Hyperliquid has 43% of daily perp volume. But 57% of the market’s open interest. Across onchain perps: → Total OI: $14.6B → Hyperliquid OI: $8.38B → Daily volume: $25.70B → Hyperliquid volume: $10.92B That gap is more important than the volume ranking. Hyperliquid isn’t just processing trades. A much larger share of the market’s outstanding positions is staying there. Volume tells you where traders are active. OI tells you where exposure remains.

What if the real RWA unlock isn’t tokenizing the asset, but tokenizing access to the revenue it generates? That’s the part of @dawninternet I find compelling. Connectivity infrastructure is expensive to deploy, but once it’s live, it produces something measurable: bandwidth being delivered and contracts being fulfilled. The USD.infra Vault is essentially trying to connect that predictable demand with on-chain capital. That shifts the conversation from: “Can we put real-world assets on-chain?” to: “Can on-chain capital help build more of the infrastructure the internet needs?”

Bitcoin is valuable collateral. But institutions holding BTC long term have often needed to navigate several disconnected systems to access dollar liquidity without selling it. Circle is bringing those operations into one workflow. On September 21, @Circle introduced Digital Asset-Backed Borrowing for eligible Circle Mint customers on @Arc and Ethereum. The process: • Deposit BTC and mint cirBTC • Supply cirBTC as collateral to a supported lending market • Borrow USDC against the collateral • Receive borrowed USDC directly in Circle Mint The institution retains its BTC exposure while accessing dollar liquidity. —— — Circle Controls The Workflow, Not The Loan Circle connects three financial functions: Bitcoin custody, wrapped-asset creation and stablecoin borrowing. cirBTC represents Bitcoin backed 1:1 by native BTC. USDC provides the borrowing liquidity, while Circle Mint connects institutions to third-party lending markets. Morpho is the first supported protocol, with Aave expected to follow. The distinction is where the financial responsibility sits. Circle facilitates the process. The lending market determines borrowing rates, collateral requirements and liquidation conditions. Circle doesn’t need to become the lender to make Bitcoin-backed credit accessible to its institutional customers. —— — Who Provides The Money? Consider an institution holding BTC that needs USDC for treasury operations. Instead of selling Bitcoin, it supplies cirBTC as collateral and borrows against the position. Several businesses participate: • Circle provides the wrapped asset, stablecoin and integrated workflow • Lending markets manage the credit infrastructure • Liquidity suppliers provide the USDC available for borrowing • The institution pays interest while retaining BTC exposure Each participant occupies a different part of the transaction. Circle benefits from greater use of its infrastructure, but the exact fees it earns from this borrowing service should not be assumed without disclosure. —— — Easier Borrowing Doesn't Remove Risk A simpler interface doesn’t eliminate the economics of collateralized lending. If BTC falls sufficiently, the borrower may need additional collateral or face liquidation. Other risks remain: • Borrowing costs • Wrapped-asset liquidity • Custody arrangements • Smart-contract exposure Circle’s 1:1 BTC backing addresses the reserve structure of cirBTC. It doesn’t eliminate the risks of borrowing against it. —— — Circle Can Grow Without Making Loans Circle is connecting Bitcoin collateral, dollar liquidity and lending infrastructure through an interface institutions can already use for treasury operations. BTC becomes usable collateral without an immediate sale. USDC becomes the borrowing and settlement asset. Third-party protocols provide the underlying credit markets. And Circle Mint brings everything together. Circle can expand the use of cirBTC and USDC without owning the loan book, while lending protocols gain institutional distribution without owning the customer relationship.

