Norwegian adventurer living in South-East Asia...... in constant pursuit of a life less ordinary.....
Phnom Penh / Yangon
Joined August 2012
- Tweets346
- Following5K
- Followers419
- Likes8.8K
Mandrake The Magician retweeted
No lies detected.
🚨 $TAO Subnet Valuation Series 👇🏼
SN4 Targon vs CoreWeave (a Nvidia-backed $24.8B company)
I had to close my laptop for a minute.
Ran the numbers on SN4 Targon of $TAO.
Nothing about this pricing makes sense.
The AI cloud compute market is worth $500B+.
@CoreWeave , the market leader:
→ $1.9B revenue (2025)
→ $24.8B valuation post-IPO
→ Nvidia-backed
→ Centralized, closed, restricted access
→ Priced at 86x forward revenue
Now look at @TargonCompute SN4:
→ $10.4M ARR: real, confirmed revenue
→ Confidential GPU compute, on-chain
→ Global, permissionless access
→ Manifold Labs + Intel partnership
→ ~$66M market cap
→ Priced at just 3.6x revenue
CoreWeave trades at 86x revenue.
Targon trades at 3.6x.
Same market.
24x cheaper valuation multiple.
Either I’m missing something.
Or this is badly mispriced.
This is Part 1 of my $TAO Subnet Valuation Series.
I’m running systematic valuations across the most underpriced subnets on Bittensor.
Next up: SN1 Apex vs OpenAI the world’s most valuable AI company.
The gap will surprise you.
Follow @CryptoAvex for Part 2.
Notifications On.. 🫶🏼
Mandrake The Magician retweeted
Most of you will ignore this. But Bitcoin OGs will get this instantly. 👇
I’ve been around long enough to know how early opportunities feel. When I was accumulating Bitcoin, most people didn’t get it. They called it useless, risky, and said it would never work.
The ones who understood it early didn’t wait for approval, they paid attention, learned, and positioned.
I’m starting to see that same early-stage misunderstanding again with @Bittensor.
This isn’t just another altcoin people trade for quick gains. It’s something deeper. It’s infrastructure being built quietly while most of the market is distracted. Instead of chasing hype, it’s focused on creating real systems around decentralized intelligence.
▪️Decentralized AI networks actually producing value
▪️Real incentives, not fake yield
▪️Subnets competing like startups, but open
What stands out to me is how early this still is. $TAO isn’t just a token you flip, it’s becoming the base layer for decentralized intelligence, similar to how Bitcoin became the base layer for money.
The difference is, this time the market is even bigger. You’re not just looking at finance, you’re looking at AI, data, and computation all merging into one open network.
Most people will ignore it because it’s not simple to understand yet. That’s exactly how Bitcoin looked in the beginning. The same doubts, the same hesitation.
But if you’ve been through one cycle early, you start to trust the pattern. I’m not saying it replaces Bitcoin, nothing does that. But I am saying this: opportunities at this level don’t come often.
Bitcoin was the first time I saw it. Bittensor feels like the next time I’m seeing it early again.
You can study it now while it’s still quiet, or you can come back later when everyone is talking about it and pay a much higher price to understand the same thing.
Mandrake The Magician retweeted
Fantastic to spend time with many of the leaders in the Bittensor $TAO community at Yuma's first ever summit. The vibes, enthusiasm and camaraderie reminded me of DCG's first Summit in 2015
nitter.gq/BarrySilbert/status/20…
Truly an inspirational group. Big things are happening
Mandrake The Magician retweeted
Insane.
So a 4 star admiral testifying to Congress now sounds like @adam3us.
Boys and girls-btc is becoming embedded within US power infrastructure.
Game over.
Valued only at $1.5T.
65% less than @Apple.
I'm not leaving. I'm not pivoting away from this space. It's early.
Mandrake The Magician retweeted
Next-gen human enhancement: muscle, cognition, mood — beyond GLP-1s
@metanova_labs
metanova-labs.ai/dashboard
Some of the most underinvested areas in frontier biology that could accelerate civilizational progress:
- Cheap, large-scale DNA synthesis (writing entire chromosomes or full organisms)
- Real-time, non-destructive RNA sequencing in living cells
- Highly accurate AI-powered polygenic scores for complex traits (disease risk, cognition, longevity) → enabling full genome design
- Ultra-precise, multiplex genome editing (far beyond CRISPR) with minimal off-target effects, scalable across millions of cells
- Safe, efficient, tissue-specific in vivo delivery systems
- Safe and effective human germline engineering
- Accelerated clinical trials via testing on decedents (with consent)
- Next-gen human enhancement: muscle, cognition, mood — beyond GLP-1s
- Ectogenesis / artificial wombs
Who’s actually building in these areas? Drop names, companies, or researchers below 👇
Mandrake The Magician retweeted
A Bittensor founder just told me it's the closest thing to Bitcoin he's ever seen.
