I’ve always loved the idea of Bittensor. Despite being a complex protocol, it felt like you could distill it down to a very simple narrative:

That is a great slogan.

You can stay relevant for a long time just drafting on the dream of that one day coming true. But what are the chances it happens? What’s really going on under the hood?

My first touchpoint was a podcast Tom Shaughnessy did with Bittensor co-founder Ala Shaabana in November 2023. I listened to the whole thing and remember struggling to follow all the moving parts. My impression was that even the co-founder had a hard time making it understandable, and there was still a lot of development needed.

I checked in on it here and there over the years. Some high signal people seemed really enamored with it, which always made me ask what I was missing. Barry Silbert was one of them. Regardless of what you think of him personally, and I have plenty of reservations about the guy, he’s been early and right on some low consensus bets. His involvement in Bittensor through Yuma kept it on my radar, along with the interest of a few other people I respect.

Recently, TAO started trending again in a lot of my personal sentiment indicators, and I casually tweeted that I thought it could rip this cycle, but still kind of thought it was vaporware.

I got a lot of DMs about this post. Which is partly why I posted it, as this can sometimes be the fastest way for people to show you what you’re missing.

Many of the DMs were quite constructive and it encouraged me to do another look at Bittensor with fresh eyes.

Why?

Because almost no crypto protocols are static. They’re always evolving and you need to constantly reevaluate things as new information is presented.

There’s a resource I’ll draw from throughout this writeup called Intro to Bittensor. I think it nails the Bittensor 101 explanation for anyone trying to get up to speed:

I like that explanation. Although “the blockchain pays” still leaves us with a pretty obvious question about where the money comes from. We’ll get to that.

To me, Bittensor is one big incentive machine, with TAO helping fund the work happening across the network. More on the token later.

Bittensor breaks this work into subnets, each with its own task and competition.

But what are they working on?

One subnet might provide AI inference. Another might rent out GPUs. Another might train models, generate data, make predictions or work on drug discovery.

The basic structure is pretty simple:

It’s a pretty meritocratic system if all elements work harmoniously together as intended. And I really like that aspect of Bittensor.

Most of the actual work happens offchain. Bittensor is the coordination layer that tracks participation, validator scores and rewards.

Instead of a company hiring a fixed team, a subnet creates an open competition. Anyone with the right hardware, model or approach can participate, and the best performers should earn more.

The scoring system has to identify useful work. If the scoring is wrong, Bittensor may reward people for gaming the test instead of building something customers actually want.

A subnet attracting miners also doesn’t necessarily mean anyone is paying to use what they produce.

But let me finish this section with a quote from Seby, who’s currently in the trenches and has built 2 subnets. He’s also one of the fine gentlemen who reached out constructively in my DMs:

TAO is the base asset of Bittensor. It has a 21 million maximum supply, similar to Bitcoin, and new TAO is issued to support the network and reward participants.

The first halving occurred in December 2025, reducing issuance from roughly 7,200 TAO per day to approximately 3,600 TAO per day. The current split is roughly 41% to miners, 41% to validators and stakers, and 18% to subnet owners.

TAO is used to participate in the network, stake behind validators and gain exposure to the subnet economy. It’s also paired with each subnet’s own token.

Since passing of dynamic TAO, each subnet has its own token called alpha. Alpha trades against TAO through a built-in liquidity pool. When you stake TAO into a subnet, you’re effectively swapping into that subnet’s alpha token. You can earn rewards, but you’re also taking on the price risk of that individual subnet.

Subnet prices help determine how much of the network’s emissions each one receives. The idea is that capital flows toward the subnets investors believe have the best prospects.

Bittensor recently tightened this system with the Emission Gate, which sharply reduces rewards for subnets below a moving demand threshold. This should cut down on weak or idle projects collecting emissions simply for existing. But the gate still uses token demand as a proxy for quality, and crypto markets can direct capital toward the best story rather than the best product.

