Hello readers,
The TDR Pro portfolio has already had its first 10-bagger (currently unrealized), and the “early bull” phase of the cycle hasn’t even been confirmed yet. In total, the portfolio is now up over 103%, and our non-BTC exposure is up 156%.
We’re now fully in a “risk-on” stance, after shifting to “risk-off” last October. If you’d like to access the portfolio and receive weekly “cycle awareness” reports + alerts when we make changes to the portfolio, you can sign up here for just $16.67/month.
Back to business.
Demand for AI inference is exploding, but so are the costs developers face to access the compute and models needed to serve it to end users.
Bittensor is attempting to turn that bottleneck into a “decentralized marketplace,” where independent providers compete to deliver inference, compute, data, and other AI services to developers more efficiently than centralized platforms.
The question is whether that market can become cheaper, better, and more scalable than the traditional AI stack.
We explore that question in this week’s edition of The Watch List and provide an update on Bittensor's fundamentals and progress.
Let’s go.
Bittensor is a decentralized marketplace for AI use cases. Independent teams build specialized AI networks on Bittensor called "subnets” for tasks such as inference, prediction, search, or compute.
We can think of it as an open-source, permissionless AWS marketplace for AI services.
Primarily other developers and businesses building AI apps/services. This is similar to an L1 blockchain that primarily sells to developers (who then attract users) in its early days.
The network issues TAO incentives to miners, validators, and subnet creators that provide useful AI services. Individual subnets can then build commercial businesses on top of them, charging users and API customers for access to their AI products.
Bittensor leverages TAO token issuance to incentivize 3rd parties to build on the network. Sort of like how BTC issuance incentivizes 3rd-party miners to secure the network with BTC subsidies (block rewards).
Ultimately, the TAO token has value if those incentives create real products and services built “on top” of its infrastructure (similar to ETH).
The broad TAM is effectively AI infrastructure + AI services: compute, inference, model intelligence, data, agents, search, and prediction. That makes the market potentially hundreds of billions to trillions.
If you’re a developer building an AI app today, you might run your own models (Llama), rent GPUs from AWS/CoreWeave, and pay database/memory/storage fees.
On Bittensor, that same developer can potentially source some of those services from specialized subnets, where independent miners compete to provide services around inference, storage, data, predictions, and other digital commodities
What’s unclear today is whether the available services meet the cost, latency, reliability, and quality requirements of the world's best AI app developers.
Over the last 90 days, the average cost to register a subnet on Bittensor was $222.4k, down 41% q/q, but up 1304% y/y.
These fees can be thought of as capital costs or entry fees for developers. They are paid by the subnet operator (in TAO) and removed from circulation by the network, but can be “recycled” at a later date (more on this later in the report).
Over the last 90 days, 3rd parties have spent $1.7m on Bittensor creating subnets, down 48% q/q but up 238% y/y.
Generally speaking, demand to create subnets has been quite subdued in 2026.
Over the last 90 days, Bittensor has generated 169.4k transactions/day, down 26% q/q but up 12% y/y.
Investors should note that most of these transactions are not inference requests. Rather, they are TAO transactions designed to coordinate the network economically and administratively.
Inference happens mostly offchain.
Bittensor currently has 129 active subnets, unchanged q/q and y/y.
ilium.io is the largest subnet by market cap today ($235k). Chutes is #2 at $229k. And Targon is #3 at $174k. Each offers decentralized compute services for AI developers. Each subnet can represent different layers of the AI stack, designed for specific developer use cases (therefore, they are not all competing with each other, similar to how a DEX is not competing with a lend/borrow app on Ethereum).
Generally speaking, we view 129 as quite small for a network that launched nearly 6 years ago (acknowledging that AI did not go mainstream until ‘23). The supply side really isn’t built out yet. Never mind actual AI user demand from those subnets and applications. Ultimately, if the best AI developers do not view Bittensor as an option to build on, there is no future for the network. We think the next few years will be a key test in this regard.
Over the last 90 days, 11 new subnets were created, down 15% q/q and 15% y/y.
65.3% of TAO in circulation is currently staked, down 2% q/q and 11% y/y.
At its peak (late ‘23), nearly 90% of the supply was staked.
Total Supply: 21 million TAO
Circulating Supply: 11.29 million TAO (53.8%)
TAO is unusual in that it has no founder, team, investor, or foundation token allocations. Rather, 100% of the supply goes to network emissions, which are designed to incentivize subnet developers to build on Bittensor infrastructure.