Not from a price angle.
From an ethos angle. And he's been building decentralized AI since before most people knew what Bittensor was.
His name is @MaxScore. He runs @webuildscore, Subnet 44 on Bittensor.
And what they're building is quietly one of the most interesting things in the entire ecosystem.
Score is a vision AI subnet.
Not sports data. Not soccer tracking. General purpose computer vision.
The idea is simple.
Instead of solving one specific vision problem like every centralized company does, Score is building a library of vision "bricks" that anyone, including AI agents, can plug into and use.
Traffic detection. Person tracking. Action spotting. You name it.
And they're doing it with decentralized miners competing for TAO rewards.
• $80k/month in revenue from real enterprise pilots right now
• Partners include one of Europe's largest fuel distribution companies and one of Africa's largest fruit companies
• No centralized lab. No $100M raise. Just miners, validators, and aligned incentives.
Max said something that stuck with me.
"Computer vision companies have to pick one vertical because it's too expensive to go wide."
"We can go completely horizontal because of Bittensor."
That's the edge.
Decentralization isn't just a philosophy here. It's a business model.
And the market they're going after is massive.
Most of their customers tried solving vision problems with ChatGPT first.
Didn't work. Too complex for LLMs.
Then tried training their own models. Too expensive.
Then found Score.
Max also dropped something most people in Bittensor get wrong.
You don't pick a subnet first.
You buy TAO first. Then you stake into subnets you believe in.
"The value flow is super clean. But people keep skipping step one."
He's right. And it matters for how you think about the whole ecosystem.
Where is this going?
Max thinks the next wave is AI agents mining subnets.
Not humans. Agents.
And the subnets that survive will be the ones building things that actually matter outside of crypto.
Score is already there.
Full interview on the channel ⏬
🤖 Made with AI
Mandrake The Magician retweeted
BITTENSOR JUST TRAINED A 72 BILLION PARAMETER AI MODEL ON A DECENTRALIZED NETWORK
Mandrake The Magician retweeted
This is the big use case for $TAO.
Training all LLMs for a fraction of the cost.
This can 20x it.
Frontier labs spend billions to train a single model
OpenAI, Anthropic, Google and Meta have racked up eye watering amounts in funding / CAPEX spend
Bittensor coordinates thousands of independent teams across 128 subnets to do the same work for a fraction of the cost
All without the single points of failure and associated centralization risk
In the coming weeks we’ll be doubling down on highlighting decentralized AI’s value proposition
Bittensor (TAO) is a strong contender in succeeding within distributed intelligence networks
It’s worth paying attention to the top 30 subnets and what they’re building
Unsure where to start? Don’t worry, we’ll be providing the tooling to help make this easier for you
Let me know in the comments which areas of the ecosystem you’re struggling to get to grips with
Mandrake The Magician retweeted
This is a massive step forward for #Bittensor ❤️
Mandrake The Magician retweeted
Many people outside of the $TAO community missed the recent partnership and joint white paper release between Targon and Intel.
The opportunity here is MASSIVE!
This Bittensor subnet is making serious moves.
SN4 $TARGON
@TargonCompute
@intel
Mandrake The Magician retweeted
Mark Jeffrey: TAO is the most efficient form of capitalism ever created
“Imagine if Satoshi didn’t do decentralized Bitcoin and instead went to VCs and just raised a bunch of money. Just imagine the insane amount of cash he would have been able to raise”
“Instead of doing that, TAO’s proposition is, we’re not going to hire anybody. You have Earth’s best minds competing against each other and they don’t have to take a job, so they can just do this until they get bored and leave”
“Because of the fierce competition, everybody benefits from the cleverness of the last winner. If you win you get $25-75k worth of subnet tokens. You can’t beat contestonomics that have financial incentives worldwide. It’s the most efficient form of capitalism ever invented”
Mandrake The Magician retweeted
Here we go -- Const hosts Sam of Templar SN3:
Covenant Labs just did a 90-minute AMA breaking down their 3 Bittensor subnets.
@tplr_ai. @basilic_ai. @grail_ai. Pre-training, compute, and post-training under one roof.
Most people missed it. Here's everything they said.
Covenant is building what they call the "end to end intelligence continuum."
Three subnets. Three layers of the AI stack. All permissionless.
Templar (SN3) handles decentralized pre-training. Basilica (SN39) handles compute. Grail (SN81) handles RL post-training.
@DistStateAndMe, the lead, put it bluntly. Decentralized training is "humanity's last dance."
Not about beating OpenAI head to head. About creating optionality. About making it cheap enough for anyone to train models.
