On My Mind

Finding experts in LLM backrooms.

I’ve had an increasing number of conversations with friends about the ways in which we use AI. It seems like there’s been a dramatic increase since the start of the year (especially among my friends in finance).

So today, I wanted to share one of the methods I’ve found most helpful: having Claude take on the perspective of category experts when trying to learn about a new, emerging, or evolving topic. In particular, I’ve used this technique most frequently in the context of informing macroeconomic views.

Throughout the first few months of this year, I set up several chats oriented around different macro-economic topics – private credit & reinsurers, the Iran conflict, etc. – and instructed each chat to take on the perspectives of a number of widely known, respected investors (e.g. Howard Marks, Ray Dalio, Mohamed El Erian, Stanley Druckenmiller, and more). The goal was to learn as much as possible, as quickly as possible, about the relevant news each day. I didn’t want just any perspective; I wanted that of experts.

Now, the first thing I do most mornings is open each chat and ask Claude to “update the investor perspectives with new and relevant information.” The result is that I keep up to date with current events much better than I used to AND I feel significantly more informed about the second-order consequences of much of what’s going on in the world.

So, why has this worked well? Each selected expert has a large corpus of public material that the LLM can reference and synthesize. Specifying the investors narrows Claude’s context window to the parts of its knowledge base I find most relevant. Instead of asking the macro questions into an abyss – and not knowing the origin of each answer I’d get back – I now have some grounding as to the rest of the “persona” and potential bias behind each viewpoint. Seeing the diversity of perspectives also helps illuminate any consistencies / throughlines. And sometimes, I’ll even ask Claude to simulate a debate between two of the different investors.

It seems to me this pattern will be extended and generalized to all types of work. What I’ve described above is a type of knowledge work. The next iteration seems skills-based, where instead of asking for expert knowledge you’re paying for expert labor. That labor could include software engineers, lawyers, designers, consultants, music producers, and more. In the same way that I ask Claude for someone like Mohamed El Erian’s perspective on macro, a company might only hire a software engineering agent modeled after Andrej Karparthy or a designer that is trained on the work of Jony Ive. The market for licensing one’s likeliness seems like it’s, in aggregate, about to get so much bigger – but for only the top 1% within each category of labor.

There’s a lot to dig into if you keep pulling on the thread: how would highly skilled people make their services legible and replicable? What does economic attribution look like? What happens to the folks who aren’t experts? And many more questions. I’m certainly not the first to write about this, and I definitely won’t be the last.

Subscribe now

Frameworking

This week’s framework is borrowed from a passage I read in Andre Agassi’s memoir Open.

The framework: are you a thermometer or a thermostat?

A thermometer reads the temperature of a room. A thermostat can change and control it. Knowing who is what can help in almost any setting.

For instance, think about constructing a team. Thermostats can be great leaders: they set the tone and energy of a project. But placing two thermostats in adjacent roles can create conflicting temperatures and, consequently, breed discomfort. On the other hand, a thermometer adjusts to the temperature in the room. Sometimes that means thermometers are great at going with the flow; other times it might look like significant ups-and-downs if the overall environment is chaotic.

One isn’t more right than the other. Both thermostats and thermometers can do well at work and in social situations. The key is simply to recognize which one you are – and which one your colleagues are – to create the most harmony.

Content I Consumed this Week

1. The Drama – the new Zendaya and Robert Pattinson movie from a24. I rate it either an A- or B+ – watchable but probably not rewatchable. Would recommend going in knowing as little as possible about the movie. Also best watched in a theater; the collective gasps from the audience definitely elevate the experience.

2. Currency Wars by James Rickards – about a third of the way through. The weather was exceptionally nice in NYC this weekend, so I stopped by my local Barnes & Noble to pick up a book to read in the park. This one caught my eye. So far, the book has provided a solid history of the ways in which currencies have evolved and how nations have used currency markets as attack vectors.

3. Redpoint’s Software Market Upside. I LOVE a good deck. Decks might be my favorite form of content. I don’t agree with everything in the deck, but there were certainly a number of great and thought-provoking charts. There’s so much uncertainty around the terminal value of SaaS companies right now. Whoever can figure out what survives (and thrives) on the other end of the AI disruption is sure to make a fortune.

Books on the docket: The Battle of Bretton Woods by Ben Steil

Costs of replication.