As such, investors should focus on TAO issuance and emissions relative to “recycled TAO” to assess future dilution risk.
Over the last 90 days, Bittensor has issued 316.25k TAO, up 30% q/q but down 68% y/y.
This represents an annualized inflation rate of 11.36%, which can be offset by “recycled TAO.”
“Recycled TAO” represents tokens that have been removed from circulation after being paid by validators and subnet operators as “registration fees.” With that said, these tokens can later be used as emissions to incentivize future developers/subnet operators to build on TAO.
41% of issuance goes to Bittensor validators/stakers. 41% goes to Bittensor miners. And 18% is shared with subnet operators (incentive to build on Bittensor).
Demand for subnet offerings. Is the compute product on par with other ways to get compute from existing providers? Is it cheaper? Where will the demand come from? Are subnets profitable? Are the best AI developers building on TAO? Ultimately, the answer to all these questions must be “yes” in our assessment of Bittensor’s future. We do not have much evidence that this will be the case today. With that said, the trajectory of AI is likely to change significantly in the coming years.
Token economics and network inflation. Given the token economic structure, there is potential for misallocation of emissions. For example, does capital flow to the best AI products, or simply to the subnets best able to attract stake/speculation through narratives? Furthermore, the 11% inflation rate is concerning, given shifting investor preferences for tokens with buybacks and deflationary token economics.
Stake/validator concentration. Roughly 2/3 of circulating TAO is staked, with 5.4 million TAO sitting on Bittensor’s Root network and delegated across validators. The validator ecosystem includes hundreds of identities, although economic weight is concentrated among just a few dozen major operators today.
Protocol complexity & changing token economics. It is our view that TAO is much harder to underwrite than BTC/ETH/SOL/HYPE and apps with product/market fit and buyback programs. This limits the pool of capital that can invest in a project like TAO.
TAO value capture. The core demand for TAO comes from operators who want to build on the network (they need to acquire TAO to pay the registration fees). Of course, if you want to transact on Bittensor (or stake), you’ll need some TAO to do that as well. But there is no direct fee/or burn mechanism tied to network adoption/use. Given that we are in the era of “buybacks and value accrual to tokenholders,” this could prove a headwind for Bittensor, given today's high inflation rate.
Over the last year, it’s become clear that AI inference costs (through centralized providers) are quite expensive for developers. Never mind the innovation that could come on Bittensor via decentralized “compute legos,” with plug-and-play capabilities similar to DeFi.
As such, a “decentralized marketplace for AI compute services” makes sense on the surface.
But this is also a very complex marketplace. And given that demand to build on Bittensor over the last year has slowed meaningfully (as AI adoption has grown), it is not clear to us that Bittensor has true product-market fit with AI developers.
That may come someday. But the messiness of decentralization in this context gives us pause concerning the long-term viability of the infrastructure network.
Of course, this will probably not stop TAO from garnering a “narrative premium” for its proximity to AI growth in a crypto bull market. And we can acknowledge that the “AI stack” is likely to look very different in a few years’ time (possibly making Bittensor more attractive).
We view Bittensor as an interesting project with upside (which may rely on narratives for now). But one where real use cases may be years out. As such, we find it too complex to underwrite at this stage.
That’s why we’ve left TAO out of the TDR Pro portfolio for now.
We’re currently up over 103% on our portfolio (+156% in our non-BTC exposure). If you’d like to access what we hold and receive alerts when we make changes, you can sign up for TDR Pro for just $16.67/month here.
Take a Report.
And Stay Curious.
Disclaimer: Individuals have unique circumstances, goals, and risk tolerances, so you should consult a certified investment professional and/or do your own diligence before making investment decisions. The author is not an investment advisor and may hold positions in the assets covered. Certified professionals can provide individualized investment advice tailored to your unique situation. This research report is for general educational purposes only, is not individualized, and as such should not be construed as investment advice. The content contained in the report is derived from both publicly available information as well as proprietary data sources. All information presented and sources are believed to be reliable as of the date first published. Any opinions expressed in the report are based on the information cited herein as of the date of the publication. Although The DeFi Report and the author believe the information presented is substantially accurate in all material respects and does not omit to state material facts necessary to make the statements herein not misleading, all information and materials in the report are provided on an “as is” and “as available” basis, without warranty or condition of any kind either expressed or implied.