The gap between academia and frontier labs is growing exponentially. Researchers can't afford to experiment. The actual training run costs 5% of the reported budget. The other 95% is experimentation.
If Covenant cracks cheap training, that entire surface area opens up.
On Templar specifically:
• Hit 39% emission on Bittensor. Highest since Apex was the only subnet on the network
• Covenant-72B trained permissionlessly with 70+ contributors on commodity internet
• 1.1 trillion tokens processed. No centralized data center
• Performance competitive with LLaMA-2-70B
On Grail, something flew under the radar.
They built Pulse. A weight synchronization method that compresses model updates by 100x.
• In RL post-training, only ~1% of weights update per step
• Pulse exploits that sparsity. Lossless compression
• Prime Intellect's comparable system took 14 minutes to sync a 30B model
• Pulse makes decentralized RL training actually feasible at scale
• Already used by Cursor
The lead researcher on Grail said they've trained on math, code, and GPU kernels. Got 40-60% improvement on benchmarks. Working toward agentic training with 100K+ token context and 30B+ parameter models.
On Basilica, the compute subnet:
The team was blunt. Just reselling GPU hours is a 5-10% margin game. Traditional compute providers already do that.
Their play is value-added services.
• "GPU as code." No dashboard. No UI. Agents interact via SDK
• Custom scheduler that places workloads across heterogeneous hardware
• Verification checks for GPU, CPU, bandwidth, memory, storage, and OS security
• Partnerships with providers like Mass Compute for 10-20% below market pricing
• Miners compete on useful infrastructure, not just GPU hours
Sam then went on a rant about the miner burn debate.
His take: Bittensor had to grow up. dTAO introduced investors. The old "miners are God" philosophy doesn't hold.
• Subnet owners have a duty to protect token value
• Miners are a resource optimization exercise, not a cost reduction exercise
• 100% miner emissions on compute subnets = immediate sell pressure
• The 41% miner allocation is arbitrary. Different business models need different splits
• Fish (who started burns) agreed. Burns usually mean the validation isn't mature enough
The bigger point. You can't police burns. Subnets just send to their own keys instead of the burn address. Subnet 28 does exactly that.
Sam's position: judge subnets on outcomes, not process.
Const has changed the protocol 9-10 times in 2 years. That iteration speed is Bittensor's actual moat.
The whole Covenant thesis is playing out in real time.
TAO is up 100%+ in a month. Jensen Huang name-dropped the network. Grayscale has an ETF filing.
But the real story is three subnets quietly building every layer of decentralized AI.
Mandrake The Magician retweeted
Covenant Labs just did a 90-minute AMA breaking down their 3 Bittensor subnets.
@tplr_ai. @basilic_ai. @grail_ai. Pre-training, compute, and post-training under one roof.
Most people missed it. Here's everything they said.
Covenant is building what they call the "end to end intelligence continuum."
Three subnets. Three layers of the AI stack. All permissionless.
Templar (SN3) handles decentralized pre-training. Basilica (SN39) handles compute. Grail (SN81) handles RL post-training.
@DistStateAndMe, the lead, put it bluntly. Decentralized training is "humanity's last dance."
Not about beating OpenAI head to head. About creating optionality. About making it cheap enough for anyone to train models.
The gap between academia and frontier labs is growing exponentially. Researchers can't afford to experiment. The actual training run costs 5% of the reported budget. The other 95% is experimentation.
If Covenant cracks cheap training, that entire surface area opens up.
On Templar specifically:
• Hit 39% emission on Bittensor. Highest since Apex was the only subnet on the network
• Covenant-72B trained permissionlessly with 70+ contributors on commodity internet
• 1.1 trillion tokens processed. No centralized data center
• Performance competitive with LLaMA-2-70B
On Grail, something flew under the radar.
They built Pulse. A weight synchronization method that compresses model updates by 100x.
• In RL post-training, only ~1% of weights update per step
• Pulse exploits that sparsity. Lossless compression
• Prime Intellect's comparable system took 14 minutes to sync a 30B model
• Pulse makes decentralized RL training actually feasible at scale
• Already used by Cursor
The lead researcher on Grail said they've trained on math, code, and GPU kernels. Got 40-60% improvement on benchmarks. Working toward agentic training with 100K+ token context and 30B+ parameter models.
On Basilica, the compute subnet:
The team was blunt. Just reselling GPU hours is a 5-10% margin game. Traditional compute providers already do that.
Their play is value-added services.
• "GPU as code." No dashboard. No UI. Agents interact via SDK
• Custom scheduler that places workloads across heterogeneous hardware
• Verification checks for GPU, CPU, bandwidth, memory, storage, and OS security
• Partnerships with providers like Mass Compute for 10-20% below market pricing
• Miners compete on useful infrastructure, not just GPU hours
Sam then went on a rant about the miner burn debate.
