橘子 _

I recently came across a technical case study on NodeGhost by @Acurast and saw a pretty interesting test result.

A phone running a quantized Qwen2.5-3B model generated about 3 tokens per second.

Answering a question of around 120 tokens took about 45 seconds.

By today’s standards for using large language models, that’s definitely a bit slow.

But what caught my attention about this case was another part of the design.

With Acurast Tunnel, a model that’s normally tucked away on a phone with no public IP address can provide an HTTPS interface to the outside world.

Then, by using NodeGhost’s OpenAI API compatibility, developers can send requests to run on the phone with almost no changes to their existing calling methods.

This means a phone is no longer just a Processor passively waiting for tasks to be assigned over the network—it can also provide standardized AI inference services to the outside world.

Of course, this case is still a reference implementation on the Canary network, with significant limitations in speed and complex reasoning capabilities.

But I think it already answers a very practical question:

If developers want to deploy their own models on a distributed network of phones, can existing applications call them easily?

At least there’s now a working example of how this could be done.

As for commercialization, latency, concurrency, costs, and long-term stability still need further validation.

Acurast’s $ACU supports network usage and computing incentives, but whether it can ultimately attract paying demand will depend on whether these applications are genuinely competitive.

I recently invited someone to try @Acurast and ran into a pretty practical issue.

Their phone connected successfully, but it still wasn’t counted as a valid user in the dashboard. The page said they needed to complete 3 more tasks.

At first, it’s easy to assume that once you scan the code and the device shows as online, you’re all set.

But after looking into how Acurast works, I realized there are a few things to distinguish.

First, connecting a phone to the network as a Processor means providing computing resources.

The device’s hardware performance affects the base $ACU rewards, and actually running Deployments can earn additional reward weight.

Acurast also has a Cloud Rebellion community campaign, where completing tasks, inviting users, and connecting devices can earn MIST points.

MIST is a campaign point, not something you can count directly as $ACU rewards. Whether an invitation qualifies also depends on the specific conditions of the campaign.

So now I think it’s best not to judge participation in Acurast solely by whether a device connected successfully.

You should also check the Processor’s operating status, reward records, and whether campaign tasks have actually been completed.

For people who are new to DePIN, these things can be easy to mix up.

Acurast has made it possible for everyday people to contribute to a computing network with their phones, but there are still plenty of details to figure out between connecting a device and understanding the reward rules.

It’s better to get one device up and running properly before inviting more people to join. That can save you a lot of unnecessary hassle.

You can also use my link: https://hub.acurast.com/rebellion?ref=kxxd8r

I recently went back through @Acurast’s mechanism and noticed a detail about $ACU that’s pretty easy to overlook.

After developers deploy tasks to Acurast, they need to pay for computation.

But this money does not simply move from a developer’s wallet to the wallet of a single phone.

When a task is actually executed, the compute fee is consumed; the Processor providing the computing power then competes for network rewards based on hardware Benchmark, task execution, and whether it participates in Staked Compute.

I think this is worth looking at separately.

Because it splits two things apart:

On one side, whether developers have real tasks. If no one deploys apps and no one consumes compute resources, the usage demand for $ACU is naturally limited.

On the other side, why phones are willing to keep providing computing power. Device performance, task execution, and staking all affect their weight in the reward pool.

Right now, Acurast’s official site already shows more than 322,000 phones connected, with cumulative Compute Deployments exceeding 675,000.

The node scale is already there, and next I care more about this number:

How much computation is actually needed by developers, and how much of it they are willing to keep paying to run.

That’s also one of the things I value most when evaluating $ACU.

It is used for both Deployment and network fees, and also participates in Staked Compute, rewards, and on-chain governance.

So when I research $ACU, I won’t just focus on staking APR or circulating supply.

The more compute tasks run, the more real token-use cases will appear in the network.

If developer demand can keep growing in the future, with phones providing computing power, tasks consuming ACU, and the network rewarding compute providers, then this loop will become increasingly worth studying.

#depin #acurast

Recently I’ve looked into quite a few DePIN projects and found a problem that’s easy to overlook.

Many projects emphasize how many nodes, how many devices, and how many countries they cover—but not many people ask one key follow-up:

After these devices are connected, is anyone actually willing to pay to use them?

Lately, in my research on @Acurast and $ACU, I’ve been focusing on this.