When looking at a business, one of my favorite things to think about are its costs of replication – that is, how hard would it be for another entity to reproduce the product being offered? That new entrant might be a startup looking to compete, or it could be a legacy company expanding to adjacent fields.

I typically bifurcate costs of replication into two parts: 1) the costs of production, and 2) the costs of everything else. Costs of production encompasses the things required to actually build and ship the product. “Everything else” might look like user acquisition, licensing, entrenched consumer bases, complexity of customer integrations, partner exclusivity agreements, proprietary data, and more.

Startups tend to emerge and succeed when incumbents’ defensibility is disproportionately weighted towards high costs of production, and then there is some disruptive force that makes it significantly easier to produce the product. A modern-day example might be defense tech: it historically required significant capex to produce the technology but now, with the capabilities of things like AI and advanced manufacturing, we are seeing a rise of new defense tech companies (“neoprimes”).

But it’s a double-edged sword. Lower costs of production may help startups get to market with less capital, but they also may make it easier for incumbents to fast follow. There’s a famous VC adage: “can the startup get distribution before the incumbent innovates?” The buffer that a startup may have had as its potential competitors were thinking through whether to allocate the resources to a buildout is now gone – the necessary internal resources (e.g. engineers) and time for a competitor to build the product have collapsed. Defensibility re-weights toward the “everything else” portion of replication.

I’ve been using this framing to think about where startups win vs. where incumbents can continue to compound. It’s especially relevant given the advances in AI. A product that used to take 5 engineers at 2 months to ship might now take one engineer, a massive token budget, and a few days to ship. In many categories, defensibility will increasingly shift from costs of building to deciding what to build and how to distribute it.

Frameworking

Peter Thiel has a famous framework for how to make money: a 4x4 of consensus vs. non-consensus, and being right vs. wrong. Being non-consensus and right leads to the most successful investments. Being consensus and right can still result in profitable outcomes, albeit often not nearly as big. And if you’re wrong, you lose money.

This week’s frameworking riffs on Thiel’s framework. It’s about how to be contrarian and, as a byproduct, the different ways to make money.

The opportunity to make a successful investment most often comes when there is space for people to change their minds. That shift – where one side switches their perspective – is often an important driver of changes in capital flows.

There are two ways to be contrarian. They can be visualized similar to a decision tree:

1. Step one: will a business exist? If many people believe that a business or market won’t exist – and then it does – there becomes a lot of room for people to change their minds.

2. Step two: everyone already agrees that a business or market will exist. The point of disagreement is about how big the business can get, or whether that growth is already priced in.

The first point of disagreement is frequently how early-stage venture capitalists make money. The latter is how growth-stage investors and (and some public market investors) tend to place bets.

In particular, betting that markets will be even larger than many believe is a core principle in underwriting extremely high P/E multiples. A fast-growing company in 2026 that trades at 100x TTM earnings is extremely expensive. But what if you believe their growth will double next year, then double again the year after that? That company is then trading at a 25x multiple using 2028’s earnings.

The most lucrative investments can come from being doubly contrarian, and doubly right – believing that a market will exist when others don’t, and then betting that it will get far larger than most people expect (even after everyone is in agreement that the market is large and valuable).

Content I Consumed this Week

1. Steve Carell on Amy Poehler’s Good Hang Podcast. Mafias and talent scenes are everywhere. I’m consistently amazed by the (largely comedic) talent that came out of Northwestern, and then Second City in Chicago, in the 1980s and 1990s. Tina Fey, Amy Poehler, Steve Carell, Julia Louis-Dreyfus, Stephen Colbert, Halle Berry, and so many more. It’s so cool to hear the stories about all of their early days, and Amy Poehler is such a good interviewer!

2. Marc by Sofia by Romy. While watching this, it occurred to me that I’ve maybe never seen a longer-form interview where the interviewer is Gen Z. Really enjoyed seeing the interaction between Sofia Coppola and her daughter Romy.

3. Why Florence Started the Renaissance by @tomaspueyo. Tomas’s blog, Uncharted Territories, is one of my favorite weekend reads. Every time I read one of his essays, I learn something new about how human civilization has evolved.

Books on the docket: The Infinity Machine by Sebastian Mallaby. If you’ve made it this far, please give me book recs!

On My Mind

Today’s topic: vehicles to enable individuals to invest in private companies.

