ChainFeeds Research roundup tracks Wintermute’s machine economy thesis, Base’s strategic pivot, and Trade[XYZ] on HIP-3

ChainFeeds on July 17 published a new edition of its daily research briefing, bringing together five featured reads on privacy tokens, the machine economy, U.S. equity valuations, Base’s strategic reset, and Trade[XYZ]’s expansion inside Hyperliquid’s HIP-3 framework.

Headline items in the briefing

The newsletter’s headline list included the U.S. Senate unanimously passing a resolution opposing a pardon for FTX founder Sam Bankman-Fried, 1inch co-founder Anton Bukov saying he is leaving the team and launching a new project called Second Tier, Moonshot AI’s Kimi K3 beginning to roll out to users, Fireworks AI closing a $1.5 billion financing round led by Index Ventures and others, and Visa launching a stablecoin platform that it said could open stablecoin services to more than 200 million merchants.

Why privacy tokens rose while most crypto sectors fell

The first long-form item, attributed to Chamath Palihapitiya, argued that privacy tokens have re-emerged because they address one of money’s core properties: fungibility. The piece said that over the past year, nearly every category of crypto tokens declined, yet privacy tokens rose 127.3%, making them the best-performing segment in the market.

The argument starts with Bitcoin’s public ledger. While that architecture makes decentralized verification possible, the article said it weakens privacy because anyone can inspect the flow of funds tied to a given address and trace the historical path of a BTC unit. If a coin was previously linked to illicit activity, a platform may refuse to accept it or may discount it. In that setup, transaction history can alter value, which means Bitcoin does not fully satisfy a strict definition of fungibility.

The piece also said wallet addresses behave more like pseudonymous usernames than bank accounts. Once an address is connected to a real-world identity through a transaction, an exchange registration, or a data leak, a user’s previous activity can become permanently traceable. That is where the article places zero-knowledge proofs. It described the technology as a way to prove that something is true without exposing the underlying information. In private digital-currency transactions, a network can validate that a transfer is legitimate, that the sender holds the assets, that funds have not been double-spent, and that protocol rules are being followed, without revealing who sent the funds, who received them, or how much was moved.

According to the article, that balance between verifiability and privacy has become the technical foundation for a new generation of privacy protocols, including Zcash. It added that the sector now sits between technical progress and regulatory pressure. In 2025, 73 exchanges delisted privacy tokens, the piece said, yet the category still ranked among the strongest-performing sectors in crypto over the same period.

The briefing noted that projects are taking different approaches. Monero uses mandatory privacy, with all transactions hiding transaction details by default. Zcash uses selective privacy, allowing users to enable private functionality when they choose. It also pointed to emerging ideas such as private digital dollars that aim to balance compliance requirements with user privacy. Looking ahead, the article outlined three possible paths: privacy tools become accepted as part of mainstream financial infrastructure, regulators tighten restrictions on anonymous transactions, or compliant privacy models gain traction by offering protection without stepping outside regulatory boundaries.

Wintermute says the next opening may come from the machine economy

The second featured essay came from Wintermute Ventures. Its central claim was that crypto’s next big opportunity may not be found by repeating the last cycle’s playbook, but by asking what the world now needs crypto to solve. The answer offered in the piece was the machine economy.

Wintermute said crypto has already spent more than a decade proving that core rails can work. Layer 1 blockchains are live. Layer 2 scaling systems continue to develop. DeFi has matured. Stablecoins are now part of market infrastructure. Exchanges, lending, perpetuals, and prediction markets have all been explored. In the firm’s telling, the earlier phase was about whether settlement could be fast, whether stablecoins could circulate at scale, and whether open networks could handle real demand. Those questions, it said, have largely been tested already.

The next set of problems comes from outside crypto. The article said AI models are shifting from passive response tools toward agents that can carry out tasks on their own. Robots are learning from human video rather than only from hand-coded training. Payment and identity standards aimed at agents are also beginning to form. None of those systems are native to crypto, but they pressure financial and trust architectures that were built around humans.

That is the point where crypto starts to look useful again in a different way. In Wintermute’s framing, the machine economy changes the role of machines from tools into economic actors. Future AI agents and robots may understand environments, retain context, make decisions, and act across digital and physical settings. Once they do, many assumptions embedded in current systems begin to fail. Payments, identity verification, authorization, dispute handling, and settlement all assume that the other side of a transaction is a person or a company that can be identified and held accountable.