Acurast went live on the mainnet on January 20 this year. The network uses computing resources provided by phones distributed around the globe. Developers can deploy compute tasks; the devices execute them, while the protocol handles scheduling and settlement.

What’s especially worth studying is its token consumption mechanism.

Developers need to pay ACU to use computing power. After a task is successfully executed, the corresponding compute fee is burned.

Devices that provide computing resources earn rewards through network incentives. Participating in staked computation can also yield additional rewards based on hardware performance, staked amount, and the duration of the commitment.

Here’s a very practical relationship:

The more users there are, the greater the burn demand generated by compute fees. But at the same time, to maintain computing power supply, the network itself will continue issuing tokens for incentives.

So if you only look at the burn mechanism, that’s really not enough.

$ACU’s initial supply is 1 billion tokens, and its tokenomics model also includes ongoing inflation.

How much of the newly issued supply can be offset by consumption driven by real compute demand in the future is the data I want to keep tracking.

There’s another detail: in March this year, Acurast introduced dynamic pricing via on-chain governance.

The computing ability of different phones is measured by hardware benchmark tests. Compute prices adjust according to device performance, no longer relying entirely on node operators’ manual quoting.

If high-performance devices and low-performance devices hold roughly similar pricing for a long time, it becomes hard for developers to judge costs rationally—and high-quality compute power may also lack sufficient incentive to join the network.

At present, Acurast already has the mainnet, compute task scheduling, fee burn upon charging, and staking governance mechanisms.

But for my long-term view of $ACU, I’ll still keep my focus on real paid tasks, network utilization, and the scale of fee burning.

After all, the number of phones connected can prove the supply size, and only when people keep paying can we further prove whether this business has a market.

When it comes to DePIN, I’m actually curious: just how much value can an old smartphone, kept in a drawer collecting dust, still generate?

Recently, I looked into @Acurast’s research and specifically checked its phone node rules.

The bar is indeed not high. Compatible devices running Android 12+ and iOS 15+ can all stand a chance to participate.

It also divides participation into two categories:

Lite can use the idle time of everyday phones to provide computing power, without needing to clear the device.

Core is better suited for Android phones dedicated to running nodes, which require restoring the device to factory settings.

But there’s one detail I think is even more important than the onboarding process.

Acurast’s base rewards are already allocated according to hardware benchmarks, so it’s no longer as simple as “one phone equals a fixed income.”

Devices with better performance usually receive higher weight, and they’re then matched to computation tasks in practice, with additional score boosts. Staked compute power involves another portion of the $ACU incentives.

So if you want to participate, I’d actually avoid buying a bunch of phones right away.

Test the devices you already have on hand first. Check the hardware score, uptime rate, and actual rewards—then factor in electricity costs, battery wear, and maintenance expenses.

After all, as the number of nodes increases, rewards become competitive. Your earnings won’t be guaranteed just because the devices stay powered on.

I think the thing worth paying attention to with @Acurast is that it enables ordinary people to participate in supplying compute.

As for how much you can ultimately earn, it still depends on device quality, network rules, and real-world operating conditions.

Sometimes, putting an old phone to use is more suitable for validating a project than rushing to invest in new hardware.

Recently I’ve been looking at on-chain AI, and I’m growing increasingly skeptical of that simplistic idea that “smart money is buying.”

Just because an address bought something doesn’t necessarily mean it’s bullish.

It could be a market maker replenishing inventory. It could be an intermediate address used for cross-chain settlement. Or it could simply be the same entity moving assets between different wallets. On-chain records don’t lie—but explaining every inflow as a viewpoint is still a long way from understanding real transactions.

This is also the hardest layer for Crypto-specific models: they can’t just read transactions. They have to identify relationships between addresses, understand why funds moved, and know that the meaning of the same action can be completely different across different times and different liquidity environments.

I follow Velvet @Velvet_Capital and the training of Velvet-1. I’m not trying to get it to guess the next coin with the biggest price increase. I’d rather know whether, inside Copilot, it can separate these easy-to-confuse behaviors: which are genuine buy orders, which are merely routing, and which funds come in without taking any price risk.

Once you can judge at this layer, AI starts to get close to being a research tool.

Otherwise, so-called on-chain insights are very likely just restating what the block explorer already shows.

