Crypto's Value Capture Problem & Why Robinhood Built Its Own Blockchain

Summary

• Santiago Santos says eight months of “turning over stones” across 20-plus sectors have not validated Inversion’s founding hypothesis: that buying a business and embedding crypto can produce an above-market return. His core mistake was filtering for businesses where crypto might release value, which selected for “melting ice cubes”—including cash-poor MVNOs and challenged remittance firms. The new mandate is to buy quality, cash-flowing businesses where AI, crypto, or other best practices can make them better, while remaining open to being “corrected now rather than later.”
• The sector autopsies are the tradeable detail: on-chain B2B borrowing really costs 8–12%, not the 2–5% headline rate, so it is not competitive with TradFi; crypto FX aggregators quoted 20–50 basis points above mid-market in exotic corridors, a liquidity problem rather than a technology problem; and remittances are a stakeholder-incentive problem, not a technology problem. Nubank spent over $1 billion in Mexico and converted around 9 million users, making an MVNO with no cash flow bundling fintech “a hard sell.”
• Jason Yanowitz’s suggested expression for being long crypto adoption in businesses is “go buy Stripe secondary and go buy Robinhood equity.” Larger incumbents may capture more of the value than small businesses implementing crypto. Santos says value accrual remains unresolved because open-source networks can push value toward consumers, and the ROI of implementing crypto is currently nowhere near the ROI of implementing AI.
• Jason argues Robinhood Chain is very real: more than $500 million in Uniswap volume—more than any chain except Ethereum mainnet—and about 150,000 new active wallets in a day. He expects Robinhood Chain, Base, and BNB to become some of the largest chains in a couple of years as centralized-exchange activity increasingly moves on-chain. Santos questions whether an operator-run validator set is really more than a database renting some security from Ethereum.
• On VVV-style dual token-and-equity structures, Jason makes the strongest case for one instrument and one narrative: retail crypto is perception- and narrative-driven, so the “Schelling point” should be one token. Santiago adds that strong disclosure and trust could allow a two-asset structure to work; Jason agrees in principle but says the market has already shown a preference for what looks like a pure token—the token is down.
• The AI segment is a workflow case study: Claude Design lets Blockworks produce landing pages or slide decks in about five minutes, reduced Figma usage to roughly 5% of its former level, and can generate a sales deck in two minutes from Granola notes, HubSpot, and Slack. But Fable is extremely expensive, so Santiago argues for model routing: cheaper models for most work and Fable for the roughly 10% that needs it. Jason recommends a dedicated internal AI engineer for routing, caching, visibility, and workflow automation. Falling token prices and low penetration make Santos “incredibly bullish” on adoption, even though the winning lab is unresolved.
• Santos argues that retail “degenerates” should be reframed: “it’s an affordability crisis. These are lottery tickets.” He and Jason discuss hiring younger, AI-native employees, while Jason says AI is “the great enabler of starting new projects”: zero-to-one can take hours, but taking a project from one to 100 still requires humans.

Deep dive

1. Inversion’s shareholder letter: the thesis is not yet validated

• Santos opens with the letter’s closing line: “if our investors have seen hard evidence to the contrary—in operating results, not pilots—we would rather be corrected now than later.” The founding hypothesis was: “Can you buy a business and make it dramatically more efficient with crypto?” After eight months of systematic work across 20-plus sectors, he has not found evidence that the resulting returns clear the risk hurdle—not because the technology is unreal, but because implementation may not create enough value.
• The load-bearing admission is that filtering for businesses where crypto can release substantial value leads to companies that are not great businesses—“melting ice cubes.” Jason translates this as catching a falling knife in the private-equity version of the term.
• Santos returns to the investing frame: “Rule number one of investing is, ‘Don’t lose money.’” He says it would be irresponsible to buy a poor business merely to prove that crypto can be implemented.

2. MVNOs and on-chain credit: the unit economics do not clear

• The MVNO thesis was to bundle telco and fintech, M-Pesa-style, to build “the largest fintech in a place like Mexico.” It broke down because MVNOs rent infrastructure, compete heavily for customers, experience substantial churn, and do not generate meaningful cash flow to fund the transformation.
• The killer comparison was Nubank, which Santos calls probably the best-run fintech in the world absent Revolut: it invested over $1 billion in Mexico and converted around 9 million users. “If they can’t do it for a billion and I don’t have cash flow to do it,” he says, the thesis is a hard sell.
• On-chain credit works mostly for crypto-backed loans. Outside that collateral, investors face smart-contract risk, variable borrowing rates, and utilization curves that are difficult to underwrite. Santos says the true rate for B2B business lending is closer to 8–12% on-chain today, not the 2–5% headline DeFi rates. Jason adds that a business borrowing at 20% is probably not a good business.