Another thing, TAO holders don’t own equity in subnet businesses. If a subnet becomes valuable, that value doesn’t automatically flow back to TAO. But once again, this is still a problem for 90% or more crypto tokens.

The emissions are also significant. At roughly 3,600 TAO per day, the network issues about 1.31 million new TAO per year. Against a circulating supply of roughly 11.3 million, that’s approximately 11.6% gross annual supply growth, before recycling.

At $250 per TAO, that would represent around $329 million of new tokens issued over a year. Some of that funds the people building the network, but it’s still a substantial incentive budget.

So yeah, that’s a pretty heavy dose of yearly inflation to overcome.

However, I don’t think that means the model is broken. Early networks often need to pay people to participate before there’s real revenue (unless you’re Hyperliquid and 2.2% will suffice). The question is whether that spending eventually creates something people are willing to pay for, or if the network is mostly paying people to stick around.

It’s also important to distinguish where the buying comes from. Some subnets now appear to be using customer revenue to buy back alpha, which is a real sign of outside demand. Bittensor also creates automatic alpha buying through its emissions system, but that tells us much less about whether customers actually want the product.

In order for the TAO thesis to workout there’s a few things that need to happen. Bittensor needs to become an important network for producing intelligence, and TAO needs to capture enough of that value to justify the dilution required to build it.

The first part, as I’ll get into shortly, is becoming easier to believe. The second still has to be proven IMO.

There are some very interesting things happening in the Bittensor space. And it’s unfair for me to call it “vaporware”. There are things you can use, and some research worth paying attention to.

This is by no means an exhaustive list and there might be more interesting things I should have highlighted but these stood out to me:

Chutes (SN64) provides AI inference, which simply means running a model to get an answer. Developers can access it through OpenRouter, where its models and pricing are listed. You don’t have to buy into the whole Bittensor thesis to use the service.

Lium (SN51) runs a GPU rental marketplace where providers earn customer rental payments alongside subnet emissions. SubConnect recently estimated that Lium is generating $8 million to $10 million in annualized revenue, making it the largest revenue-producing subnet they tracked. Customers buy compute credits, which are then used to buy and burn SN51 alpha.

ORO (SN15) is doing something a little different. It used output from its subnet competition to train Qwen3-4B, taking its ShoppingBench success rate from 18% to 42.7% with less than a day’s worth of subnet data. It’s a pretty interesting example of Bittensor producing valuable training data alongside compute.

OK this came out just as I was publishing, but this is cool:

And a bit more on this:

BitMind (SN34) uses an ongoing competition between deepfake generators and detectors to keep improving its models. In a July paper covering 19 datasets, it matched or beat the best commercial detector cited on Deepfake-Eval-2024 and comfortably outperformed the best open-source models. The research came from the team and hasn’t been peer reviewed, but the evaluation tools and tested model are public.

Zooming out, SubConnect estimates that 24 Bittensor subnets are generating a combined $28 million to $35 million in annualized revenue, with 14 using outside revenue to fund alpha buybacks. SubConnect rated 15 of the 24 revenue estimates as high confidence. Still, these numbers come from inside the Bittensor ecosystem and aren’t audited financials. As mentioned above, at $250 per TAO, the network is issuing around $329 million in new tokens annually, so reported revenue is still only around one-tenth of the incentive budget. Early, but no longer zero.

So yes, there’s really something here. The revenue picture is getting better, but it is still early and much smaller than the emissions supporting the network. The next test is whether customers keep paying, revenue keeps growing and these businesses still work as subsidies fall.

In fact, Aylo might have summed things up the best in his response to my original post:

On Bittensor, Covenant AI trained a 72-billion-parameter model from scratch across permissionless, globally distributed infrastructure, showing decentralized AI can work at meaningful scale. Then after doing that the team left Bittensor in April 2026. I remember when this happened and at the time I wondered if it was a death knell for Bittensor.

What stood out to me was the possibility that Bittensor could help a team build something valuable, only for that team to leave once it had enough funding or customers to operate independently.

