After Mining Score: Why I Think It's a High-Potential Subnet

I want to be upfront about what this is: my own opinion, formed after actually mining on Score, not a research desk's neutral coverage. Take it as one builder's thesis, not financial advice — and see the disclaimer near the end before you act on any of it.

The problem every subnet has

Here's something that becomes obvious once you spend time across a few different Bittensor subnets: almost all of them are drawing from the same pool of people. The same few thousand crypto-native ML engineers rotate between whichever subnet has the best emissions this month. That's fine for keeping a subnet alive, but it caps how big the whole network can get. Bittensor's ceiling isn't set by how good its miners are — it's set by how many people outside crypto ever have a reason to touch it.

Under dTAO, that ceiling matters more than it used to. Every subnet has its own alpha token and its own liquidity pool, and a subnet's emission share is the network's own running vote on how much value it's creating. A subnet that pulls in real, outside demand — people staking or building on it who were never going to mine on some other subnet anyway — is one of the few things that actually grows the pie instead of just reslicing it. That's the setup I think Score is positioned for.

A quick look at Score right now

Score is Subnet 44 — "Making every camera intelligent." As of July 27, 2026, here's where it stands on-chain:

Metric	Value
Rank by market cap	#7 of 128 subnets
Alpha price	~0.0335 TAO
Market cap	~123,800 TAO
FDV	~185,300 TAO
Volume	~2.02M TAO
Validators / Miners	10 / 246
Emission share	2.8%

View Score's live subnet page →

That emission share — 2.8% of everything the entire network mints, across 128 subnets — is the network's current, stake-weighted opinion of how much value Score is producing. It's not a fixed number; it moves with how the root network's validators score it over time, same as every subnet.

Why football, specifically

Score's actual product is computer vision for sports, starting with football (soccer). The initial wedge is Game State Recognition — tracking players, the ball, and match events automatically from raw video, at a fraction of what manual annotation costs today. Score's own materials frame the target market at roughly $600 billion for football overall, with $50 billion of that in betting and $30 billion in data services.

I don't think the interesting part is the total addressable market number by itself — every subnet pitch has a big TAM slide. What's interesting is who that market is made of: broadcasters, sportsbooks, leagues, and scouting departments who have never mined a token in their life and have no reason to care that the model underneath was trained on a decentralized network. If Score's output ends up embedded in a betting platform's live odds engine or a club's scouting pipeline, that's a mass audience arriving at Bittensor sideways, through a product, not through a CT timeline.

The competition inside Score is real

It's easy to assume a "football AI" subnet is a novelty — bounding boxes around a ball, nothing more. Having actually mined on it, that undersells what's going on. The tasks miners compete on span detection, segmentation, description, reasoning, OCR, counting, and action recognition, across CCTV, dashcams, and drones, not just match footage. The miner pool is more diverse than people assume, and it shows in what the network has actually produced.

The clearest proof: in June 2026, a distilled 19-megabyte model trained through Score's network — small enough to run on a four-thread CPU with no GPU — outperformed GPT-4o, Gemini, Grok, and Claude at object detection, scoring 0.848 mAP on the UA-DETRAC vehicle benchmark. It ran roughly 9x faster than the best specialist detector it beat and 70–130x faster than the frontier chat models it was compared against (source). That's not a subnet coasting on a niche; it's a network whose competitive pressure produces results that get benchmarked against the biggest labs in the world — and wins on the specific task.

The new approach that I think changes the equation

On June 26, 2026, Score announced it's going multimodal. Instead of only running narrow, single-task vision skills, it's training a family of on-device vision-language models called Satori, distilled down from larger frontier models into compact ~2B and ~0.5B checkpoints.

The detail that matters most to me: Satori 1.0 2B reportedly covers nine separate perception tasks — detection, open-vocabulary detection, segmentation, description, reasoning, OCR, counting, action recognition, and early temporal video — from a single ~2B checkpoint, in one forward pass, with no teacher model needed at inference. By Score's own account, even a frontier model like GPT-4o only natively covers two of those nine. And critically, it's built to run on the camera itself, not in a data center.

Here's why that's a bigger deal than a model release: it turns Score's existing skill-distillation competition into an input, not the end product. Miners keep competing to sharpen narrow skills; those skills get folded into Satori; Satori powers Manako, Score's no-code vision agent platform, which is what actually gets sold into real businesses. Skills sharpen the model, the model powers the product, the product is what a non-crypto customer touches. That's the flywheel — and it's a much shorter path to the "mass audience" outcome than hoping individual skill-marketplace transactions add up on their own.

The market this is actually chasing

"Every camera intelligent" isn't a slogan-sized market. A few independent estimates as of 2026:

Market	2026 est.	Later-year est.	CAGR
Video analytics	~$15.0B	~$33.7B by 2030	~22%
Computer vision (overall)	~$24.1B	~$72.8B by 2034	~20%
Smart cameras	~$51.4B	~$159.6B by 2036	~12%
AI-specific video analytics	~$6.2B	~$17.2B by 2031	~23%

These are independent industry forecasts, not Score's own numbers, and forecasts this far out should be read as directional, not precise. But directionally, they all point the same way: cameras that can understand what they're seeing, cheaply, at the edge, are a market being counted in tens of billions today and low hundreds of billions within a decade. Score doesn't need to "win crypto" to matter — it needs to take a sliver of a market that's already large and already growing without any help from Bittensor.

Why I think this pumps more than just Score

This is the part that's genuinely my own belief, not something I can point to a chart and prove: if Score (or any subnet) starts pulling in demand from people who were never going to touch Bittensor otherwise — sportsbooks integrating its output, businesses running Manako, camera manufacturers licensing Satori — that demand shows up as real staking activity and real emission-share growth for Score specifically. But it doesn't stop there. New attention on one subnet is attention on the mechanism itself: dTAO, the root network, and the idea that a permissionless, incentive-driven network can out-execute centralized AI labs on a specific task. That's a story that's easier to tell with one concrete, benchmarked, non-crypto-facing product than with 128 subnets' worth of abstract pitches. A rising subnet doesn't guarantee TAO goes up — but a subnet with a real, provable, mainstream story is one of the more plausible ways the broader ecosystem gets re-rated.

Risks, and why this isn't financial advice

I could be wrong about all of this, and you should treat it that way. Specific things worth weighing before you do anything:

• Alpha token price is volatile — Score's alpha, like every subnet's, is priced through an on-chain pool and can fall as easily as it's risen. See Bittensor subnet staking explained for what that exposure actually looks like.
• Manako and Satori are new — a strong benchmark result and an ambitious roadmap aren't the same as proven commercial adoption yet.
• Competition isn't limited to Bittensor — centralized computer vision vendors aren't standing still, and other subnets are pursuing overlapping ground.
• Emission share can fall as easily as rise — it's a live, stake-weighted vote, not a fixed allocation.

Nothing here is personalized investment advice, and none of it should replace your own research. If you're evaluating Score or any other subnet, how to evaluate a Bittensor subnet is a reasonable framework to run it through yourself rather than taking my word for it.

Think Score deserves a spot in your portfolio?

In-app buying isn't live on BitExplorer yet — join the whitelist and we'll notify you the moment you can buy in.

⚡ Join the whitelist — be first to buy Score →

Related Articles

• How Bittensor's root network decides which subnets get funded
• What alpha tokens are and how their price is actually set
• How staking into a subnet works, and what it risks
• A framework for evaluating any subnet yourself