Acurast has onboarded 335,000+ smartphones while running real deployments instead of just incentivized testnet activity. @Acurast is a decentralized compute network that turns ordinary phones into verifiable compute nodes using Trusted Execution Environments. What changes: ▸ compute supply comes from existing consumer devices instead of new data centers or GPU clusters ▸ execution is confidential and verifiable because workloads run inside the phone’s TEE ▸ developers can deploy serverless jobs, APIs, and LLM inference without managing infrastructure ▸ providers earn $ACU by contributing idle or spare phone capacity The data (as of late 2026): ▸ 335,000+ phones onboarded ▸ 1.1B+ on-chain transactions ▸ 685,000+ deployments ▸ presence across 175+ countries What it means: ▸ capacity can scale by adding phones people already own ➜ lower marginal cost than traditional compute new workloads are possible, but sustained demand still depends on developers actually using the network beyond early experiments My take: if Acurast moves from mostly provider-side growth to consistent real workload usage, it becomes one of the more practical decentralized compute layers. that only happens if confidential and verifiable execution creates enough demand that people keep phones online for more than just rewards. https://t.co/QbE53D1PaC

I think people don’t understand how crazy @trdEverything’s growth is They just cleared $40B in total trading volume That’s more than many perps platforms do in their first year: - Started with $334M in month 1 - Hit $10B in just 60 days - Now $40B in ~7 months And they’re still accelerating every month...

I’ve always liked the idea of crypto networks being used to improve real AI systems instead of just generating more content. At the very least, it’s a more useful application than most speculative AI tokens. @perturbaix is taking that idea pretty far with continuous adversarial testing and hardening. Think of it as turning model vulnerabilities into an ongoing, incentivized process, where independent miners keep finding ways to break models and the findings get used to make those models stronger. That could change how teams approach robustness, since adversarial testing usually: → Happens internally or once in a while → Relies on limited in-house expertise → Stops after a single report Perturb’s idea is to make that pressure continuous, verifiable, and easier for teams to act on. But the upside goes beyond the subnet. The output is tied to making production AI systems less fragile before they reach users. This is the side of decentralized AI I’d love to see grow. $TAO #Perturb #SN26

Things are moving fast for @alpharidge_ai right now. From a quiet early-access tool a couple of months ago → the product going live with paying users → the AI analyst and RidgeScript landing in one workspace → and now the news layer from SN45 sitting inside the screener. Stocks, crypto, and FX. Charts, screening, and rules that keep running with the browser closed. The part I still like most is the news engine: miners process ~290k articles a day → articles get clustered into real events → sentiment and impact become screenable columns. If AlphaRidge can keep turning those signals into something traders actually filter on, this mechanic will get a lot more interesting. But first, the live product with real users is still the real checkpoint. Watching how the research-to-alert workflow lands. 👀 $TAO #SN45 #Alpharidge

➥ AI security is entering the next phase, and @perturbaix is one project I'm keeping an eye on. Until now, adversarial testing has mainly been about finding attacks with internal teams or open-source tools. But Perturb is tackling something more interesting: continuous, incentivized hardening. ▸ Global miners compete to find real adversarial examples ▸ Verified attacks get used for model training, not just reports ▸ Robustness scores and datasets companies can act on ▸ Moving from subnet activity into an actual product This is the shift from one-time red teaming → ongoing external pressure that I think production AI needs. I'm currently watching how their product launch develops and whether teams outside the subnet actually use the output. #TAO #SN26 #Perturb

Up to $1 of alpha for every $1 a referred developer spends SayGTM just went live on @say_gm_, Bittensor subnet 28. A share of emissions now sits in an on-chain pool for people who bring paying customers. a few things i think are worth paying attention to: ▸ Proof of Customer: only real top-ups count. Signups and clicks pay nothing. ▸ Cap: $1 of SN28 alpha per $1 spent in the first 30 days, priced at the round's close. The rate drops if referred spend outgrows the pool. ▸ Split: 10% of that round's referred revenue = 10% of the alpha allocated. Same rate for everyone. ▸ Holds: 7 days on crypto top-ups, 120 days on card. Payout lands on your coldkey as alpha, then moves with the market. How to participate: ➜ Sign in with a Bittensor coldkey ➜ grab your referral link ➜ send developers who actually top up. Pls read this carefully: Only accounts created after launch, attributed to your link, and paying inside their first 30 days qualify. Shared card or payer coldkey gets grouped, and only the earliest one counts. Self-referral is allowed, but it follows the same cap. Let me explain what SayGTM is. SayGM's go-to-market pool. Affiliates mine revenue instead of clicks: emissions pay the people who bring customers that pay. Personally, i find tying a slice of subnet emissions to referred spend more interesting than a 3% cut of nothing. Still early, and the 1:1 rate is a ceiling, not a promise.