The open source work can remain available, but the people and future business don’t automatically stay with it. Bittensor has since introduced locked-token mechanisms that allow subnet ownership to change, so another operator may be able to take over. But that operator still doesn’t inherit the departing company’s employees, customer relationships, or contracts.

For me, the real answer is that successful teams need to want to stay. Bittensor’s contributors, funding, and services have to keep adding value as those teams grow. I’d want to hear the strongest subnet teams explain exactly what they would lose by leaving.

That may still require some tinkering with the incentive design, and that doesn’t mean the model is broken. Solana is debating an acceleration of its disinflation schedule, while the Ethereum community is considering issuance changes of its own. These systems are hard to get right on the first try. They need fine tuning.

However, like Bitcoin, I see the Bittensor ecosystem as highly reflexive. In good times, speculation can drive TAO higher, injecting a lot of adrenaline into the entire subnet ecosystem because TAO is essentially the lifeblood. But that same reflexivity can work against it in bear markets.

I’m more interested in Bittensor than I was going into this. I came in with a bearish tilt and actually expected to confirm my priors, but that’s not the case. There are products and research here that I wasn’t giving enough credit to, and calling the whole thing vaporware feels extremely dismissive. Shame on me!

I’m very bullish on the idea of using an open network to fund and coordinate AI development, and I understand why people are excited about it. Frankly, I think it’s imperative that this happens, and I believe it will.

We’ll see whether Bittensor is ultimately the right model, but so far, it appears to be way out in front. The ecosystem also seems willing to keep tinkering with its incentives and refining the model as it goes.

All that said, I find TAO harder to underwrite than something with clearer revenue and cash flows to token holders. But honestly, that difficulty is part of the intrigue and part of the upside.

And there’s enough of a big idea here for the the TAO token to still outpace fundamentals. Which is why I think it’s still potentially a good trade this cycle.

It’s so difficult to put a ceiling on TAO because the TAM is massive. You also have to assign some probability to Bittensor winning the open source intelligence race. And that probability should rise as more research breakthroughs emerge and revenue grows.

However, I’m still a fundamentals guy and what would make me more bullish from here is fairly straightforward: customers coming back, businesses needing less subsidy, and successful teams continuing to build on Bittensor because it helps them. Then I want to see how much of that success makes its way back to the tokens. One way to keep an eye on this? Follow Subconnect and watch for their revenue updates (and if someone has a dashboard for this, please send it my way).

For now, I’m glad I took another look. I still don’t have enough conviction to make TAO a long-term hold, although that could change within a few months. I think there’s also an opportunity for people with the bandwidth to research individual subnets. I wouldn’t be surprised to see some real winners emerge there (this is probably where the alpha is the tokens are literally called alpha!).

I will end with this. In my opinion, Bittensor is something you need to keep paying attention to and revisit often. One genuine breakthrough could change the entire story on a dime.

7 day benchmark stablecoin rates from Portals: 3.16%:

Here’s the top yielding stablecoin vaults (real yields) for the past 30 days:

Checking in on Stablewatch to see the 7-day TVL changes.

This week the top movers were: Smart Dex (USDN) + 25.5%, USD.ai (sUSDai) +12.5%, f(x) protocol (fxSAVE) +12.2%:

7 day benchmark ETH staking rates from Portals: 2.26%

Here’s the top yielding ETH vaults (real yields) for the past 30 days:

Here’s a snapshot of the front page of looping this week from yieldz.io (30d APY):

Taking a look at the front page of Stablecoin yields and RWA yields this week on Pendle:

Stablecoins

RWA

Thanks for reading!

DISCLAIMER: Nothing written in The Edge Newsletter or said on The Edge Podcast is a recommendation to buy or sell tokens or securities. This content is for educational and entertainment purposes only. Nothing shared here is financial advice. Any views expressed in our content are solely the opinion of that writer, host, or guest. Always do your own research. DeFi Dad, Nomatic, and guests may have positions in the assets or other matters discussed in this content.