Xavier Lyu explains his business with a Formula One race. "If your pit crew is screwing on the wheels 2 seconds slower, you might not win the race, even if you have the fastest engine," he sa
Xavier Lyu explains his business with a Formula One race.
"If your pit crew is screwing on the wheels 2 seconds slower, you might not win the race, even if you have the fastest engine," he said.
In his analogy, the engine is the AI model — the thing every lab brags about. The pit crew is everything that makes the model cheap and fast to run. Lyu is betting the race is now won in the pit.
He's one of three founders building on Bittensor who told TheStreet Roundtable how they're using the network to do something crypto isn't famous for: finding customers who pay.
First, the plumbing, in plain words. Bittensor is a blockchain network powered by the TAO token. It hosts independent projects called subnets, and each subnet pays contributors around the world — known as miners — to compete on one specific AI task. Best solution wins the rewards. The founders running these subnets treat the setup as a global research engine: post a hard problem, let hundreds of competitors attack it around the clock.
Crypto has a reputation for hype over revenue. These three are making a different case.
Related: Analysts rank 16-year-old chip firm 61 places above NVIDIA
Lyu came to Bittensor from Pantera Capital, the crypto investment firm, where he worked in quant and fundamental research. His company, Pareton AI, doesn't build AI models. It makes them cheaper and faster to run.
The business logic: picking the right model is no longer what separates AI companies, Lyu argues — efficiency is. Margins are thin across the industry, so even single-digit savings add up. Pareton uses Bittensor's miners to search the enormous range of possible optimizations faster than an in-house team could.
"We're able to get to our solution faster and cheaper," he said. Early pilots with design partners look very promising, he added.
The bigger prize, as Lyu sees it, sits outside tech: hospitals, schools and small businesses that want private, locally run AI — organizations that can't send sensitive data to a cloud provider and need their models lean enough to run on their own machines.
Cameron Wang left the University of Chicago and has been building startups since 2017. His read on robotics: the bodies are ready, the brains aren't.
Robot hardware is already good enough, he said, but the intelligence inside is closer to that of a two- or three-year-old. His company is building the missing piece. "You can think of it as AI models for robots," he said of Open Roboto's work.
The bottleneck is real-world data — robots learn from recordings of humans actually doing physical work, and there's never enough of it. Open Roboto turned that scarcity into the business itself: contributors wear recording devices while doing manual labor and get paid through Bittensor rewards. The footage trains the models.
Two months after launch, Wang said, Open Roboto has partnerships with robotics companies, a base model that reached state of the art on a robotics benchmark, and factory pilot programs worth a couple hundred thousand dollars.
"There's a huge demand over there," Wang said.
Gavin Zaentz worked on the stock exchange at Nasdaq before co-founding Leadpoet, which automates lead sourcing and sales outreach for clients including Dropbox.
His use of Bittensor is the most direct of the three: he points the network at his own product. Up to 256 miners compete to improve Leadpoet's sales agent, an arrangement he describes as letting the company "outsource our R&D."
The result, he said, is "an agent that we could never have built internally ourselves."