To end the month, @Circle just gave @Binance two ways to benefit from USDC adoption, and neither requires Binance to issue its own stablecoin. On September 17, Binance completed a $100M investment in Circle through a private placement, acquiring approximately 1.24M shares. The companies also expanded their USDC partnership to five years, with Circle agreeing to pay Binance monthly incentives linked to qualifying USDC balances. The economics are worth unpacking. Circle generates income from the reserves backing USDC. Binance controls one of the largest distribution channels in crypto, with users, trading infrastructure and wallets where those dollars can circulate. Under the expanded arrangement: - Circle benefits from additional USDC distribution and eligible balances. - Binance receives recurring incentives tied to qualifying holdings. - Its $100M equity investment also gives it exposure to Circle’s business. This effectively creates two revenue opportunities for Binance from the same stablecoin ecosystem. But there’s another side to this arrangement. For Circle, growing USDC supply increasingly involves sharing reserve economics with the platforms responsible for distribution. Higher USDC supply doesn’t necessarily translate into proportionally higher retained income if distribution costs increase. The actual economics depend on the incentive rates, eligible balances and reserve yields, none of which can be fully calculated from the disclosed terms. — My Take Stablecoin distribution is becoming a business model in its own right. Exchanges and wallets don’t need to become issuers to capture value from the dollars circulating through their platforms. And the growing importance of these distribution agreements raises a question worth examining: How much of the stablecoin industry’s economics will ultimately remain with issuers, and how much will flow to the platforms controlling access to users?

.@Aave’s size is impressive. But the amount of capital deposited into a lending protocol tells you something different from how much credit its users actually demand. According to Aave Insights, the protocol currently reports: • Total deposits: $33.2B • Outstanding loans: $13.0B • Implied borrowing-to-deposits ratio: 39.2% • Difference between deposits and loans: $20.2B Around 39% of Aave’s reported supplied value is matched by outstanding loans. That distinction is important when assessing the scale of onchain lending. —— — The Difference Between Capital Supply and Credit Demand Aave has built an enormous liquidity base, but depositing capital and borrowing against it are two separate economic activities. Suppliers contribute assets to earn yield. Borrowers create credit demand by paying interest to access liquidity. The relationship between those two groups determines how intensively the protocol’s capital is being used. Aave’s $33.2B in deposits demonstrates its ability to attract capital. Its $13B in outstanding loans provides a more direct measure of the credit being extended. The remaining $20.2B is the difference between those two figures, not necessarily liquidity that can immediately be borrowed. Availability depends on individual asset markets, borrowing caps and other constraints. —— — Where the Next Layer of Growth Comes From Consider two possible outcomes for Aave. • Deposits increase from $33B to $40B while borrowing remains around $13B. • Deposits remain around $33B while borrowing expands from $13B to $18B. Both represent growth, but they describe different developments. The first expands the protocol’s supplied capital. The second represents greater outstanding credit against an existing deposit base. And because borrowers pay interest, expanding loan balances can have a more direct relationship with lending fees and protocol revenue, depending on borrowing rates and reserve factors. —— — The Aggregate Number Doesn’t Tell the Whole Story Aave is not one homogeneous lending pool. A market-wide borrowing ratio can conceal substantial differences in how individual assets are used. Some pools can operate near their optimal utilization levels while others hold considerably more available liquidity. That is why the next level of analysis should examine where borrowing demand is concentrated, which assets account for available liquidity, and which markets generate the largest share of lending revenue. A single aggregate ratio is a starting point, not a complete assessment of capital efficiency. —— — My Take Aave’s scale is established. The more useful question is how much of that capital is actively supporting outstanding credit. $33.2B measures deposits. $13B measures outstanding loans. Those are two different dimensions of lending growth. And treating them as interchangeable makes it harder to see whether Aave’s next phase of expansion is coming from attracting more capital or putting more of its existing capital to work.

Jupiter added a lot more active card wallets last week. Spend barely moved with them. • Active wallets: from 7,800 to 9,235 (+18.4%) • Weekly spend: from $1.73M to $1.78M (+2.7%) That’s a pretty big gap. More people are using the card, but the dollars being spent aren’t growing anywhere near as fast. Which means the average active wallet is currently spending less than it was before. That doesn’t make the growth bad. It just tells you what kind of growth it is. Right now, @JupiterExchange is doing a better job expanding card distribution than deepening card usage. Acquisition came first. Usage hasn’t caught up yet. And for a payments product, that distinction is important. 9,235 active wallets looks good. How much those wallets actually spend tells you much more.

