I joined VMware, where I worked on distributed computing systems — not distributed AI per se, but adjacent problems. I later moved to Instacart, and that’s when Bittensor started as a side project before becoming my full-time focus.
The Node: What initially drew you to digital assets and blockchain?
Ala Shabaana: It came from a very concrete technical problem. Together with Jake Roberts-Steeves, we were exploring whether AI models could share knowledge with each other — essentially, collaborative intelligence.
The challenge wasn’t whether it worked technically, but rather incentives. People kept asking: who cares, why would anyone do this, how does it make money? We looked at projects like Folding@Home or SETI@Home. They were distributed, but they lacked proper incentives, and ultimately they failed. That’s when it became clear that blockchain was the missing piece, a way to align incentives so people would actually contribute compute and intelligence.
Many still see the combination of AI and blockchain as a gimmick. Why do you think they work well together?
In a sense, it’s understandable as a lot of projects have been gimmicks. But AI has a fundamental verification problem. Blockchains are very good at verification. That’s where the fit becomes natural. In decentralised systems, you don’t rely on a single authority like OpenAI, Google or Amazon, companies that are increasingly losing trust. Instead, you get a trustless, permissionless environment where work can be verified transparently. In many ways, AI verification is the use case blockchains have been waiting for.
Would it be correct to describe Bittensor as a brain?
One way to think about it is as a brain. Another is as a city. Bittensor is made up of “subnets”, which you can think of as neighbourhoods or specialised regions. Each subnet solves a different problem — storage, inference, training, object generation, defect detection — all incentivised differently. Like a brain, different parts do different jobs, but together they form a single intelligence network.
Who is actually using Bittensor today?
It’s a mix. You have companies using it directly, for example, teams generating 3D objects on Bittensor that are then used in video games, with partnerships extending into engines like Unity. You also have researchers, including people from traditional Web2 AI labs, who are increasingly interested in decentralised training because it produces more resilient, generalist models.
Then, there are institutional investors speculating on subnets, and finally individual miners who contribute computing and earn TAO by generating intelligence.
Why is a token essential to the Bittensor network?
It all comes down to incentives. Without a token, everyone races to solve the same problem and earns nothing, which is exactly why earlier distributed compute projects failed.
By aligning incentives properly, Bittensor gives everyone a fair opportunity to earn based on the value of their work. In many ways, it’s a better deal than contributing intelligence to a closed system owned by a single company.
Bittensor has seen rapid growth recently. What’s driving this momentum?
Several things converged at once. The OpenAI governance drama highlighted just how centralised AI power has become: the idea that a handful of people could control AGI scared a lot of folks. That pushed people to look for alternatives.
At the same time, Bittensor is fully open-source, incentivising transparent AI development. And importantly, the ecosystem has reached a point where builders are moving faster than the core team. Subnets are evolving faster than we can personally keep up with — which is exactly what you want in a decentralised system.
Do you see decentralised AI as a way past the current limits of large language models?
Potentially, yes. LLMs are already far beyond what can run on a single machine. Training and inference now require massive resources, traditionally only available to organisations with OpenAI-level funding.
Decentralised AI changes this by letting you talk to many models at once, with validators selecting the best responses. Instead of one ChatGPT, you’re speaking to a room of them. That brings resilience, nuance, and scale, and it opens the door to models with trillions of parameters.
Looking ahead, how do you see Bittensor evolving?
I think Bittensor could become something like TCP/IP for AI. Most people don’t think about TCP/IP, but it underpins the entire internet. I see Bittensor becoming that underlying layer, quietly powering AI development everywhere. At some point, training AI in isolation just won’t make sense anymore. You’ll deploy it into a subnet and let it grow within the network.
Are we still early in decentralised AI?
Very early. We were among the first to coin the term “crypto AI”, and now you’re seeing massive valuations being raised for problems we’re already solving. That tells you two things: the market is only starting to understand the value here, and the wave is coming. We’re just at the front of it.