People want to invest in the companies building the future. The software engineer who uses Claude Code to 10x her productivity may want to invest in Anthropic, both because she sees first-hand how it’s disrupting coding and because she wants to hedge against any risk that AI puts her out of a job. A lawyer using Harvey may feel similarly. Same with the musician using Suno. The list goes on.

Yet these companies are staying private – often with significantly higher valuations than historical averages. I jokingly tweeted that we now need a name for private companies that exceed $1T in valuation. There’s undeniable demand for investors to get exposure to these companies. Everyone wants to own a piece of the future.

So, on my mind is the question: what’s the best vehicle to give investors exposure to these assets?

There’s two ways to decompose the set options: with each company’s permission, or without. With a company’s permission is challenging. One can invest directly in a private funding round (largely inaccessible for the average Joe). Sometimes companies allow employees or early investors to sell secondaries (also often difficult to get access to). There are a growing number of SPVs that attempt to provide exposure – but they’re typically riddled with fees and may lack investor protections.

The path I’m more interested in is the latter: products that provide exposure without needing a company’s permission. These are likely to be some form of a derivative. And we do have some early attempts, like Robinhood’s synthetic share tokens and select pre-IPO perpetual markets. One could imagine prediction markets eventually serving as a directional venue for exposure, where markets for the over/under on a company’s next private round valuation provide some upside related to the company’s appreciation (even if far from a 1:1 with the payoff structure from a direct investment).

But there are challenges with all of the above. And personally, I’m unconvinced any are the form factor most investors have been looking for. Low liquidity limits volumes. Severe information asymmetries may undermine the integrity of the markets and/or impact the price oracles.

This week I was struck by another construction: venture funds going public. Investing in (certain) VC firms may provide exposure to baskets of assets of many of the most important private companies. The idea was sparked by the launch of Robinhood’s Ventures Fund I, which will be a publicly traded vehicle. Its focus is to invest in highly desired private companies. By going public, the shares will be highly liquid, widely accessible (i.e. no accredited investor credentials required), and marked to market daily. The premium or discount to NAV may even become a mechanism for price discovery for some of the underlying assets the fund holds!

My guess is that, in the years to come, we see more venture firms go public. And maybe we even see an entirely new crop of venture firms emerge with the primary intent of going public to give retail venues to build exposure to the companies building the future.

Subscribe now

Frameworking

Designing a slide deck.

I kind of love building slides. I’ve written about the art of a pitch deck before. I even spent last fall making a 150 page deck on crypto trends as a passion project. It’s an art.

This week’s frameworking was a buzzer beater, inspired by a conversation I had last night with a founder and friend. The question: how do you build the perfect deck?

Claude or ChatGPT can likely design a good slide. But it’s more like a paintbrush than the actual artist. You’re still the painter. And the role of the painter is to figure out the content that should go on the canvas (each slide).

So this week’s frameworking is: always ask yourself what question(s) each slide should answer.

People often put too many slides in the core of a deck. They think about the content they want to show and the work they’ve put in, not the set of questions they need to answer. By flipping this framing – and focusing on the questions first – the author puts themselves in the audience’s shoes.

If you can’t figure out what question a slide is answering, the slide probably isn’t that important. And if the question it’s answering isn’t a good one, the slide can probably go in the appendix.

Subscribe now

Content I Consumed This Week

1. The Stablecoin Podcast (self-promo) – I appeared on this week’s episode of The Stablecoin Podcast. It was great to jam on where value accrues, how I anticipate the infrastructure will evolve, and even share a few tips for how I separate the signal from noise when looking at markets.

2. Stripe’s Annual Letter – always a fun read. My favorite section was on the “five levels of agentic commerce.” I can’t wait for the day I can have an AI help me find the perfect jacket for 45-55 degree weather. That being said, I am maybe more skeptical than the letter authors of soon seeing a day when we’ll fully outsource purchasing decisions to agents. Nonetheless, a great example of frameworking.

3. Open (by Andre Agassi) – Agassi is one of the greatest tennis players of all time. I’m about halfway through his autobiography and would highly recommend. The most shocking part of the story? He continuously emphasizes and restates how much he hates tennis.

Books on the docket: Setting the Table by Danny Meyer; Hit Refresh by Satya Nadella.