The easiest bull case for decentralized compute is GPU pricing. @akashnet reported: • A100 80GB: $1.54/GPU-hour • H100: ~$2.58/hour @LambdaAPI lists: • A100 80GB: $2.79/hour • H100 SXM: $3.99/hour That puts Akash roughly 45% cheaper on A100 and 35% cheaper on H100. Compelling. But enterprises don’t buy GPU-hours. They buy completed workloads. ——— Say I’m running an 8-GPU training job for seven days. At $2/GPU-hour I’m paying: 8 × 24 × 7 × $2 = $2,688 At $3/hour: 8 × 24 × 7 × $3 = $4,032 That’s $1,344 saved. Easy decision, right? Only if both jobs actually behave the same way. If the cheaper cluster disappears halfway through a training run, I suddenly care about a lot more than the $/GPU-hour. Where’s my latest checkpoint? Does the storage persist? Can I get another 8 identical GPUs immediately? Do I have to move the dataset? How long until the workload is running again? And how much engineer time did I just burn fixing all of this? That’s where the cheap GPU can become the expensive one. For production workloads you’re really buying some combination of: 1. GPU capacity 2. Availability 3. Persistence 4. Scheduling certainty 5. Recovery The pricing table only shows #1. ——— @akashnet is actually a good example of why I started thinking about this differently. The marketplace can expose cheap GPU capacity across independent providers. But persistent storage is provider-local. It can survive container and provider restarts, but not every failure scenario, lease termination, or migration between providers. I don’t necessarily see that as a criticism of Akash. It’s a tradeoff. You’re getting access to a distributed marketplace of compute instead of one vertically controlled cloud. The mistake is pretending those are identical products and then comparing them entirely on price. They’re not. And this is where I think the decentralized compute conversation gets more nuanced. Not every AI workload even needs the same reliability. You can probably tolerate more flexibility with: • rendering • batch inference • experiments • checkpointed training • non-urgent jobs If something gets interrupted and can cheaply resume elsewhere, price becomes extremely important. But try applying the same logic to: • persistent inference • customer-facing APIs • long distributed training • latency-sensitive workloads Now availability starts becoming part of the product itself. ——— Different workloads also need different reliability. Flexible workloads like: • rendering • batch inference • experiments • checkpointed training can tolerate interruptions if restarting is cheap. But for: • persistent inference • customer-facing APIs • distributed training • latency-sensitive workloads availability becomes part of the product. ——— Decentralized also doesn’t automatically mean interruptible. @ionet, for example, describes its on-demand instances as non-spot capacity. So the real competition may become reliability profiles, not just GPU prices. One network may win flexible workloads through lower costs. Another may charge more for stronger persistence or availability. Centralized clouds may remain worth the premium where downtime is expensive. ——— Which gets me to the metric I actually want to see. Not: $/GPU-hour But: cost per successfully completed workload. Show me: • what percentage of jobs completed without interruption • average restart/recovery time • capacity availability by GPU class • workload duration before interruption • checkpoint persistence • effective cost after failed/restarted jobs Then we can actually compare decentralized compute with centralized clouds on something enterprises care about. My Take: Cheap GPU-hours get decentralized compute into the conversation. But enterprise adoption will depend on whether that discount survives once availability, persistence and recovery are priced in. The networks delivering cheaper workloads win.

RWA tokenization is moving beyond Treasuries and real estate. @dawninternet is taking a different route: putting real digital infrastructure cash flows onchain. AI compute, connectivity, edge infrastructure capital goes into actual deployments, while the revenue generated by that infrastructure flows back through the financial product. That’s a much more interesting RWA model to me. The USD.infra Vault opens Sept. 22.

I’m building my sBTC position gradually. The plan is to use @bitflow’s DCA function instead of trying to nail every entry. @Stacks Sats has three ways to earn BTC rewards around market activity: → Daily Stack: a qualifying $25+ swap enters you into the daily BTC draw → Trading: rewards scale with your share of eligible volume → LP: liquidity near the active price earns more The LP part makes sense to me. Two positions can hold the same amount of capital, but the one sitting closer to where trades happen is more useful for execution. Bitflow rewards that difference. My approach is pretty simple. Build the position gradually, trade when there’s a setup, and consider LP when the economics justify the exposure. No need to force all three. PS: If you’re joining Stack Sats, register on Bitflow and pick the route that fits how you use the market. app.bitflow.finance/stack-sa…