One of the most overestimated things about trading AI is that it can lay out a very smooth-sounding logic.

But on-chain, many good trades aren’t good because the story is seamless.

When an asset goes up, and AI only tells you things like “increased capital attention” or “market sentiment is heating up,” that’s not very useful. I’d rather see: what *didn’t* happen.

It rose, but liquidity didn’t keep up.

People were calling for it, but the old wallets didn’t keep buying.

Trading volume increased, but most of it came from just a few large orders.

Price made a new high, but on-chain holdings are actually even more concentrated.

These “things that feel off” are often more valuable than pretty conclusions.

So my expectations for tools like Velvet @Velvet_Capital AI Copilot aren’t for it to summarize the market for me.

I’d rather have it act like a trading assistant that throws cold water—before I place an order—bringing up the gaps I didn’t notice.

The most expensive mistake in crypto isn’t missing opportunities.

It’s after you spot an opportunity, automatically ignoring all the evidence that doesn’t support your own judgment.

The most annoying thing in on-chain trading isn’t failing to find a token.

It’s that once you’ve found one, you still have to keep watching it.

Many opportunities aren’t suitable to buy immediately at the start, so you can only put them into your watchlist first:

Will the wallet activity keep coming in?
Will liquidity get thicker?
Is there any KOL discussion spreading?
Is the price still sitting at a spot nobody has noticed yet?

The problem is, no one can keep staring at it forever.
You go to sleep, open a meeting, go out for dinner—then come back to find the conditions have changed, but the most critical window has already passed.

So when I look at Velvet @Velvet_Capital working on an AI Copilot and its own on-chain model, I don’t really want to interpret it as:

“AI helps you catch 100x coins.” That sounds like nonsense.

What I want instead is something that can help me monitor conditions.

For example, if I’m watching a new token on Arbitrum or Base, I don’t want the AI to tell me whether I should buy it.
I want it to remind me when conditions change:

Liquidity suddenly spikes;
A few old wallets start buying in;
Trading activity is no longer just low-volume wash trading;
Social attention has already clearly moved to lag behind the price.

Something like this is simple, but it’s genuinely useful.

What traders spend most of their energy on every day isn’t making a single decision—it’s maintaining dozens of half-formed judgments.

Which ones are still worth watching, and which ones have already stopped working;
Which signals are just noise, and which ones need to be re-evaluated.

Human brains easily mix these together, and in the end you trade based on instinct.

If Velvet’s AI can help on this layer, I think it’s more practical than “giving me an answer.”

Many trades aren’t missing answers—they’re missing an observer who stays clear-headed, doesn’t get emotional, and doesn’t get fed up.

Sometimes, a great tool isn’t meant to make you more aggressive—it’s meant to help you not miss the things you were already seeing, but would otherwise overlook.

When many people hear “RWA,” the first thing that comes to mind is usually yield, U.S. Treasuries, real estate, and credit.

But collectible-based RWA can’t be viewed that way. The earliest source of value for collectibles wasn’t cash flow—it was culture, aesthetics, scarcity, and community consensus.

Why would someone want to buy a card?

Not because it’s like a bond, and not because you can calculate a stable return curve.

It’s because someone genuinely wants to own it, show it off, trade it—and even treat it as part of their identity and interests.

That’s also what I find interesting about Renaiss @renaissxyz.

It doesn’t plug into traditional financial assets—it moves into a market that already has real users, real emotions, and established offline trading habits.

Pokemon, One Piece—TCGs like these are easy for regular users to understand, and collectors also have their own set of judgment criteria.

The problem wasn’t that nobody liked these things. The issue was that their value has historically been difficult to distribute, price, and manage more efficiently.

So when I look at Renaiss, I don’t really like summarizing it with the three words “financialization.”

It’s more like: on top of the collectibles’ existing market, add another layer of tools:

Can the asset’s status be made clearer?

Can the price reference be more trustworthy?

Can the trading radius be expanded?

Can users rely less on gut feeling and more on information to make judgments?

If you get this wrong, of course it turns collectibles into packaged short-term chips.

But if you get it right, it can actually make collectors much clearer about what they’re buying and holding—and what price the market is ultimately willing to give them.

That’s also the difference between Renaiss and many purely narrative RWA projects.

It’s facing a group of collectors that already exists, rather than first creating an asset and then spending time educating the market to believe it has value.