I’m trying a new style of piece this week. It’s a selection of half-baked ideas I’m tinkering on for long-form essays combined with great content I consumed recently. The goal is to publish in this format weekly, with occasional longer-form essays as I develop the ideas mentioned here.

Subscribe now

On My Mind

M&A within crypto.

A number of crypto (or crypto-adjacent) companies that went public recently have been getting crushed in the market. Gemini is down over 80% from its IPO, sitting at just $700M. BitGo is down nearly 30% and valued at about $1.2B. Circle is down over 70% from its peak. EToro down >50%.

Part of this may just be air leaving the market as multiples (across many sectors) compress. Some of it may be macro. And some of it is certainly correlated with broader crypto asset price declines.

On my mind, though, is the question: at what point do these companies become attractive acquisition targets?

It feels like there are two powerful trends on a collision course. The first I mentioned above: crypto companies (that had strong enough businesses to go public) are struggling in the markets. The second is that many more traditional financial institutions – banks, exchanges, payment service providers, and brokerages – are rolling out crypto service offerings. I wouldn’t be surprised if the value from these crypto companies’ infrastructure, licensing, and customer relationships starts to make them attractive acquisition targets for strategic acquirers.

Frameworking

In December, I was lucky to meet a legendary investor and someone whom I admire greatly. What struck me most about our conversation was that he seemed to have a framework for everything. Any question I asked and he responded with a structure for how to think about the scenario. It was the type of thing where each answer came across so polished and considerate that he had surely iterated each framework over time. But at the same time, there were so many – and they all felt so natural, and specific – that there’s no way he could have practiced and prepared each one.

The lasting takeaway has been that I should spend more time building and applying frameworks. There seems little downside to having a prepared mind. If thinking is a muscle, frameworking feels like a healthy exercise to add to my mental gym routine. So here goes.

This week’s frameworking: give people something to react to.

If you want an answer, always provide a first attempt and let the other person react to it. If debating a topic, pick a side and choose to defend it as the starting point for discussion. Just avoid starting with an empty slate.

The framing is consistent with Cunningham’s Law: the notion that the best way to get to an answer on the internet is not to ask a question, but to post the wrong answer. People can’t help but tell you how you’re wrong and, by doing so, end up sharing what’s right.

Subscribe now

Content I Consumed This Week

1. Everything is Market by Jesse Walden. I particularly enjoyed the “markets as APIs” framing. It’s an expansive and provoking view of how software and finance will increasingly intersect, especially as two horizontal technologies (AI and crypto) hit inflection points.

2. Love Story – a new TV show on FX/Hulu that explores the relationship between JFK Jr. and Carolyn Bessette. I’m obsessed with it. One of the most striking (and admirable) things about the show is Carolyn’s wardrobe. I loosely knew that Bessette was a style icon in the 90’s but didn’t fully appreciate it until watching this show. (As an aside: it seems like half the female population in NYC has become inspired to start dressing like Bessette in the past week). The show also made me realize that putting effort into how one dresses will probably grow in importance in a world of AI. When more and more of the world is automated, one of the few things that can’t be edited or augmented is how someone shows up in person. Individual effort is and always will be a differentiator.

3. Andrew Reed on the Dialectic podcast. Lots of fascinating stories about how Sequoia has evolved and grown. TLDR: Jackson is a great interviewer and Andrew should go on more podcasts!

Books on the docket: Open by Andre Agassi; Setting the Table by Danny Meyer.

1. Write down people’s birthdays in your reminders apps, regardless of how close you are to them. It’s always a fun surprise to receive a happy birthday text from someone you wouldn’t expect.

2. If most of your time is spent responding to what other people tell you to do, you’re already behind the 8 ball.

3. Don’t watch the final episode of Game of Thrones.

4. Actually, never watch the final episode of any TV series.

5. Build personal lore – hobbies, interests, niche side quests and more that make you a unique person. And do it with gusto! Most of that lore should only be known to people close to you, though.

6. When you find socks that you like, buy way way more pairs than you think you need.

7. Start using Spotify DJ as soon as it comes out.

8. When you hear a song you like, add it to your Liked Songs playlist. You’ll never regret it.

9. Burnout doesn’t come from working hard; it comes from lack of momentum and/or direction.

10. The only way through is through.

11. Send more cold DMs.

12. Publish online. If the content is bad, no one will pay attention. But if the content is good, it’ll act as a force multiplier for many aspects of life.

