AI Fatigue, Robinhood & Every Market Becoming Crypto | Weekly Roundup

Summary

• CLARITY’s odds just hit an all-time low near 27-30% on Polymarket, and Rob — the resident “clarity guy” — wouldn’t buy the dip: “I wouldn’t buy the 30% on poly,” it’s “appropriately priced.” Public optimism from figures like Brian Armstrong masks private bearishness over two blockers: ethics language on Trump-family crypto ownership (who litigates — FBI or state AGs?) and BRCA developer protections, where Sen. Cortez Masto floated “agreed” language Republicans and industry “would not have agreed with.” The consolation: SEC and CFTC rulemaking proceeds regardless, so “we’ll be in a better spot regardless” in 6-9 months.
• Every asset class now trades like crypto, and the mechanism is leverage plus quant flow. Santiago cites Goldman reporting prime leverage in the final half of 2026 at its highest since before the GFC, with ~20% bunched in a few AI memory names; Leopold supposedly sold his entire $14B notional in a single block trade at 4x leverage into memory stocks down 50% in a month, and Rob thinks Korea has had more circuit breakers this year than in the market’s entire history. Santiago’s takeaway: of “ladies, liquor, and leverage — definitely avoid leverage.”
• The best-performing crypto liquid funds over the last five years are all delta neutral — “market neutral funds and lawyers are the best performing people in crypto.” Rob argues the long-term fundamental investor “is struggling more than ever” because LPs can’t stomach the vol (“just don’t send me quarterly statements, please”), while Ken Griffin’s newfound interest in long-term fundamental investing reads as a contrarian tell.
• Rob is hearing AI fatigue from big institutional LPs for the first time, with one comparing it to crypto’s 2021 cycle. The rumored $10B OpenRouter deal on a $140-150M revenue business “is insane to me”; his stealth router portfolio company cut one client’s token spend by 80%, and he thinks the premium for frontier models “is probably going away in the vast majority of industries within the next 12 to 18 months.” Yet Santiago calls Anthropic’s ARR the “fastest growing in the history of mankind,” and says, “I don’t know anybody who thinks that goes public for less than 1.2 trillion.”
• Santiago calls Robinhood “the most important company” to track for crypto flows — if AI fatigue rotates money back, “you’re likely going to see it first” on Robinhood Chain, not in earnings. Q results: $1.3B in earnings, up 32% YoY, with prediction markets ($156M revenue) out-earning crypto ($100M); the Arbitrum-built chain did $12B in DEX volume and 100M transactions in four weeks and, per Frank’s tweet, leads all chains in tokenized stockholders — though Rob notes active accounts have trended down post-launch, “same thing happened to Base.”
• Away from Robinhood, on-chain is bleak: all other Ethereum L2 TVL is back to ~$5B (2023 levels), and H1 saw a record 212 exploits with $1B+ lost, half-plus from Lazarus. Rob’s three-factor explanation is low yields (basis trade from 10%+ to 3-4%), hacks, and AI stealing the momentum trade. Santiago’s preferred metric is active on-chain users (a16z: under 100M, range 40-60M), not TVL, half of whose decline is just ETH/BTC price denomination. Rob says apps like Uniswap and Ethena benefit more than the infrastructure layers.
• Santiago links Ondo’s private institutional L1 to Hyperledger déjà vu — “just a database that looks like a blockchain” — while steelmanning it: settlement stays public, execution private, because “you can’t have order flow being public.” Rob’s bottom line is that “the direction of travel is tokenized assets. Everything’s going to be tokenized” — but which chain wins, “nobody knows.” Visa’s forward-deployed stablecoin platform could accelerate institutional adoption. Meanwhile crypto VC hit a record-low deal count, fund sizes shrank across Paradigm, a16z, Haun, Framework and Variant, and the re-up math is brutal: under 80-85% of existing LPs returning means “it’s a conversation. Every time.” Santiago says a $50M fund can still work if crypto includes stablecoin fintech, while a $1-2B crypto-only fund is difficult.

Deep dive

1. CLARITY at all-time-low odds: public confidence, private bearishness

• Rob’s read on the timeline optimism: it’s confidence theater — people talk it up “because they think that helps get it done,” but “if you talk to people in private… there’s a lot of bearishness.” Polymarket has slid from 46% last week to ~27%. Two open items: the ethics language on carving out Trump-family ownership in crypto projects (and whether the FBI or state AGs could bring complaints), and BRCA developer protections, where Sen. Cortez Masto put out supposedly agreed language that Republicans and industry “would not have agreed with.”
• Timing risk compounds: maybe a vote next week, possibly on partisan lines, then September — “the longer this goes on, the less likely it is that we get to an agreement, but I’m still hopeful.”
• Rob won’t trade it: “I wouldn’t buy the 30% on poly, I think it’s appropriately priced” — but both the SEC and CFTC are moving on rulemaking regardless, so crypto is better positioned in 6-9 months either way, even if rulemaking can be rolled back under different chairs.
• Santiago’s case for passage in its current or a close-to-current form: a bill “provides a legitimacy” and can be amended later; failure would be “much more detrimental” than any stakeholder’s marginal gripe. Rob: “100%.”