The piece gave several examples. An AI agent might book flights, negotiate prices, make payments, and process refunds for a user. A warehouse robot might accept work autonomously, pay for computing resources, and distribute revenue to an operator. A research system might design experiments, buy materials, and carry out a research flow without direct human supervision.

Wintermute argued that blockchains are structurally better suited to these cases than traditional financial rails because traditional systems depend on identity and intermediaries, while blockchains rely on publicly verifiable code, onchain records, and rules executed by networks. If more economic activity is machine-driven, the article said, then open, programmable, permissionless, fast-settlement crypto infrastructure may fit that environment better than systems built for human institutions.

The essay also pushed past the usual crypto-versus-AI framing. It said the next generation of founders is unlikely to choose one field over the other. Instead, they may combine them in areas such as Crypto + AI, Crypto + Robotics, and Crypto + Autonomous Science. Wintermute said current market attention is concentrated in foundation models, robotics hardware, stablecoins, and trading venues, where competition is already intense. The less crowded opportunity may lie in the connecting layer beneath them: economic infrastructure for agents, machine-native payment systems, and identity and authorization systems for autonomous actors.

For AI agents, the key issue is not simply whether a payment can be made. The harder questions are who has authority, who bears the cost of mistakes, and how merchants can connect to an agent-driven economy without rebuilding existing systems. For robots, the missing piece may not be physical execution, but wallet systems that let them manage funds, pay costs, and receive income on their own. The article added that automated science could become another major area, with AI linking hypotheses, experiments, data, and discovery in ways that could speed work in fields such as materials and pharmaceuticals. Its conclusion was direct: crypto’s largest opportunity in the coming years may be to serve as a new trust layer for AI and the machine economy rather than a replacement for the old financial system.

Are U.S. stocks overheating?

The third long read, from arndxt, focused on U.S. equity valuations. It said that nearly all of the standard market-wide valuation gauges being tracked now place the U.S. market at or near the top end of historical observations, including the Shiller CAPE, Tobin’s Q, and long-term trend-deviation measures.

Still, the article said those highs should not be treated as hard ceilings. A market reaching a historically rich level does not mean it cannot go higher. It pointed to Japan’s bubble-era valuation peak at roughly 100 times earnings and to similarly extreme valuation episodes in China, arguing that a Shiller CAPE in the 40s for the U.S. is not a magic number that automatically stops further expansion.

What matters more, the author wrote, is where those rich valuations are concentrated. The current setup is centered in a small group of large-cap growth names, which the piece compared with the structure seen near the top of the late-1990s internet bubble. Back then, a limited cluster of mega-cap growth stocks traded at very high valuations while a broader set of smaller companies remained reasonably valued or even cheap. The article tracked the current split using two long-run relationships: equal-weighted versus market-cap-weighted S&P 500 performance, and the relative performance of the biggest firms in the S&P 100 against the broader S&P 500. During episodes such as the late 1990s, 2020, and 2024 to 2025, the market-cap-weighted index tended to lead sharply while the equal-weighted market lagged.

It added that over the past few months, those indicators have begun to tilt back toward the broader market, though the move has been volatile. At the same time, the valuation gap between the most expensive assets and the cheapest ones has reached an extreme. Using a valuation-dispersion measure adjusted for quality, the article said the spread between value and growth sits around the 90th to 95th historical percentile. Even after accounting for quality, the spread remains around the 85th to 90th percentile.

The piece said that kind of divergence is close to what was seen at the bottom of the 2009 financial crisis and during the 2020 pandemic selloff, but with one major difference: this time the broader market index has not gone through a comparable drawdown. The author tied that gap to earnings divergence. From 2022 through the end of 2025, small- and mid-cap companies went through an earnings slump that the market underappreciated, while profits at large technology companies rose almost in a straight line.

That leaves a difficult question around the Magnificent Seven. After multiple expansion has already pushed valuations so high, the article said, it is hard to judge how much room remains. For investors who believe in long-run mean reversion, the more reasonable opportunity may sit in neglected areas such as small-cap value, micro-cap value, and mid-cap value.

The second half of the essay turned to AI. Large language models are changing productivity structures, the author said, but it is still unclear whether the biggest economic gains created by AI will necessarily accrue to today’s model leaders. The personal experience of using large language models, the article said, can feel like having an extremely fast and usually accurate junior analyst who occasionally makes mistakes. Even so, that does not guarantee durable excess returns for the companies currently in front.