13. Quality > quantity, but you’re also your own worst critic so bias toward putting the art out anyway.

14. Never post a photo online of where you are while you’re physically in that location.

15. Stairs rule: always take the stairs if you’re going <4 floors down or <2 floors up.

16. Let others get out of the elevator first.

17. Never lose the plot.

18. Books are excellent presents.

19. Getting good at asking follow up questions is a superpower.

20. Improv classes won’t make you funnier.

21. Barry’s Bootcamp is one of the best things you can do for your health.

22. Staycations are underrated. Sometimes you just need a week to read, do chores, and think.

23. The more frameworks you can develop and refine over time, the easier it becomes to make sense of the world.

24. If something makes you cringe, don’t recoil. Instead, try to dig in further. A strong reaction to anything is worth paying attention to.

25. When you see something that makes you think of someone, send it to them.

26. Be nicer to your twin. She’s awesome.

27. Don’t be afraid of getting older; embrace it! Birthdays are worth celebrating and you’re lucky to have each one.

    

I started working on this essay shortly after watching Roger Federer’s 2024 Dartmouth commencement address. In it, he compares many aspects of tennis to life. Looking effortless actually requires immense effort. Winning the game doesn’t mean winning every point. And there’s more to life than the stage on which one performs.

Federer’s speech got me thinking about other ways in which tennis teaches valuable life lessons. And as we begin the new year, it seemed like as good a time as any to finally hit publish.

One of the more meaningful skills in tennis is to avoid unforced errors. An unforced error is when you lose a point by your own doing – hitting the ball out of bounds, into the net, etc – that should have been easy to place in the court. They’re considered unnecessary mistakes. In contrast, a “winner” is a shot that the opponent cannot return (thereby directly winning the point).

Many of the best players in the world still lose a significant number of points because of unforced errors. But the very top avoid unforced errors more than the rest. Points in a Federer vs. Nadal match, or an Alcaraz vs. Sinner one, are won with winners.

Avoiding unforced errors put these players in positions to hit more winners.

I’ve found this a useful framing for approaching life. Want to get super fit? The “winners” might be all of the health optimizations – focused athletic training, supplements, maybe even cold plunges and sauna. But the avoidance of unforced errors matters more: limit packaged foods and refined sugar, try to move your body daily, and get enough sleep. Want to do well at work? Strive to be reliable and kind. Being unpleasant to be around is a massive unforced error that often supersedes any of the winning qualities that might otherwise be present (strong work ethic, quality performance, etc).

Just as reducing unforced errors enables top tennis players to hit more winners, the avoidance of mistakes in whatever you choose as your “sport” helps create the conditions to fully excel. It sets a higher baseline. It puts you in more positions of opportunity. And it lets your winners shine.

One could argue that a major difference in tennis vs. life is that, in tennis, you’re playing against a direct opponent. Avoiding unforced errors means your competitor has to play a little longer, a little harder, a little more strategic. And maybe your endurance prompts them to make an unforced error, leading you to win the point. That dynamic doesn’t apply 1:1 in life. Your main “opponent” is often yourself.

But maybe these are less different than they seem. When a player hits an unforced error, they’re defeating themself. Losing a match because you played poorly feels far worse than knowing you left it all on the court. The most fulfilling successes come from hitting real winners, not from one’s opponent losing. That’s true in tennis, in business, and in life. And it often starts with avoiding unforced errors.

I’m excited to publish the first annual edition of my Crypto Trends Report!

This report takes the perspective that the crypto industry’s growth can be viewed as a series of three compounding s-curves: asset creation, asset accumulation, and asset utilization.

Phase 1 is creation: the tokenization of value. The creation phase started with Bitcoin’s launch in 2009. Since then, it has grown to include layer-1 monetary assets, project tokens, stablecoins, content coins, memecoins, NFTs, tokenized equities, and more. From 2024-2025, the crypto industry moved up the steepest part of the s-curve. We went from ~20k listed, tradable tokens to millions in the course of just a few years. There is still significant room for innovation and growth (tokenized credit, onchain structured products, additional real-world assets, and much more), but the most dramatic zero-to-one moments have already occurred.