3. Remittances and FX: incentives and liquidity, not technology

• After hearing Western Union claim it could substantially reduce prefunding, Santos told the team to examine almost any remittance company they could buy. The conclusion was that fixing remittances is “not a technology problem—it’s a stakeholder-incentive problem.” Stablecoins can move money in real time, but stakeholders may not want to implement them. Santos would rather wait and see whether Western Union can make the model work before reengaging.
• In FX, Inversion ran an RFP for a large multinational industrial company’s global treasury flows and asked crypto providers for their best rates. Most quoted 20–50 basis points above mid-market, especially in Colombia, Egypt, and Israel. Santos calls this a liquidity problem rather than a technology problem and says it is not the risk Inversion wants to undertake now.

4. Distribution was the wrong lens; quality is the filter

• Santos revises his earlier “buy cheap distribution” framing: “not all distribution is created equal.” A large installed base does not matter if a company lacks pricing power, brand loyalty, profitability, or receptive customers. Upselling is not automatic.
• In his revised framework, a business that has produced cash flow across different market environments has already demonstrated something important. He also makes the counterintuitive claim that a business easily transformed by technology may not be a good business.
• Jason offers media as an example: newspapers were transformed by technology, but the newspaper companies were not the primary winners; value shifted to platforms such as TikTok and Facebook.

5. The new mandate: Berkshire and Constellation, using a holding-company model

• The update is to broaden the mandate and “go buy good businesses” where technology—AI, crypto, or other tools and best practices—can make the business better. Santos points to Berkshire Hathaway and Constellation Software as models.
• He describes Mark Leonard starting Constellation with C$25 million in 1995 and building the business through many $2–4 million software acquisitions. Over time, Leonard accumulated a proprietary understanding of what makes a good software business—margins, growth, sector, vertical, horizontal, and related metrics—so capital-allocation decisions became much faster.
• The process also validated Santos’s belief that a holding-company model is better than a fund model. A fund with a crypto mandate can become trapped by its promise to LPs even as managers become interested in robotics, AI, or other areas. He cites Paradigm’s pivot toward AI as an example of adapting to evidence. The game of investing, he says, is truth-seeking, and it would be irresponsible to buy poor businesses merely to preserve a thesis.

6. “Bearish crypto?”—bearish the ROI, not the technology

• Santos distinguishes between the technology and its value capture. He says the technology, business interest, willingness to implement, stablecoins, and tokenization are real, but “the value accrual is still not resolved.” Open-source networks can push value away from companies and toward consumers, with technology’s disinflationary curve becoming consumer surplus.
• He still does not know whether Tempo will become a profit center for Stripe or primarily improve Stripe’s existing business. He similarly expects Stripe and Robinhood equity to benefit, while remaining uncertain about the value captured by smaller businesses.
• The competitive-moat problem is that if one acquired company uses a technology solution to lower costs, competitors may adopt the same solution. Once everyone adapts to the new normal, the cost advantage disappears and the businesses may simply rerate lower. Santos says many crypto founders he asked could not name traditional-business clients worth considering as acquisition targets.
• Jason, not Santos, offers the direct trade expression: “If you really wanted to be long this thesis of businesses implementing crypto, honestly, go buy Stripe secondary and go buy Robinhood equity.” Santos’s broader conclusion is that the ROI of implementing crypto is currently nowhere near the ROI of implementing AI.

7. Fable and Claude Design collapsed Blockworks’ workflows

• Jason says Claude Design can turn Blockworks’ design system and creative assets into a landing page or slide deck in about five minutes. After the Messari acquisition, a batch of landing pages that previously would have moved through product, design, copy, and engineering took 24 hours, with the engineering team using the generated ZIP file. He says the design team now uses Figma at roughly 5% of its former level.
• Sales teams can combine Granola notes, HubSpot data, and the relevant deal’s Slack channel to generate a tailored deck in about two minutes. Jason frames this as a major change for sellers who would rather close deals than build presentations.
• Jason says Fable 5 is the first tool that made him think people will definitively spend less time at a computer. Santos describes Inversion’s AI-assisted research similarly: enriched company databases and call notes that once required weeks, an army of analysts, and expensive subscriptions can now be produced within hours.

