Sτeve

Into everything crypto!

[Sτeve @blockhash7]

What this “burn or lock buybacks?” question makes me think of is an old Bittensor narrative I’ve been missing for a while.

Back in 2024, @const_reborn, @KeithSingery and other Bittensor opinion leaders used to describe $TAO as something close to a claim on an inexhaustible well of intelligence: own a piece of $TAO, and you effectively own a piece of whatever commodities the network keeps producing.

With Dynamic $TAO, that narrative somehow faded into the background.
Maybe it’s time to bring it back — but at the subnet level.

What if Alpha wasn’t merely a token whose value reflects the perceived value of a subnet? What if it represented an actual claim on the subnet’s output?
If I own 1% of all Chutes Alpha, why shouldn’t that ownership entitle me to a proportional share of the compute Chutes produces?

The Alpha stays. You consume the fruit, not the tree.

And crucially, I wouldn’t have to pay again every time I use the service. I already paid for the productive asset that generates my claim.

That would turn Alpha from “a token tied to a subnet” into something much more tangible: a perpetual claim on the commodity that subnet produces.

Own the Alpha. Keep the Alpha. Consume the intelligence.
Now that sounds like the old Bittensor promise again: a spring of intelligence that never stops flowing.

[Sτeve @blockhash7]

@0xSunRun @lcnxyz @const_reborn @KeithSingery I agree. There’s nothing wrong with that. And many alpha holders would be perfectly happy for the subnet owner to monetize their share of the intelligence produced on their behalf.

[Sτeve @blockhash7]

@0xSunRun @lcnxyz @const_reborn @KeithSingery Absolutely. If I don’t use my share of the intelligence produced, the subnet could monetize that unused capacity on my behalf and use the proceeds for buybacks.

Or, if you ask me, even better: reinvest it straight back into growing the subnet.

[Leo Lucian 🉐 @lcnxyz]

Burn or lock buybacks?

we have always tried to experiment with chain upgrades, so when conviction launched we saw the benefits to lock alpha acquired with revenue

this meant showing commitment, cementing our spot as the subnet operators and takes alpha out of circulation through the locked stake yield

when presenting this to the community, great points were made for both burn and lock options however, ultimately consensus was ‘burn is better’

admittedly it’s a small sample size of opinions so i’d like to get more data: What is everyone’s opinion on burn vs lock for subnets generating revenue?

[vaN ττ @vaNlabs]

I set up @lium_io [SN51] today for some GPU compute I need for a project.

I've used a fair few GPU providers at this point, and this was probably the smoothest setup I've had.

Created an account, topped up with $TAO, generated an API key and handed the docs to my agent. A few minutes later it was spinning up GPUs.

The part I really liked is how agent-friendly everything is. llms.txt, clean docs, JSON output, budget caps and TTLs. There was no fighting with dashboards or figuring out some obscure API.

Well done to the team. This is how infrastructure should be built now.

[Florian S @airesearch12]

@markjeffrey If someone would give me a subnet slot (can't afford myself), I'd be happy to build that subnet.
I think I might be a very qualified person, as the maintainer of JevBench, the biggest benchmark for Jev-class models.
https://t.co/00vs9BiR2U

[const @const_reborn]

The most dangerous mind in history is the one aligned to a state.

[σℓαigє τ @_olaige]

Remember when Jon tweeted that he had a "once-in-a-generation-holy-shit idea"?

Well, that idea was Parallax, and they've been absolutely crushing it.

Currently at 316.3 billion tokens trained, $11/b tokens, and 62.6% peak MFU.

Don't fade Jon, Parallax, Chutes (SN64) and $TAO.

[Good Morning @say_gm_]

GPT-6 Astra is live on Gm.

OpenAI’s smartest model yet. Tell it what you need done. It opens the browser, writes the code, runs the terminal, ships the result.

Routed through a verifiable TEE, at 8.3% discount on SayGm.

https://t.co/sk0GHhAmTh

[Barry Silbert @BarrySilbert]

Eventually investors will connect the dots to Bittensor bittensor:native

[Chutes @chutes_ai]

We’ve been thinking a lot about this, lately... How do you build something that can last a thousand years?

Almost no company has done it. Most countries haven't either.

Chutes must become a network that outlasts any single company, team, or founder. Including us.

[Andy ττ @bittingthembits]

🚨 $TAO SN51 @lium_io

Lium just did something almost nobody else on earth can do.

They put NVIDIA B300 SXM6 on demand.

Not waitlisted. Not "contact sales." Not reserved for enterprise accounts with seven-figure commitments.

Rent it in under a minute.

Why:

B300 is Nvidia's newest Blackwell Ultra chip. Most major clouds are still rationing H100 and H200 supply. B300 inventory anywhere close to on-demand is close to nonexistent right now.

Lium has 352 pods live.

The pricing:

1x B300, $8/hour
8x B300, $64/hour, roughly $8/GPU

Real specs on the listings. 2,149 GB VRAM on the 8x configs.

Locations in Helsinki and Reston, uptime and reliability published per node, provider hotkey visible for every single pod.

No black box. No TRUST US!

Install is one line:

pip install https://t.co/MZL1qmXlAh && lium up --gpu A100

Same permissionless access whether you want a single card or a full cluster.

This is the part that keeps getting missed about Bittensor subnets.

People assume decentralized compute means leftover consumer GPUs nobody else wants.

Lium proved the opposite.

They're routing some of the scarcest, most in-demand silicon on the planet through a subnet, with pricing and access that centralized providers can't currently match.

Scarce hardware plus permissionless access plus transparent pricing is not a small thing.

That's infrastructure the whole AI industry needs, sitting on Bittensor mainnet right now.

$TAO | SN51 | DYOR.

[mogmachine (ττ) @mogmachine]

If Open Router is valued at $7bn and @say_gm_ does what they do but on average 45% cheaper and without sniffing your prompts.

It has to be worth checking out: https://t.co/g2cVeQhTX3

[@jason @Jason]

Predict bittensor:native models will consistently be top 20 vs their contemporaries over the next 12 months, while being cheaper than the average model in that category

Over 24 months, there is a non-zero chance they show up in the top ten consistently, and like Bitcoin, they shock the world in the amount of compute being added to the network

If you have extra compute, add it to @bittensor and you’ll create AI sovereignty while minting an accelerating token (or two!)

Follow @const_reborn and @markjeffrey

Donate to the vision by buying one bittensor:native — be delighted if it goes 10-100x as a pure lottery ticket

[lium.io @lium_io]

we have the cheapest h200s on the market, still a couple left for under $3/hr.

The only place in the world to get on-demand H200s for this price.

[lium.io @lium_io]

Calling all AI/ML engineers. If you are working on a cool project, we would love to be a part of it to fulfull your GPU needs.

Fill out this form for a GPU credit grant to help you build.
We have it all: H100s, H200s, B200s, or whatever else you need.
https://t.co/qztCED24Lg

[engy.ai @engyai]

Kimi K3 is live on https://t.co/bsUzzt6L9y.

next: DeepSeek V4 Flash and Qwen3.8.

[Algod @AlgodTrading]

@engyai launched the cheapest Kimi K3 on the market, if you’re using K3 on openrouter you can now save up to 50%

Tps/ttft is similar to the official moonshot api

[Sτeve @blockhash7]

@TheTNetHunter You can still stake directly to individual subnets. Root Reborn only changes how staking to Root works.

[lium.io @lium_io]

can't find high quality gpus available anywhere?

over 100 b300s and 100 h200s available now for rental on lium!