Subscribe now

Phase 2 is accumulation: the more assets that exist – and the more valuable they become – the more that people want to hold those assets. This provides tailwinds for multiple sub-sectors within the industry: custody products, exchanges, and security solutions. Different custody solutions service different user groups: stablecoin-enabled apps may use embedded wallets like Turnkey, institutional investors often use qualified custodians, and active onchain users may use super-app-esque wallets like Phantom. The growing demand to buy, sell, and hold crypto assets also catalyzes increased distribution. Many incumbent exchanges (e.g. Coinbase) have seen volumes skyrocket. Traditional fintechs (e.g. Robinhood) have doubled down on crypto access, while newer venues (e.g. Axiom) have emerged with explosive growth. Asset managers are offering crypto access in retirement accounts, public companies are beginning to hold bitcoin and stablecoins on their balance sheets, and even some sovereign wealth funds are beginning to accumulate these assets. We have just begun to climb the steepest part of the s-curve.

Phase 3 is utilization: putting assets to work. Once people hold assets, they want to start using them. Crypto assets are the most composable, accessible, and programmable financial assets that have ever existed. A number of exciting, robust use cases have begun to emerge: stablecoin payments, supplying & borrowing capital on lending protocols like Morpho, providing liquidity to onchain exchanges, and network staking. Yet we are still at the very early innings of the different ways in which tokenized assets can and will be used; we have barely begun to climb the s-curve. I see the design space for expanding and enabling greater asset utility as some of the largest and most exciting blue ocean opportunities over the coming years.

The deck is split into five broad categories: Macro, Stablecoins, Centralized Exchanges, Onchain Activity, and What’s Next. Each section loosely uses the framing of the three compounding s-curves to help contextualize the major trends and opportunities that exist in the crypto industry today.

Subscribe now

I’d recommend reading the deck in full. But if you only have time for the highlight reel, here are some of the major trends we see in crypto today:

1. Macro. The major crypto assets keep getting bigger. Even as the total market cap of crypto has grown, the concentration of value among the top 10 crypto assets has remained shockingly consistent. It has also been challenging for new assets to enter the top five. There is real Lindy-ness and mimetic power to many of the top assets.

2. Stablecoins. Stablecoins have experienced explosive growth, across all three s-curves: new issuance (asset creation) is occurring at record pace, there are many new venues for on-ramping (accumulation), and we see an increasing number of ways to put stablecoins to work (utility). There are strong network effects within the category: the more stablecoins in circulation, the more useful they become. Similarly, the more entities that hold stablecoins, the greater one’s ability to build products and services around those assets. Several areas that have benefited to date from stablecoin productization are payments, earn products (e.g. lending protocols), and exchanges. Yet we believe there is still significant room for innovation across all three s-curves.

3. Centralized Exchanges: Centralized exchanges are some of the clearest winners from the “accumulation” s-curve. Both volumes and revenue have grown dramatically over the past five years. And notably, a second-order effect of the growth in trading is that users then hold those assets on exchanges – enabling exchanges to offer an expanded set of utility (e.g. staking, yield, and lending products). Many of the new ways to utilize crypto assets will end up getting built directly onchain, but may find strong distribution through centralized exchange integrations.

4. Onchain Activity: Curious about new, novel ways to utilize crypto assets? Then look onchain. It’s a hotbed of experimentation. Every element of the asset lifecycle – issuance, liquidity provision, exchange, listing, and productization – is permissionless and open on blockchain rails, whereas every one of these steps is gatekept in traditional finance. Such a setup dramatically expands the design space for creation and utilization. Furthermore, several categories of onchain protocols have begun to mature into solid businesses with strong fundamentals. Decentralized exchanges, lending platforms, token launchpads, and perps exchanges all have exciting signs of product market fit.

5. What’s Next: Building on crypto rails can transform products into platforms. This section uses the growth in prediction markets as a prime example of a new market that’s supercharged by crypto rails. It also summarizes a (non-exhaustive!) selection of the startup opportunities we’re looking for at Variant.

Enjoy!

Find the full Crypto Trends Report here: https://docsend.com/v/sfvqr/cryptotrends

Subscribe now

In past cycles, major appreciation in the price of Bitcoin has preceded similar appreciation (if not outperformance) of many longer-tail cryptoassets. But over the past few years, we haven’t seen this pattern play out. Bitcoin dominance currently sits at 58% and has been in a steady uptrend since November 2022.