2. Every market now trades like crypto

• The scene-setting: semis up 20% in a day, trillion-dollar assets running “100, 120 vol,” and Rob thinks Korea has logged more circuit breakers this year than in the entire history of its stock market. Santiago’s frame: “crypto has foreshadowed what is happening in every other market” — the world molded to crypto, not the reverse.
• The mechanism: Santiago cites Goldman saying prime leverage in the final half of 2026 was the highest since before the global financial crisis, with ~20% of it in a few AI memory chip names; retail active traders are often the marginal buyer; the takers are increasingly quant. Result: “the long-term fundamental investor is struggling more than ever” — partly because LPs can’t handle it. Rob recounts LPs begging managers, “just don’t send me quarterly statements, please… send me something in 4 years, 5 years, 10 years.”
• The cautionary specimen: Leopold supposedly sold his entire $14B notional in a single block trade because he was 4x levered into memory stocks down 50% in a month. Hence Charlie’s rule — “ladies, liquor, and leverage. Out of the three, definitely avoid leverage.” JPMorgan now says 90% of system leverage has been rinsed; Goldman’s momentum index is worse than in 2008.
• Who’s actually winning: the best crypto liquid funds over five years are all delta neutral — Santiago says “market neutral funds and lawyers are the best performing people in crypto.” Rob also describes a London market-neutral crypto shop managing close to $4B, though this year has been tougher than 2022-24. Contrarian tell: Ken Griffin, historically focused on short-term pod books, now says he’s more interested in long-term fundamental investing.

3. AI fatigue reaches the LPs — but capex and Anthropic’s ARR hold the trade together

• Rob, on the West Coast doing LP meetings, heard something new: “a little bit of AI fatigue” from endowments and fund-of-funds where three months ago appetite was insatiable. One LP compared it to crypto 2021: these are “still just software” companies that changed business model — “there’s no reason it deserves a 100x multiple.” The rumored $10B OpenRouter deal on a $140-150M revenue business: “it is insane to me.”
• Santiago’s parallel: AI is bucketed the way crypto once was, and needs slicing — hyperscalers, neoclouds, memory, training and inference — with Gavin Baker’s take that pockets are “extremely overvalued,” especially private rounds that go $3B → $8B → $16B in three months where nothing changed “other than you missing out on investing in the first round.” He sees the most risk at Anthropic-style valuations if open-source models eat into token-spend margins — “I’d rather own other stuff in long AI.”
• The bull-side ballast: Meta, Google and Microsoft all maintained or increased capex guidance, Satya Nadella posted on AI ROI, and Rob thinks Meta said the useful life of these racks is much longer than anticipated. Rob cited a tweet saying Anthropic was close to $150B ARR as of July; Santiago called it “the fastest-growing ARR in the history of mankind.” Santiago said he doesn’t know anybody who thinks it goes public for less than $1.2T; Rob cautioned that the extraordinary growth is also part of the reason there could be a very big correction, without invalidating the long trade.

4. Token spend is the next battleground

• Rob’s stealth portfolio company — a router whose “only other competitor” is OpenRouter — cut one heavy client’s cost by 80%; Coinbase built an internal router, and Ramp launched a more closed-loop one. His call: with cheap open-source models improving and most workflows not needing frontier capability, “being able to charge people for the most frontier models… is probably going away in the vast majority of industries within the next 12 to 18 months” — forcing OpenAI and Anthropic toward net-new business models.