As of today, @pendle_fi has $691.06M in TVL on Ethereum versus $213.37M on Monad. Ethereum holds roughly 3.2x more capital. But monthly fees are much further apart: → Ethereum: $476,014 → Monad: $37,206 That’s a 12.8x fee gap from only a 3.2x TVL gap. This is why TVL alone can be misleading. Two deployments can attract significant capital while producing very different levels of economic activity. ——— — What $1 of TVL Actually Produces Using current TVL as a rough denominator: → Ethereum: $689 monthly fees per $1M TVL → Monad: $174 That gives Ethereum roughly 4x higher fee intensity. It’s not a perfect efficiency metric because trailing fees are being compared with point-in-time TVL. But it still shows how differently the capital is being used. Pendle separates principal from future yield through PT and YT. So equal amounts of TVL can produce very different economics depending on how actively those positions trade. ——— — Where The Difference Comes From Imagine two Pendle markets with $100M each. One is mostly PT buyers holding to maturity. The other has traders repeatedly rotating between PT and YT as implied yields change. Both show $100M TVL. But the second generates more turnover and more fees. That’s what makes the Ethereum comparison interesting: it has 3.2x more capital, but 12.8x more fees. Its capital is simply more economically active. ——— — What Drives The Fee Gap Pendle deployments may share the same interface, but their market structure can be very different. Underlying assets, maturities, liquidity depth, implied yields and user behavior all affect turnover. Ethereum can support deeper secondary markets and more frequent repositioning. Monad can still attract large deposits while more capital remains parked until maturity. TVL counts both equally. Fees reveal the difference in activity. ——— — Why Expiry Matters Pendle markets naturally create recurring activity because PT and YT expire. At maturity, capital can: → Exit → Roll into a new maturity → Rotate into another yield opportunity That makes rollover activity important. A deployment where capital repeatedly rolls can monetize the same capital several times. A deployment dominated by hold-to-maturity users can maintain strong TVL with much less trading. Both are real usage. They just create different economics. ——— — Lower Fee Intensity Isn’t Automatically Worse Lower turnover does not mean Monad’s deployment is inferior. A user buying PT for fixed yield can hold until maturity without trading again, and Pendle has still done its job. The narrower conclusion is this: Monad currently produces much less fee-generating activity per dollar of deposited capital than Ethereum. That is more informative than simply comparing TVL. ——— — What The Numbers Actually Tell Us The entire comparison comes down to three numbers: → TVL advantage: Ethereum 3.2x → Monthly fee advantage: Ethereum 12.8x → Fee intensity advantage: Ethereum 4x Pendle’s Ethereum and Monad deployments differ in more than size. TVL tells us where the capital sits. Fee intensity tells us how economically active that capital is.

Bitcoin holders realized $336.6M in profit on September 10. But it wasn’t evenly distributed across the holder base. One cohort dominated: • 100–1,000 $BTC: $128.2M • 1,000–10,000 $BTC: $56.6M • Total realized profit: $336.6M The 100–1,000 $BTC cohort alone generated roughly 38% of the entire day’s realized profit. Combine both whale cohorts and that rises to around 55%. There’s another useful comparison. The aggregate realized price of 100–1,000 $BTC wallets is roughly $67.2K, versus $52.2K network-wide. So this cohort has a significantly higher cost basis than the average $BTC holder, while still sitting on substantial embedded gains at current prices. That helps explain why relatively modest coin movement can generate such a large share of realized profit. One important caveat: realized profit does not automatically mean selling. Coins can move to exchanges, OTC desks, collateral accounts, custodians or simply another wallet controlled by the same entity. So the chart doesn’t prove whales are dumping. It shows something more specific: Bitcoin’s current profit realization is heavily concentrated among larger holders, with the 100–1,000 $BTC cohort responsible for nearly $2 of every $5 realized on September 10. That’s the chart.