His take: Bittensor had to grow up. dTAO introduced investors. The old "miners are God" philosophy doesn't hold.
• Subnet owners have a duty to protect token value
• Miners are a resource optimization exercise, not a cost reduction exercise
• 100% miner emissions on compute subnets = immediate sell pressure
• The 41% miner allocation is arbitrary. Different business models need different splits
• Fish (who started burns) agreed. Burns usually mean the validation isn't mature enough
The bigger point. You can't police burns. Subnets just send to their own keys instead of the burn address. Subnet 28 does exactly that.
Sam's position: judge subnets on outcomes, not process.
Const has changed the protocol 9-10 times in 2 years. That iteration speed is Bittensor's actual moat.
The whole Covenant thesis is playing out in real time.
TAO is up 100%+ in a month. Jensen Huang name-dropped the network. Grayscale has an ETF filing.
But the real story is three subnets quietly building every layer of decentralized AI.
Mandrake The Magician retweeted
Sat down with Bittensor co-founder Jacob Steeves (Const) for a deep dive into the state of Bittensor in 2026. We covered how Subnet 3 trained a 72B parameter model in a fully permissionless decentralized way, how subnets create markets for digital commodities like gradients and inference, the dynamic TAO tokenomics that lock emissions into subnet liquidity pools, and why open ownership of AI matters more than ever as centralized labs head toward trillion-dollar IPOs.
0:00 Intro: Bittensor Co-Founder Jacob Steeves
1:13 Jacob's Background, From Deep Learning to Crypto
4:45 Founding Bittensor: The Monetary Computer
7:58 How Subnets Work as Markets for Digital Commodities
11:18 Subnet 3: Decentralized Model Training
15:28 Permissionless Training at Internet Scale
18:52 Why This Matters: Open Source AI Ownership
25:53 Subnet Tour: Optimization, Inference, and Compute
29:55 Affine: Beating Qwen 30B via Market Incentives
34:26 TAO Tokenomics and Dynamic TAO
51:54 How Investors Participate: Just Buy TAO
54:05 Open Ownership vs Fiat AI
Mandrake The Magician retweeted
Templar builds the foundational knowledge. Grail sharpens the reasoning through RL post-training.
@erfan_mhi's PULSE paper showed how to make distributed RL practical by cutting weight sync bandwidth by 100x. This week @FireworksAI_HQ confirmed the same approach works at 1T scale for @cursor_ai Composer 2.
When Templar's next models are ready, Grail's infrastructure is proven at scale.
Paper: arxiv.org/abs/2602.03839
Pretty wild to see our work on PULSE show up in a real 1T-scale post-training run done by @cursor_ai.
Cursor built Composer 2 in collaboration with Fireworks and trained it across multiple datacenters, getting huge savings by syncing only the weights that actually changed between RL checkpoints. Fireworks reports that more than 98% of BF16 weights can stay bit-identical from one checkpoint to the next, and they cited our paper on this, too.
That is basically the exact sparsity pattern we showed in our paper, where we introduced PULSE, a lossless method for 100x more efficient weight-sync communication for RL training. Their system is very close to this idea in practice: exploiting the fact that only a tiny fraction of weights actually change between RL steps.
The deeper reason for this is not that RL gradients are sparse. They are not. The gradients are still dense. What becomes sparse is the realized weight update. In RL, learning rates are tiny, and with Adam, the update size stays bounded around the learning rate. Then BF16 adds a hard threshold: if the update is too small relative to the weight, it just rounds away, and the stored weight does not change at all. So from one checkpoint to the next, most of the model literally stays identical.
That is why this is such a useful systems idea. Lower precision, like using BF16, does not just save compute. It can also save communication, because more tiny updates get absorbed and fewer weights need to be shipped. At that point, compute efficiency and comms efficiency stop being a tradeoff. They start reinforcing each other.
If you want the deeper story on why RL updates get this sparse, the theory behind it, and how to push weight-sync bandwidth down by 100x+, take a look at our paper:
arxiv.org/pdf/2602.03839
The Fireworks blog on Composer 2 that cited our work:
fireworks.ai/blog/frontier-r…
The animation is taken from Fireworks!
Mandrake The Magician retweeted
Full @const_reborn discussion with @notthreadguy on Bittensor 👇
(Skip to 28 mins)
Its a great jumping off point for those looking to understand what #bittensor is all about
Enjoy 🍻
$TAO
nitter.gq/notthreadguy/status/20…
IM LIVE: Trump's TACOs Aren't Working... - IRAN vs USA Continues, TAO is RIPPING, Crypto is BACK!? Interviewing @const_reborn (4pm et)
Twitch.tv/threadguy nitter.gq/i/broadcasts/1nJOLEnDg…