honored to be working with @pantheonvaults. pantheon vaults are tokenized creator indexes. each creator gets their own vault. community stakes in. value flows back to the person actually building the audience. not another points farm. a financial layer for web3 creators. my vaults solana:3xaDR6gxCp92SZRJ1zMigoRAUMkQGNmFF9HvGT94pump are live. the model is simple, and it fits how this industry actually works. creators bring distribution. the vault turns that into an economic network. early, but the mechanism is the interesting part.

I think the way @CrucibleLabs is shipping Crucible Wallet shows a pretty clear playbook for Bittensor access. Instead of a separate tool for every action, they are opening more ways to use the same TAO position. ➜ Crucible Wallet is now expanding in 3 directions: • Chrome extension covers the daily flow: unified swap, subnet views, claim rewards, portfolio PnL. • iOS and Android take the same wallet off the desk, with Ledger over Bluetooth. • Smart Allocator puts root rewards to work across subnets without a new product or a spreadsheet. One wallet, more surfaces, more distribution, more utility.

Green Compute gives stranded renewable power another place to work. If a farm has biogas, solar or hydro it cannot fully sell into the grid, @green_compute_ puts permanent GPUs next to that power. An enterprise client pays for inference. A slice of that revenue buys and locks ALPHA. A few ways the pieces plug in: ▸ Energy sites: UK farms already producing power from waste can host RTX 4090/5090 clusters instead of curtailing output or selling it cheaply back to the grid. ▸ Enterprise buyers: clients that need identical servers, on-site support and capacity still online in 6–24 months rent dedicated green compute, not spare GPUs. ▸ Subnet participants: infrastructure miners hold multi-year hardware, while a percentage of contract revenue is routed into ALPHA buybacks. The design is the neat part. The power is already there and often constrained. The GPUs are permanent, not idle capacity. Revenue does not depend on emissions to make the site viable, so more contracted inference widens buy pressure instead of diluting it. At Exploit, Josh said the prior $3M site is live, with another 240 RTX 5090s. The next UK site is another $4M, targeted for the end of October, and they think it could add roughly $300k in monthly revenue. They also flagged a bigger buyback as commercial clients and the iota work expand. Clients can contract the capacity directly. Miners can keep the hardware pointed at SN110. ALPHA still needs real revenue behind the buybacks, and the new site still needs those contracts to fill.

GM, what I find interesting about the latest @green_compute_ update isn’t really the $4M data centre itself. It’s what they chose to build. I’ve seen plenty of compute subnets optimize for: emissions → temporary miners → spare GPUs With Green Compute, capacity only really counts once it’s permanent, powered by verified renewables, and pointed at commercial clients that need it months from now. So a GPU count by itself doesn’t mean much unless it converts into contracted inference. And I think that fits pretty well with how enterprise AI compute is starting to scale. On the demand side, the loop looks like: client → contract → inference → revenue But the supply side needs its own loop too: stranded green power → permanent GPUs → usable capacity → ALPHA buybacks Seen this way, the Exploit announcement looks like a mechanism for turning renewable sites into commercial compute. At the Montreal talk Josh said the prior $3M farm site is already live, with another 240 RTX 5090s. The next UK site is another $4M, targeted for the end of October, and they think it could add roughly $300k in monthly revenue. They also flagged a bigger buyback as clients expand, including work with iota and new commercial inference accounts. Earlier, 10% of contract revenue was already being routed into ALPHA. That’s probably the metric worth watching over time. Not how many cards get announced. How much of that new capacity actually turns into revenue, and how much of that revenue comes back as buy pressure. How are you reading the $3M live site vs the $4M October plan? 👀