So, is this cycle going to skip alt season? Is it just that alt season isn’t here yet? Or maybe… is alt season already happening in a different market entirely, and just no one is looking?

My sense is that it’s the latter. The real alt season is happening in crypto equities.

What are the characteristics of an alt season setup?

• Price appreciation attracts new capital → question is, where is that new capital coming from?

• Price appreciation leads to a rotation of profits → question is, who’s taking profits and where is it being redeployed?

There’s certainly new capital looking to get exposure to crypto. But much of this is institutional, not retail. Retail tends to be fast adopters, while institutions are slower and often wait for external legitimization. Well, that’s happening now. The SEC approved the Bitcoin and Ethereum spot ETFs in 2024. Chairman Atkin recently announced Project Crypto. Nasdaq CEO Adena Friedman has advocated for the tokenization of equities. The list goes on.

The institutions are here with fresh capital. My guess is that most of that capital is being directed to crypto equities rather than crypto assets. Equities are familiar and accessible. Institutions already have operational setups (custody, compliance processes, dealer relationships, and more) are already in place, whereas purchasing crypto assets may require brand new capabilities. And buying equities falls within their mandates – vs. direct crypto tokens (much less long-tail alts) may simply be out of scope.

As a result, institutions are plowing money into crypto (or crypto-adjacent) equities. Coinbase is up 53% YTD. Robinhood is up 299%. Galaxy is up 100%. Circle is up 368% since its IPO in June (75% if you use the end-of-day close on its first trading day). Compare that to Bitcoin (up 31%), Ethereum (35%), or Solana (21%) and crypto equities’ outperformance becomes clear.

A similar story holds true if we look at performance since Bitcoin’s bottom on 12/17/22:

There’s good reason to believe this trend will continue. We have a slew of crypto equity IPOs on deck, with many more later stage companies likely to file in the years ahead.

Like a typical alt season, not every asset will perform well. And I expect there to be some rotation, as traders take profits from one expensively priced asset (e.g. CRCL trading at 26x P/S) and redeploy that capital into other assets.

In crypto, we often get different metas: the market might move from DeFi assets to gaming tokens to AI coins. Equities are likely no different. An alt season within crypto equities might see rotation from stablecoin stocks to exchange stocks to digital asset treasuries (or another trend).

There are a number of other reasons I believe that the crypto equities alt season may end up looking more like a historical alt season than any future alt season we get in cryptonative markets:

• Asset concentration. There are only a handful of equities that provide exposure to crypto. This is similar to how, in past crypto cycles, there were arguably under 100 tokens that buyers found attractive. And notably, it’s a very different dynamic than the crypto-native market today – which has millions of tokens and, as a result, is experiencing broader dispersion among deployment.

• Access to leverage. Last cycle, a number of crypto-native lending desks collapsed. We haven’t seen many rebuild. Equity allocators do have access to leverage, though, which means the booms can get bigger (and the busts can really bust).

We probably will have another alt season in crypto-native assets. But it will take time as the new marginal sources of capital gradually set up operational capabilities that enable them to deploy into cryptoassets.

So for now, it may not be the alt season many expected – but we are in an alt season nonetheless.

Thank you to Hootie Rashidifard and Mason Nystrom for feedback on this essay.

This comes just a few months after Robinhood’s Lost City of Gold event in San Francisco, where the company rolled out products like on-demand cash delivery, an AI investment advisor, private banking, and the ability to see your net worth on one screen.

In December, I wrote about what was happening at Robinhood. We now have about half a year of additional game film. Given the recent product announcements, I figured it was a good time to revisit the original analysis and provide an updated perspective on where I think Robinhood may be headed.

Robinhood’s Mission

Six months ago, I wrote that Robinhood’s stated mission was to serve the active trader. Their mission now actually seems much broader: to be the financial home for the next generation. This still includes servicing traders, but it also involves providing security for long-term investors (across every asset class), offering banking rails, and extending credit.

My “aha” moment was Robinhood’s “net worth” demo – a screen that calculates your total assets on / connected to Robinhood and can update by the second. It’s addictive and entertaining. And for it to showcase your entire net worth, it requires that Robinhood be the home for all of your assets (not just those you’re actively trading). Understanding Robinhood through this lens – rather than that of just serving the active trader – can help put their newer product initiatives into perspective.