That said, the yardsticks are also very practical: are users really more likely to discover good assets? Is pricing more transparent? Is trading smoother? And in the end, are collectors willing to stay for the long term?

Alright, first Robinhood, then Stable, and now Arc

After a roundabout detour, the funds ultimately end up back at BSC memes

Round after round of storytelling

And big brother @CZ said it too—over the next few weeks, he may personally buy or sell one or two, to test out new things

May the best meme win

Believe $BNB is buying—$BNB , let’s build together!

Many people look at Hertzflow and their first reaction is: trading, joining a pool, doing invitations—which one should I choose?
But if you switch perspectives, it might be simpler: what role do you want to enter @Hertzflow_xyz as, and do it earlier?
For trading-oriented users, you can start with the product itself.
What Hertzflow wants to do isn’t just opening long and short positions—it’s to bring more Oracle-supported assets into the on-chain markets.
In the future, on-chain perp trading may not be limited to only BTC, ETH, and a few mainstream coins.
The earlier you get familiar with the product, the easier it is to find your trading rhythm when new markets emerge.
For capital-oriented users, focus on Pools and Vaults.
If Hertzflow is going to build yield-generating assets in the BNB ecosystem, liquidity is definitely core.
A Pool corresponds to a specific market, while a Vault concentrates different yield structures a bit more.
For regular users, it’s not just depositing funds and waiting for returns—it’s providing liquidity to the market while taking on risk.
For community-oriented users, the invitation system may be lighter.
If you don’t have a large amount of capital and high-frequency trading isn’t suitable, you can bring new users in through invite codes, tutorials, and community Q&A.
A truly effective invitation isn’t just dropping a link—it’s helping others understand how to use Hertzflow, where the risks are, and whether it fits them.
So in Hertzflow’s early stage, the opportunity may not be to build every feature all at once.
Finding the role that suits you comes first. Some people trade, some provide liquidity, and some are responsible for explaining the product clearly.
Whether these roles can form a cycle is what will determine whether Hertzflow ultimately becomes a common trading protocol—or a new on-chain trading entry point in the BNB ecosystem.

Verifying my Binance Square account for YZi Labs Amplify via EchoHunt: EH-5VYA2M

Ondo Chain doesn’t stay stuck in the narrative of “building a chain for RWA,” but instead evolves further into the Ondo Network

The problems it addresses are very straightforward:

On-chain, it can achieve non-custodial operations and public verification, but for long periods the matching speed, transaction privacy, and user experience have lagged behind CEXs.

Ondo Network (@Ondo) chooses to handle execution, verification, and settlement separately—while preserving asset self-custody and on-chain verifiability, it pushes the order-matching experience to CEX-level performance.

The first application running on this architecture is @OndoPerps

Recently, its 24-hour trading volume at one point exceeded Lighter—indicating that the market needs more than just “tokenized stocks,” but also leverage, liquidity, and trading tools built around these assets.

Ondo is moving from an RWA issuance platform toward a complete on-chain financial market infrastructure.🙌
$ONDO

📊 The premise of the compound interest effect is finding a vehicle for continuous positive accumulation.

COSM has several key conditions for compound growth:

🔹 Returns are continuously reinvested—what you earn today will be added on top tomorrow
🔹 Benefits keep getting updated—each time you participate, you accumulate new entitlements
🔹 The ecosystem keeps expanding—the more users there are, the richer the resources, and the thicker the value
🔹 Ongoing institutional support—this isn’t a one-time investment, but long-term co-building

Improve a little every day, and over the course of a year the gap becomes exponentially large.
COSM is creating a compound-interest environment for long-term participants.
The earlier you enter, the longer the compound-interest cycle.

#COSM

There’s a real issue with on-chain trading: self-custody feels secure, but the trading experience and execution efficiency often can’t keep up.

@grvt_io uses a hybrid architecture of off-chain order matching and on-chain settlement. Trading speed is handled by a high-performance matching system, while funds custody, margin, and risk management are executed by on-chain contracts—together with ZK technology to protect trading data.

For traders, this route is quite practical. You don’t have to hand full control of asset custody over to the platform, and you also don’t have to endure complicated and slow experiences in the name of “decentralization.”

With safety, privacy, and execution efficiency addressed at the same time, on-chain derivatives have a chance to truly reach a larger market.

#grvt