ChainFeeds on July 14 published a research digest covering Ethereum, AI equities, crypto positioning, private AI infrastructure and yield generation for tokenized real-world assets.
Ethereum: application value remains large, but L1 capture is under pressure
In a long-form thread on Ethereum’s ecosystem, Nick Researcher said the network’s investment case is changing. The second quarter of 2026 showed some improvement in Ethereum revenue from the prior quarter, but Layer 1 fee capture remained well below last year, on-chain yield was close to historical lows, and DeFi activity softened.
The piece said users have continued moving to Layer 2 networks, while some have also left the ecosystem because the L2 experience did not meet expectations. At the same time, transaction fees have fallen, blob supply has grown faster than demand, and L2 activity has not translated into enough value for Ethereum mainnet.
The report highlighted one core set of numbers: Ethereum L1 generated $88.4 million in real economic value in the second quarter of 2026, up 7% quarter over quarter but down 68% year over year. Over the same period, applications on Ethereum L1 produced about $1.79 billion in fee revenue. The argument is straightforward. Applications inside the Ethereum economy still generate meaningful value, but the base layer captures only a small share of it.
The report also pointed to Ethereum’s role in hosting major protocols such as Tether, Circle, Lido, Aave and Uniswap. Stablecoins remain one of the chain’s strongest advantages. Ethereum L1 stablecoin supply reached $172.9 billion in the second quarter of 2026, down about 4% from the prior quarter but still at large scale.
Scale alone is not enough, the piece argued. What matters as much is the speed of capital turnover. If stablecoins sit idle on-chain without trading, settlement or collateral use, they do not create enough economic value.
RWA was presented as a possible next growth driver for ETH. On-chain RWA on Ethereum L1 has already exceeded $15.7 billion, up about 90% year over year, including tokenized Treasuries, commodities and equities. Still, the report noted that Solana’s average daily RWA trading volume in the second quarter of 2026 was higher than Ethereum’s despite lower RWA TVL, suggesting Ethereum’s edge lies more in institutional depth while Solana’s lies in velocity.
The piece said ETH would need three conditions to support a re-rating: more institutional assets entering the Ethereum ecosystem, more financial settlement activity taking place on Ethereum, and higher real transaction frequency from on-chain assets.
On tokenomics, the report said ETH’s annualized net dilution rate was about 0.85% in the second quarter, close to BTC. But it also flagged risk. Total on-chain yield had fallen to a record low of 2.68%, with 94% of that yield coming from ETH issuance rather than real user fees. In the report’s framing, Ethereum’s valuation upside depends on whether it can become the settlement layer for an institutional financial system.
BlackRock’s AI view: strong rally, but long-term valuation signals are flashing
Another section of the digest cited a BlackRock note based on Morningstar data. It compared the current AI boom with the internet bubble. From 1993 to 1999, U.S. tech stocks rose 1,097% while the broader U.S. market gained 292%. In the AI cycle running from 2019 to June 30, 2026, tech stocks rose 569% and the broader U.S. market gained 237%.
The note said the current run has been powerful, but the path has looked different. Tech stocks fell 28.2% in 2022, rebounded 57.8% in 2023, then gained 36.6% in 2024, 24.0% in 2025 and 19.8% in the first half of 2026. Unlike the internet bubble, which accelerated sharply in its final stage, the AI trade widened in 2023 and then slowed in pace.
Valuation is now the central point of debate. The S&P 500’s Shiller CAPE has climbed to about 40x, back near levels seen during the dot-com era. BlackRock said the 12-month forward P/E, at roughly 21x, gives a different read because earnings expectations have also moved higher. S&P 500 second-quarter earnings are expected to grow 23% year over year, marking a seventh straight quarter of double-digit growth.
The report added that the so-called Mag 7 now trade at roughly 26x earnings, with profit growth expected above 30% and aggregate earnings growth around 27.6%. As of May 31, 2026, technology stocks accounted for 37.5% of the total U.S. equity market, above the level seen in the late 1990s bubble period.
It also said market leadership is spreading beyond the traditional Mag 7 to a broader group of AI beneficiaries. A new acronym, MANGOS, was used to describe Meta, Anthropic, Nvidia, Google, OpenAI and SpaceX. Morningstar’s Global Next Generation Artificial Intelligence Index gained about 45% in April and May 2026 before pulling back in June.
BlackRock’s conclusion was narrow but clear: calling AI a bubble is itself a major claim because it assumes AI will not deliver long-term productivity gains. The more important question now is not how much further AI can rise, but how long earnings growth can keep up.
Multicoin’s Tushar Jain names Solana, Hyperliquid and Zcash
The digest also featured comments from Multicoin Capital managing partner Tushar Jain. He said he still sees Solana as the right technical architecture for internet capital markets, arguing that a permissionless open-source chain is needed to bring activity onto one platform.
