Subnets / SN9
iota SN9
iota (subnet 9) is a Bittensor subnet held by 5,707 unique coldkeys. Its liquidity pool holds 55.5k τ, with the alpha token trading at 0.0270 τ. Roughly 34% of tracked holders are in profit. Seller exhaustion is elevated: 82% of recent sellers have emptied their wallets, thinning the supply overhang. Full insider flow, whale tracking and the complete proprietary factor set for iota are available to SubnetStats subscribers.
ALPHA PRICE
0.0270 τ
IN TAO
HOLDERS
5,707
UNIQUE COLDKEYS
POOL DEPTH
55.5k τ
TAO IN POOL
7D MOMENTUM
-7.9%
ALPHA vs TAO
Project activity
Less than two weeks until Exploit, and the picture of what we're bringing is coming into focus.
A summer of runs, turned into proof. What distributed training actually delivered, and what those numbers mean commercially, laid out in the open.
It's a key shift for iota, moving u
At @proofoftalk, we revealed Project Orion, on @IOTA_SN9
and asked a simple question: can distributed AI training actually compete commercially?
We've spent the whole summer answering it and at Exploit, we will show what we found.
https://t.co/uYUIIldh1N
Orion’s 16B run hit a major milestone passing 100B training tokens and it's still improving.
Our co-founder @macrocrux elaborates on the achievement:
“This is the largest LLM ever pretrained using DPP: Currently at 10^22 FLOPS using heterogeneous commodity GPUs from permissionl
If you own GPUs, you already know the number that matters isn't how many you have. It's how many are earning right now.
Allocation isn't utilisation.
Capacity can be fully booked on paper and still sit idle in practice and every idle hour is depreciation you're paying for again
Most companies building on AI today are building on rented ground.
The model you depend on is one someone else owns, hosts, prices, and can change. That's a reasonable trade but the more central AI becomes to how your business actually runs, the more that dependency can start t
Every AI compute conversation runs on the same anxiety: not enough.
Not enough GPUs, not enough data centres, not enough data, not enough power.
All real, all justifiable, but for most teams that's not the wall they hit first.
The wall is that the compute you can get is rarely
There is a quiet assumption when it comes to most AI training: that the compute you're using is stable, uniform, and always there.
Even in a single datacenter, your training setup is always shifting, even when you own every GPU in the building. Machines drop out and come back,
The headline is a 16B-parameter model, but the model size isn't the hard part.
The hard part is everything beneath it: coordinating hundreds of independent GPUs, in different datacenters, over the public internet, while machines join and leave whenever they want.
That's the iot
Distributed training is possible, that question is settled.
The harder question is can a system take compute that constantly joins, leaves and restarts, over ordinary internet, without degrading the model or eroding the cost advantage that made it worth using?
The next part of
Our distributed training runs use compute that iota doesn't own and can't fully control. Machines join, leave, stall, and occasionally fail – these are normal operating conditions for @IOTA_SN9.
So, how does iota ensure fault tolerance when the GPUs that are used for training ar
Project Orion's 16B run isn't training on one type of machine, in one place, from one supplier.
The run is built to handle 256 machines. Right now, around 188 are online; drawn from multiple independently operated providers, running different hardware, in different locations, un
iota is built to bring distributed capacity online: hardware that would otherwise sit unused, coordinated into one functioning training run.
Project Orion's 16B run trains across 256 globally distributed, heterogeneous GPUs, supplied by an open network of independently operated
Project Orion, on @IOTA_SN9, began with a series of deliberate training runs designed to answer the questions we needed to solve before we could scale to new heights.
Large-scale distributed training is not unlocked by one breakthrough. It is earned through repetition: running e
Most large-model training happens behind closed doors.
The final model is announced, followed by results and benchmarking, with very little shown of the actual training in between.
Our public dashboard for Orion's 16B- training run opens the doors — live training, visible to th
A 16-billion parameter model is now training live on @IOTA_SN9; globally distributed over three continents, heterogeneous, permissionless, owned by no single entity.
Today, we present the next stage of Project Orion - Orion-16B.
Over the coming weeks, we'll show what it means t
Recent changes to Bittensor emissions incentivizes subnets to reduce their miner burn to zero.
In light of this, we are rolling out zero-burn initiatives across our subnets which balance productivity gains with price stability.
We are commencing today by reducing miner burn to
AI access is becoming a geopolitical dependency.
Chinese open models like DeepSeek and Qwen spread fast because they're capable, accessible, and cost effective - and Western companies have built products around them.
Now Beijing is reportedly weighing restrictions on foreign ac
The next evolution of Orion has seen the team pushing our experimentation stack: 70+ autoresearch sweeps ablating the params that maximise MFU for production.
Several million tokens, 200 GPUs, and one autoresearch loop.
More insight and announcements on the evolution of Orion w
Liquid training is the mission.
iota is the refinery for this future.
Read @WSquires thoughts on infrastructure, efficiency, and how @IOTA_SN9 can disrupt AI infrastructure through heterogenous, interruptible, distributed training.
OpenAI just unveiled Jalapeño, its first custom inference chip.
https://t.co/Pn3zhsat8O
With this, we see that AI infrastructure is being reorganised around workload economics. Jalapeño is a clear signal that training and inference should not be fighting over the same GPUs - it
The AI market has been actively embracing open-source models, many of which are from China. Models from this region have moved from 1.2% of global usage share in late 2024 to 30% in 2025.
Airbnb, AWS, Azure, HSBC, and GCP are just a handful of the corporations bringing these mod
We hosted a workshop at Imperial College London yesterday, preparing the crowd for the upcoming AI agent hackathon starting on Sunday with @iclblockchain.
We introduced students and developers to Macrocosmos, giving them an overview into how distributed architectures can change
“Frontier labs spend millions of dollars a year on workload prioritization. iota is a way to democratize that for training.”
Co-founders @WSquires and @macrocrux appeared on Hash Rate, discussing @IOTA_SN9's economic benefits, showing how flexible training can fill idle GPU capa
Tomorrow co-founders @WSquires and @macrocrux will be appearing on @markjeffrey’s Hash Rate podcast.
We will be discussing @IOTA_SN9 and our Orion-100B pretraining run, along with our @Apex_SN1 competitions and upcoming hackathon at Imperial College London. https://t.co/Dp42jp6s
We occupy an integral part of the cutting-edge decentralised AI sector. With @IOTA_SN9, we are building distributed AI tools to facilitate model training across the globe, on heterogeneous compute.
Decentralised training is the most important technology we perceive in unlocking
SELLERS DRAINED
82%
28D SELLERS EMPTIED OUT
CAPITULATION
+0.12
BOUNCE COMPOSITE (z)
HEAT 7D
0.08×
VOLUME ÷ POOL DEPTH
HOLDERS IN PROFIT
34%
TRACKED WALLETS · ENTRY BELOW PRICE
See who's buying and selling iota — every trade classified miner / validator / owner / outsider, plus holder cost basis and the full 23-factor screener.
OPEN THE INSIDER TAPE →