Prediction markets have a pretty funny problem once you think about how they’re supposed to work. The whole product gets better when someone knows more than everyone else. If a market says an outcome has a 30% chance and you have better information suggesting 70%, you buy. Price moves, other traders react, and eventually the market incorporates that information. Great product. Until the person buying at 30c doesn’t have a better model. They literally know the answer. ESMA flagged prediction markets today over growing concerns around insider dealing and manipulation as these platforms get bigger. And this gets weird fast because prediction markets can create insiders for almost anything. — Everyone Can Be an Insider Somewhere Stocks have a relatively obvious insider class: executives, employees, advisers and people with material nonpublic corporate information. Event markets are messier. • Will candidate X drop out? Campaign staff might know first. • Will player X start Sunday? Coaches, players and team employees might know. • Will company X announce product Y? Employees might already have the launch calendar. • Will country X conduct military operation Y? Government and military personnel can possess information the public literally cannot obtain. These people aren’t better forecasters. They’re standing closer to the answer. That’s a completely different edge. — And We’ve Already Seen What This Looks Like This isn’t hypothetical anymore. @Polymarket explicitly bans trading using stolen confidential information, illegal tips or information where the trader can influence the outcome. It even references the 2026 case of a U.S. service member charged after allegedly using classified information to profit from prediction-market contracts. @Kalshi has gone further with proactive screening. Political candidates, campaign staff, election officials, athletes, coaches, referees and relevant league personnel can be restricted where their position creates an informational conflict. Kalshi also says surveillance caught a video editor trading markets related to content he worked on. So the industry already knows this problem exists. — But Removing Insiders Creates Another Problem Prediction markets are useful partly because informed money corrects bad prices. You don’t want to remove everyone with superior information. A political analyst with a better election model should trade it. A weather researcher with better hurricane data should trade. Someone who analyzed satellite imagery and figured something out before CT should trade. That’s price discovery. But someone sitting inside the campaign meeting where the candidate already decided to withdraw? Different game. Same profitable trade. Completely different source of edge. The line isn’t really: Informed vs uninformed It’s: Discovered information vs privileged information. — Wallet Transparency Doesn’t Solve This Crypto makes surveillance easier. Onchain markets expose wallets, positions and trade timing. Everyone can see when some wallet buys an unlikely outcome hours before the news. Cool. Now tell me who owns the wallet. Kalshi requires KYC and has added employment verification for higher-risk markets. Polymarket says it uses real-time surveillance and external partners including Chainalysis and Palantir. The market can be transparent while the person behind the trade remains difficult to classify. — This Is Probably Prediction Markets’ Hardest Design Problem Prediction markets need informed traders to improve prices. But they can’t reward privileged access. Every market creates different insiders: • Politics → campaigns • Sports → teams • Companies → employees • National security → officials The event itself is the asset. Markets need information asymmetry, not one side already knowing the answer.

Crypto products increasingly reach users through applications they already use. Exchanges connect balances to lending. Wallets bring swaps into the portfolio screen. Cards turn holdings into spending power. Payment platforms introduce stablecoins through checkout and payroll. The landscape now spans six entry points, each influencing which products users discover and where their money goes next. 1. — Exchange Customer Funnels • @Coinbase: trading and integrated onchain lending. • @binance: retail trading and ecosystem access. • @krakenfx: trading, funding, and rewards products. • @OKX: exchange and Web3 access. • @Bybit_Official: spot and derivatives trading. • @bitget: trading and copy-trading discovery. • @cryptocom: trading and consumer payments. • @Gate_io: broad token discovery. • @kucoincom: altcoin discovery and trading. • @MEXC: emerging-token trading. 2. — Wallet-Native Discovery • @Phantom: in-wallet swaps and staking. • @MetaMask: swaps and application access. • @TrustWallet: multichain asset management. • @Rabby_io: EVM application access. • @Backpack: wallet-based ecosystem access. • @Zerion: portfolio tracking and onchain trading. • OKX @wallet: multichain Web3 discovery. • @Binance Wallet: exchange-connected onchain access. • @exodus: asset management and swaps. 3. — Fintech and Banking Integrations • @Revolut: crypto inside a financial account. • @PayPal: crypto and PYUSD access. • @Venmo: crypto within social payments. • @RobinhoodApp: crypto alongside traditional investments. • @CashApp: Bitcoin inside a money app. • @n26: crypto inside a banking app. • @Strike: Bitcoin and money transfers. 4. — Crypto Card Payments • @RedotPay: crypto-funded everyday payments. • @Ether_fi Cash: direct spending or asset-backed borrowing. • @KASTxyz: stablecoin-funded account and card spending. • @Bybit_Official Card: spending from exchange funding balances. • @ethena Pay: card access alongside stablecoin services. • @MetaMask Card: spending from a self-custodial wallet. • @Nexo Card: debit and crypto-backed credit modes. 5. — Merchant and Payroll Distribution • @Stripe: stablecoin checkout for merchants. • @BitPay: crypto payment acceptance. • @CoinGatecom: merchant payments and payouts. • COINPAYMENTSNET: crypto checkout integrations. • @TripleAHQ: digital currency payments and payouts. • @NOWPayments_io: merchant payment integrations. • BtcpayServer: self-hosted Bitcoin payments. • B2BINPAY: business payment infrastructure. • @Bitwage: crypto and stablecoin payroll. • @RequestFinance: business payments and payroll. 6. — Embedded Applications and Developer APIs • @privy_io: wallets embedded inside applications. • @dynamic_xyz: wallet onboarding and authentication. • @turnkeyhq: programmable wallets and signing. • @crossmint: embedded wallets and financial APIs. • @thirdweb: in-app wallets and developer tools. • @0xsequence: wallets for games and applications. • @Coinbase Developer Platform: developer wallet infrastructure. • @Stablecoin: stablecoin issuance and payment APIs. PS: Some businesses appear across multiple categories because they control several entry points.