At the same time, he said derivatives volume is shifting toward Hyperliquid. He said he has large positions in both and remains constructive on both. In his view, Solana leads in spot trading and could host spot trading for tokenized securities, while Hyperliquid is ahead in derivatives.
Looking to 2026, he also singled out Zcash. Jain said the position is smaller because of liquidity and market cap limits, but Multicoin has accumulated a significant share of total supply. He said he likes Zcash’s momentum, use cases and community, and that it reminds him of early Bitcoin.
Jain also laid out the assumptions behind Multicoin’s Hyperliquid thesis:
- Crypto derivatives grow at a 35% compound annual rate, versus 45% over the last five years.
- DEXs reach 32% of the derivatives market after rising from nearly zero in 2022 to 16% now.
- Hyperliquid maintains a 30% share of decentralized derivatives.
- USDC collateral grows linearly with trading volume.
He added that Hyperliquid currently accounts for 59% of real open interest across the network, which he described as a harder figure to fake. As subsidies on other platforms fade, he said Hyperliquid’s effective share should rise.
IOSG on private AI: the next fight is around trust and architecture
IOSG Ventures used its weekly note to examine why large financial firms are saying no to ChatGPT and Claude in sensitive workflows. The piece framed private AI around one basic issue: where plain-text data exists from the moment a prompt leaves a user’s device, travels across the network, reaches the model server and returns as an answer, who can read it, and how users can verify protection.
The note broke the market into several privacy models. Protocol-level privacy relies on provider promises, including enterprise zero-retention setups. Anonymous relays can hide identity but not the text of the prompt. TLS protects data in transit between machines, but the recipient can still read the content. Oblivious HTTP, or OHTTP, splits knowledge of identity from knowledge of content. It has become an IETF standard and is starting to appear in production environments.
Still, the note said those approaches are close to the limit for closed flagship models because model weights are the core asset of AI companies. Training a top model can cost tens of billions of dollars, and labs protect valuation by keeping a performance gap and not opening their weights or complete service code.
That is where structural privacy comes in. Trusted execution environment, or TEE, confidential computing was presented as the path closest to commercial adoption. TEE places model inference inside a hardware enclave, and the chip can produce attestation to show users that a specified model and code are running. But the note also said prompts are only protected once they enter the enclave, leaving room for exposure at proxy and relay stages beforehand.
End-to-end encryption closes more of that gap by letting the user device encrypt prompts directly with enclave keys, though it increases engineering complexity because every feature that depends on plain-text data has to be redesigned. Fully homomorphic encryption, or FHE, and multi-party computation, or MPC, go further by trying to compute directly on ciphertext. The report said costs remain a major obstacle. Transformer models require complex operations, and encrypted inference can cost tens of thousands of times more than plain-text inference.
The report called local inference the most complete privacy model because the model runs on the user’s own device and does not depend on servers or relays. The tradeoff is model capability and hardware cost.
IOSG then shifted to agent workflows. Current private inference systems mainly protect the path from prompt to model, but AI agents also need external tools such as calendars, databases, search engines and internal enterprise systems. Those create fresh plain-text exposure points. The note said future value capture may sit in unsolved areas such as running training loops inside enclaves, protecting tool calls end to end, and building search systems that do not expose query content.
Tokenized gold and the next step for RWA
The final piece argued that the biggest constraint in today’s RWA market is not only asset size, but asset efficiency. Most on-chain RWAs are still concentrated in lower-risk assets such as U.S. Treasuries, with expansion into equities and other categories still developing.
Gold is the largest tokenized commodity on-chain, with more than $4.9 billion in on-chain gold already issued. But most products still stop at spot exposure and do not offer ways to make those holdings productive.
The report said the next phase for RWA is to move beyond tokenization alone and turn on-chain assets into yield-bearing positions. In traditional markets, gold covered-call ETFs already offer one model. The piece cited GLDI, which charges about a 0.65% management fee and deducts that fee directly from investor returns.
It then described Enhanced’s PAXG Volatility Income Vault, the firm’s first Thesis Vault product. The vault uses PAXG and a covered-call strategy to generate income from gold volatility. It runs on an RFQ model, with deposited assets sent through batch auctions where market makers quote prices before options are executed on-chain and users receive premium income upfront.
According to the report, the vault uses European-style options that can only be exercised at expiry, with funds locked for each cycle. Users can deposit either PAXG or USDC, and the system converts USDC into PAXG automatically. Options run on a two-week cycle, about 26 times per year, with strike prices expected to sit 3% to 7% above the current gold price.
Two payout modes are available. In compounding mode, USDC premiums are automatically converted back into PAXG and rolled into the next cycle. In income mode, proceeds are kept separately and users can withdraw USDC at any time. The piece argued that this model aims to solve a core RWA problem: not just putting assets on-chain, but making them generate real economic value.