The article offered three reasons. First, functional depreciation in compute infrastructure such as GPUs may be much faster than in past infrastructure cycles. It said GPUs have an effective life of about five to seven years, while railroads and fiber networks preserved value over much longer periods. Second, it remains uncertain whether AI ends up dominated by a single model or by several competing models that converge in capability. If the latter happens, competitive pressure could compress margins. Third, another outcome is also possible: the industry spends huge sums building powerful AI systems, but consumers absorb most of the value rather than the model creators. The article compared that outcome with websites around 2000. Having a website was once an advantage; today it is basic operating infrastructure. AI, the author suggested, may follow a similar path.

Jesse Pollak admits Base got social wrong

The fourth article, from ChainCatcher, centered on a strategic shift at Base. It said that on July 15, Base founder Jesse Pollak published a long post announcing that he would hand leadership of Base App back to Coinbase and focus fully on the Base blockchain itself, with the goal of turning Base into a “global financial blockchain.” Pollak will continue to lead the Base chain, but he will no longer run Base App. That role, according to the article, will be taken over by Cobie, Jordan Fish.

The piece argued that the key development is not the personnel change itself, but Pollak’s rare public admission that Base misread the importance of social products over the past two years. Base had tried to position itself as a consumer-facing gateway to crypto, leaning on Farcaster, Zora, creator coins, miniapps, and Base App to push an onchain social and creator-economy model meant to pull ordinary users into blockchain products.

What happened instead, the article said, is that Base was right about builders but wrong about social. Pollak’s line was quoted directly: “We bet right on builders, but bet wrong on social.” In the article’s reading, that sentence sums up the outcome of Base’s social experiment. Onchain social products did not become the main driver of the next wave of adoption. Prediction markets, perpetuals, stablecoins, and tokenized assets moved further instead. Users were not rejecting onchain activity altogether; they were simply unwilling to go onchain for social use alone. They were more willing to do it for trading, payments, yield, and speculation.

The result is a narrative shift from consumer social to financial infrastructure. The article said Base’s earlier push was not illogical. Jesse Pollak helped shape Base’s culture, and the chain had spent the last few years trying to define itself as a consumer-oriented ecosystem distinct from traditional DeFi. The rapid rise of friend.tech inside the Base ecosystem once gave the market reason to think onchain social products and creator finance might become a new user-acquisition engine. It showed that when social relationships become financialized, onchain products can attract intense attention very quickly.

That momentum helped reinforce confidence in the broader stack around Farcaster, Zora, creator coins, miniapps, and Base App. The idea was that Coinbase could supply a compliant gateway, Base could supply a low-cost onchain environment, Farcaster could supply a social graph, and Zora could supply tools for turning content into assets. Together they might support a new onchain consumer ecosystem.

The article said that logic did not translate into durable growth. The problem was that onchain social products can easily become onchain speculation. friend.tech drew attention more because social ties were turned into tradable financial instruments than because users got a meaningfully better social experience. Creator coins ran into a similar pattern. Once influence and community relationships were tokenized, trading activity often outweighed content consumption. When speculative interest faded, those social relationships did not remain by themselves. The piece said Farcaster faced a cold-start problem, Zora faced tension between content and asset issuance, and creator coins could turn into short-cycle attention trades.

Base had hoped those products would attract mainstream users, but the article said the audience that remained was more often crypto-native users, airdrop hunters, short-term traders, and creator-coin participants. Base is now moving toward onchain finance instead, with a focus on trading, payments, stablecoins, AI agents, and financial settlement rails. The article said that direction better matches what has actually grown over the past year, including stablecoin payments, tokenized stocks, prediction markets, perpetuals, RWA products, onchain lending, and AI-agent payments.

Even with that reset, the piece said Base still has clear strengths: Coinbase’s compliance brand, exchange distribution, a developer ecosystem, stablecoin use cases, and institutional relationships. It added that Base is also exploring AI, naming Venice and Virtuals as representative ecosystem projects. If AI agents become a new class of economic actor, Base may serve not only human users but also wallets, payments, settlement, and trading for agents. The article described the biggest opening in this next phase as linking stablecoins, trading, tokenized assets, and AI agents into one infrastructure layer. It also warned that pressure is building. Robinhood Chain is expanding through tokenized stocks and financial distribution, while Solana and Hyperliquid continue to compete on trading experience and market structure. Base, in that telling, no longer has much room to rely on narrative alone.