5. Robinhood is the tape to watch for crypto flows

• Earnings: $1.3B in earnings, up 32% YoY, beating consensus — with prediction markets generating $156M in revenue versus $100M for crypto. Santiago also said the relevant volume was up 10x year over year from a low base. Santiago notes the prediction-market result was expected with the World Cup and says he will be watching July and August.
• Structurally, Robinhood drove over half of Kalshi’s volume, bought NYX/the old LedgerX business, JV’d with Susquehanna on its own exchange (“RothEra” as spoken) and is migrating flow there: “they want to own that full stack.”
• The chain, four weeks in on Arbitrum: $12B in DEX volume, 100M transactions, ~328k holders — per Frank’s tweet the largest network by tokenized stockholders, surpassing Solana, BNB, Ethereum and Base. Rob’s pushback: wallet-count methodology is tricky and active accounts “have actually trended down after that initial bump… Same thing happened to Base.” Santiago’s wanted catalyst: merge the genuinely good Robinhood wallet with the core app, Coinbase-style — “a humongous unlock.”
• Santiago’s thesis: “Robinhood is probably the most important company” to track — if AI fatigue rotates money into crypto, “you’re likely going to see it first” on Robinhood Chain, the role BNB Chain used to play. Tokenized stocks are ex-US only, and he’s watching that metric as the true test of the Hyperliquid/HIP-3-style thesis.
• The incumbent tell: CME is suing the CFTC over perps while sounding positive on perps in its own earnings call, and probably plans sports-linked prediction contracts — “talking out of both sides of their mouth.” Rob: “you try to get in front of the innovation; when you can’t stop it, you embrace it.”

6. The rest of on-chain: $5B L2 TVL, record hacks, and the metric that matters

• The Block has all other Ethereum L2 TVL back to roughly $5B — last seen in 2023. Santiago flags that L2BEAT put the ecosystem at ~$33B and questions the methodology, while Rob concedes the trend is one direction — and adds that about half of the industry TVL decline is just ETH and Bitcoin prices, since TVL is denominated in them.
• Rob’s three-factor explanation for the bleed: low on-chain yields (the basis trade fell from 10%+ to 3-4%), record hacks, and AI becoming the momentum trade. He says H1 was the worst in crypto history — 212 exploits, over $1B lost, 3-3.5x last year, North Korea/Lazarus half or more, mostly social engineering rather than newly fragile smart contracts; Kelp and Drift were the biggest examples mentioned.
• Santiago’s preferred gauge over TVL: active on-chain users — he says an a16z report pegs them under 100M, in a range of 40-60M. Santiago identifies two countervailing forces: crypto-native vault activity re-rating risk-reward after April and the Kelp hack, while retail arrives through Robinhood and neobanks that abstract the chain away.
• Rob says applications such as Uniswap and Ethena benefit more than the Arbitrums and Ethereums.

7. Ondo goes private — Hyperledger déjà vu versus Robinhood’s open bet

• Santiago describes Ondo’s announced institutional L1 as private but verifiable, non-custodial but without necessarily allowing independent validators, and links it to the Hyperledger/Enterprise Ethereum era a decade ago: “sort of just a database that looks like a blockchain” without much of the trust-coordination utility of blockchains.
• Rob is skeptical that he understands the architecture, later likening it to “a dark pool with some sort of blockchain settlement.” Santiago’s steelman is that settlement, and perhaps verification, stay public while execution is private because “you can’t have order flow being public.” He notes that TEE work is drawing more excitement than it has in a while.
• Santiago would like a Canton-vs-Ondo-vs-Figure discussion, noting that Figure has tried to make Provenance open and permissionless but “just haven’t been able to get other people to build there.”
• The through-line: Santiago says “the direction of travel is tokenized assets. Everything’s going to be tokenized… but whether that is Ethereum or Base or Avalanche or Monad or Solana or Canton and a private Ondo — nobody knows.” Direct-to-consumer players take more risk than institution-serving players, and Robinhood’s history of hurting the Schwabs “doesn’t say they’re not right about what the world will actually look like.” Rob says both models can win.

8. Visa’s forward-deployed stablecoins and the crypto VC freeze

• Santiago’s closing positive: Visa’s new stablecoin platform is a “forward-deployed model” from AI applied to stablecoins; his portfolio company Velocity does the same institutionally. A year past GENIUS, every institution wants a strategy but cannot assemble the pieces, and Visa’s approach could cause “another wave of adoption or a quickening of growth.”
• The VC picture: record-low new crypto deal count, with dollars holding up via later-stage deals; Series B raises are “really, really tough,” and outside defense, robotics and AI-adjacent frontier tech, everything struggles to fund. “VC is a herding culture” — yet Santiago heard from “a couple of very big allocators” this week that “now is actually the right time to invest in crypto.” The open question is what their ICs say, since those contacts are the crypto champions at their firms.
• The re-up math: Santiago says “if you have less than 80, 85% from your existings, it’s a conversation. Every time”; Rob agrees that if existing investors come down, the fund size usually must come down. Sizes shrank across Paradigm, Andreessen, Haun, Framework and Variant.
• Santiago says a $50M fund can still be deployed if crypto includes stablecoin fintech, while a $1-2B fund that is deeply crypto-only is difficult right now. Andreessen’s $2.2B fund also does liquid investments such as Bitcoin and Solana.