RWA perpetual volume fell 13.5% to $122B in August, ending six consecutive months of growth. That alone isn’t especially interesting. The timing is. August was also crypto’s broadest rally of 2026, with roughly 83% of the top 100 assets finishing higher. So the first down month for RWA perps arrived exactly when crypto-native assets became much more attractive to trade again. That suggests part of the category’s growth wasn’t only structural adoption. It was substitution. … — Crypto Traders Needed More Beta RWA perp volume started the year around $23.1B in January. By July, it had reached $141B. That’s more than 6x growth in six months. Over the same period, perp venues expanded aggressively into: • equities • commodities • indices • precious metals This gave crypto-native traders access to traditional-market beta without changing venues, collateral systems or execution workflows. The product wasn’t simply “tokenized TradFi.” It was a way to keep traders active when crypto itself lacked enough movement. That distinction carries weight. A trader using Hyperliquid or Binance doesn’t necessarily care whether the opportunity is $BTC, NVDA or gold. They care about liquidity, volatility and execution. RWA perps expanded the opportunity set without forcing that trader to leave crypto rails. … — August Exposed the Demand Mix Then crypto started moving. $BTC gained sharply. $ETH outperformed. Breadth expanded across the top 100. And RWA perp volume fell from $141B to $122B. That’s the first clean evidence that some RWA perp demand was substitutional. When crypto-native volatility was weak, traders moved toward equities, commodities and macro products. When crypto offered broader directional opportunities again, part of that flow returned. So RWA perps aren’t only competing with CME, brokers or traditional derivatives venues. They’re also competing with crypto assets for trader attention. … — But $122B Is Still a Large Market The decline doesn’t invalidate the category. August volume was still more than 5x January levels. That tells us the market kept a large amount of activity even after crypto-native beta returned. So the right framework isn’t: RWA perps are growing or RWA perps are fading It’s: structural demand + substitution demand Structural demand comes from traders who specifically want stocks, commodities and macro assets through crypto rails. Substitution demand comes from crypto traders rotating into those assets when native markets are less attractive. August separated those two demand sources for the first time this year. … — The Business Model Is Bigger Than Tokenization This is the part I find more important. Perp venues are gradually becoming multi-asset exchanges. Once the same account can trade $BTC, $ETH, gold, equities and indices with the same collateral and execution stack, the venue is no longer competing only inside crypto. It’s competing for the trader’s entire derivatives wallet. RWA perps therefore don’t need crypto-native assets to stay boring forever. They only need traders to prefer one unified venue over fragmenting their capital across multiple platforms. That’s already a much larger market than tokenized assets alone. … — My Take August showed that part of RWA perp growth was driven by crypto’s lack of movement. But it also showed something more important. Even after crypto produced its broadest rally of the year, RWA perp volume still held at $122B. So the category has already moved beyond a temporary boredom trade. The durable advantage is distribution. Crypto exchanges are turning into multi-asset derivatives venues, and RWA perps are becoming another product inside the same liquidity network. The trade isn’t “RWAs versus crypto.” It’s who owns the account traders use for both.