How Trade[XYZ] built 92 markets and 98% of HIP-3 volume

The final feature, from TechFlow, examined Trade[XYZ] inside Hyperliquid’s HIP-3 model. It opened with a provocative question: will Trade[XYZ] kill Hyperliquid? Its answer was no.

The article first laid out Hyperliquid’s core idea. Long-run competitiveness in onchain finance, it said, comes from the infrastructure layer itself: a high-performance Layer 1, a liquidation system, and a matching engine. Hyperliquid has put most of its effort into those components and assumes that strong operators will choose to build on top of them. In that structure, HIP-3 is not described as a simple horizontal-scaling tool. It is closer to an open application layer for market access. Hyperliquid does not want to directly operate every market. It wants a neutral platform where strong operators in different verticals compete, and liquidity determines the winners.

The article said traditional finance already shows how powerful liquidity network effects can be. There is one CME, one NYSE, one Hong Kong Stock Exchange. Markets without a single deep center of liquidity struggle to become mainstream. Hyperliquid’s target, in this account, is to become the base layer for trading all financial assets, with HIP-3 as the mechanism that lets operators launch products on top. It does not preselect a winner. It opens the field and lets builders compete to create the deepest markets. Successful markets can feed trading fees, user growth, and value back to HYPE, while deployers also capture their own revenue. Under that model, market concentration is not a sign that the framework failed; it reflects how financial markets tend to work.

The article then moved to stock perpetuals and argued that listing an asset is the easy part. The harder task is creating a market with enough depth to handle meaningful capital. Trade[XYZ]’s data, as cited in the piece, points to three core requirements for a successful RWA perpetual market: list assets quickly enough to catch demand, attract professional market makers who can supply liquidity, and keep that depth in place long enough for the market to gain real economic value.

On speed, the article said Trade[XYZ] performs well. The median time from onchain asset registration to the first trade is 3.3 days. Sixty-five percent of markets go live within a week, and 47% are deployed within three days. Yet the article argued that speed is not the real moat. Liquidity is.

It cited several order-book figures from flagship markets. XYZ100 shows about $2.6 million in depth within 10 basis points of the mid-price. The S&P 500 market shows about $964,000. Gold shows about $759,000. By contrast, the median depth of ordinary markets is only around $20,000. That makes market-maker acquisition and retention the hardest capability to replicate. In 73 markets, the article said, the number of distinct market-maker wallets active each day is clearly negatively correlated with spreads: more market makers, lower spreads. Higher volume also lines up with stronger liquidity. In that reading, the durable edge comes from continuously attracting capital and operating markets well, not from simply opening a new trading venue.

The article also focused on the risk problem in stock perpetuals. Market makers earn the bid-ask spread, but every fill leaves them with directional exposure. If they cannot hedge that exposure efficiently, inventory risk builds. Crypto perpetuals can usually be hedged around the clock on other venues. Stock perpetuals are different because the underlying hedge is often a stock, an ETF, or a futures contract, and those markets have trading-hour limits. During regular sessions, a market maker can hedge inventory with spot equities and quote tighter, deeper books. After the equity market closes, that hedge is no longer immediately available. Under ordinary logic, that means wider spreads, less depth, or even a halt in quoting.

Trade[XYZ], according to the article, addresses that with three mechanisms: Discovery Bounds, liquidation protection, and funding-rate adjustment. Discovery Bounds limit how far the mark price can move away from a reference price, which keeps one-step price moves inside a predictable range. Liquidation protection helps prevent unreasonable liquidations caused by temporary price dislocations. The funding-rate system helps keep perpetual prices anchored to spot value while easing pressure on market makers.

The article said that in the top 10 stock markets on the platform, after-hours order-book depth is maintained at about 116% of regular-session levels. It presented that as evidence that the platform’s risk controls can help market makers keep quoting outside normal equity-market hours. Even so, the piece was clear on where it thinks the real advantage comes from: not a simple technical trick, but long-term market operation, risk management, and accumulated relationships with market makers. The hard part is not launching a market. It is keeping that market running across different assets, different times of day, and different risk conditions.

The July 17 ChainFeeds briefing said the daily digest was produced jointly by the ChainFeeds team and AI.