8. AI pricing is a management problem—and a bullish adoption signal

• Santos says Fable is “ridiculously” expensive and consumes tokens at a rate unlike any model he has used. He estimates that 80–90% of Blockworks’ work could use less expensive or earlier models, with roughly 10% warranting Fable. He mentions Opus 4.8, prior ChatGPT models, GLM 5.2, and Kimi 2.7 as part of the routing question.
• Jason emphasizes the operational response: an LLM gateway, better routing, caching, leaner context, spending visibility, and a full-time internal AI engineer focused on workflows, model dependence, memory, and unnecessary software spend. The transcript’s explicit recommendation for that dedicated employee comes from Jason.
• Jason cites similar performance at dramatically lower cost from models released by Cursor and Meta. Santiago frames falling token prices alongside very low AI penetration as a Jevons-paradox argument and calls the adoption outlook “incredibly bullish,” while saying it remains unresolved whether Anthropic or OpenAI wins.
• During the episode, Santiago says a Fable 5.6 livestream is underway and that Sam Altman announced ChatGPT Work, a new desktop app, and hosted sites. He and Jason also discuss a pricing statement from Zuckerberg and an “epic pricing war” among AI companies.

9. Robinhood Chain: real numbers, verticalization, and a live disagreement

• Jason reports more than $500 million in Uniswap volume on Robinhood Chain—more than any chain except Ethereum mainnet—up 10x day over day, plus roughly 150,000 new active wallets in a day. He argues that Robinhood is verticalizing the stack to control its users and margins rather than sending activity to Ethereum, Base, or Solana.
• Jason believes centralized-exchange trading will increasingly move on-chain and that Robinhood will want to own the destination. He expects Robinhood Chain, Base, and BNB to become some of the largest chains in a couple of years, and calls Kraken’s execution under Arjun phenomenal.
• Santiago questions whether the chain adds enough efficiency over Robinhood’s existing database. More infrastructure creates more surface area for failure, so he asks where the benefit appears in the P&L. He also questions whether an operator-run validator set is really a chain or simply a database renting some security from Ethereum.
• Both discuss faster onboarding and regulatory constraints. FOMO reportedly added 12,000–15,000 users in one day, many arriving from TikTok and Instagram rather than from crypto-native circles. Santiago notes that on-chain venues can offer faster onboarding and broader asset access, while Jason summarizes the recurring pattern as “it’s always been regulatory arbitrage.”
• Jason relays Mike’s view that Coinbase’s older holders of Bitcoin and ETH create a natural borrow-lend market for Base, including protocols such as Morpho, while Robinhood may attract more retail and memecoin activity. The resulting distinction is that Coinbase could look more like a corporate chain on Ethereum, while Robinhood looks more like Solana. Jason says he finds the take persuasive but wants to think about it further.
• Santos argues that “degenerate” is the wrong label for much retail behavior: “It’s an affordability crisis. These are lottery tickets.” He says markets are more momentum-driven, hedge funds are adapting, and this is the new normal. Jason cites Thread Guy’s argument that memes and crypto fundamentals may be foundational to modern finance rather than a temporary phase.

10. One instrument, relevance, and AI-driven busyness

• On VVV-style dual token-and-equity structures, Jason lays out the strongest one-instrument case: crypto is retail-driven and highly dependent on perception, narrative, and a clear focal point. He argues that ETH, SOL, and Hyperliquid demonstrate the appeal of pure tokens and that the Schelling point should be one token. He invokes Liberty Media’s tracking stocks as a more specialized structure supported by disclosure and tax considerations.
• Santiago adds an important caveat: if a project has two assets but maintains strong communication, disclosure, and transparency, it could potentially work. Jason agrees that such a structure is possible, but says the market has already shown its preference for something that looks like a pure token. His conclusion is that the token is down, “end of story.”
• Both identify Jeremy Giffon on Invest Like the Best as their content of the week. Santiago recalls a health-startup dinner where billionaires wanted to sit next to Nobel laureates, leading him to argue that relevance can eventually matter more than money. Jason pushes back that 99.99% of people would still be thrilled to be billionaires, while agreeing that wealth alone may no longer confer relevance in elite circles.
• Santiago says many billionaires who felt important five or ten years ago now feel less relevant, while Jason jokes that posting on Twitter is required to remain relevant. They also discuss younger employees: Santiago favors constantly hiring younger people, and Jason says some of Blockworks’ best new products have come from employees under 30 who understand the market and are AI-native.
• The closing AI discussion is about busyness. Jason says AI is “the great enabler of starting new projects,” which has contributed to Blockworks doing too many things. Zero-to-one can now take hours, but moving from one to 100 still requires people, execution, and new skills. Jason says Robert, whom he estimates to be roughly 23–25, turned a request for Friday into “I’ll have it to you by tonight,” then asked, “Have you not used Fable?”