Subscribe now

New Crypto Initiatives

The crypto announcements yesterday were exciting. Highlights include:

• Staking on cryptoassets

• Tokenized stocks, available to European citizens

• Tokenized equity in private companies, with the first two being OpenAI and SpaceX

• Robinhood Chain – an Ethereum-based layer 2, similar to Coinbase’s Base chain

• Perpetual futures (for Europeans)

One clear message is that Robinhood views crypto not just as a new asset class, but as a technology for expanding access and distribution. Importantly, the product announcements serve both missions mentioned earlier. Perps are oriented toward the active trader. Tokenized stocks and tokenized private equity appeal to both traders and long-term investors. And Robinhood chain lays the foundation for the company to build a suite of even more composable financial products.

Another message seems to be: anything Coinbase can do, Robinhood can do too. Coinbase has staking? Robinhood does too. Coinbase has a chain? Now so does Robinhood. On the consumer fronts where Coinbase still leads – like the ability to borrow against your BTC using Morpho – it seems likely that Robinhood will soon follow suit.

Staking was the lowest hanging fruit. It creates a stickier crypto asset base and helps Robinhood better monetize each dollar that comes in for crypto (previously, crypto was the only category in which Robinhood had only one path to monetizing a dollar – i.e. just via trading fees). It’s also probably the easiest to model from a near-term cash flow perspective.

Tokenized equities, perpetual futures, and tokenized private companies are more challenging. I say this mainly because these products are only being offered to Robinhood’s European userbase and it’s not clear to me how many of Robinhood’s 25.9M accounts are European vs. American. Some of these – like perps and tokenized private company stock – are products that will be attractive to roll out in the US, once there’s greater regulatory clarity. But others, like tokenized equities, are more of a question mark.

The Bigger Picture

Robinhood has gone from strength to strength. It transformed from being the ridicule of the 2021 GameStop / memestock mania to a widely trusted, deeply respected financial institution in just four years. What a turnaround.

The company is clearly investing in long-term-oriented growth initiatives. Approximately 77M Americans are aged 18-35. That’s 23% of the population – and the number is growing. Robinhood wants to be their brokerage and bank.

Other recent initiatives support this perspective. Robinhood has invested heavily in acquisition campaigns to bring both more users and more assets onto the platform. It recently offered a 2% net asset match for new stock deposits. A similar intiative is live now for crypto assets. It’s waving management fees, offering sweepstakes, and boosting APYs.

Remember, every dollar that comes into Robinhood can be monetized in multiple ways. Each net new asset deposited creates an amplification effect on the company’s top line. And more users ultimately means more assets, especially as younger generations accumulate wealth. Winning the wallet for the recipients of a(n eventual) generational wealth transfer is an opportunity that’s hard to understate.

Subscribe now

A Splash of Cold Water

Hopefully it’s clear that I’m excited by Robinhood’s future. That being said, I do occasionally take off the rose colored glasses. Many of these initiatives will probably take time to generate meaningful returns.

Robinhood ultimately has two levers for growth: increase ARPU or grow the number of users. The majority of revenue still comes from the active trader segment (what I’m proxying as users trading derivatives and crypto). I think the short-term effect of Robinhood’s recent initiatives will boost (active trader) ARPU, while the longer term goal is to increase the overall number of users.

As of its Q1 earnings, Robinhood generated $145 quarterly per every user on its platform. Increase ARPU by 15% – from $145 to $167 quarterly, or $580 to $667 annually – and it represents a $2.2B boost to annual top line revenue (assuming 25.9M funded accounts). Calculating these numbers in real-time, it doesn’t actually feel like I took off the rose glasses. The caveat, though, is that most of the new products that should meaningfully boost ARPU are only available to Europeans. If only 20% of Robinhood’s customer base is European, the actual top line growth would be closer to $450M. These numbers are illustrative but do directionally show the type of impact these products could have to Robinhood’s revenue growth.

Robinhood’s greater lever comes from increasing the number of users on the platform. Each 1 million customers added generates an additional $580M in revenue (using Q1 ARPU numbers, which was near an all-time high). And remember, Robinhood has 25.9 million accounts out of a potential 77 million in its strike zone. There’s significant room to run. But that will take time, especially if people already have sticky relationships with other brokerages or banks. My 2c is that it seems like the current market is pricing this as nearer-term growth than what may be realistic.

Long-term, I continue to believe that Robinhood is a generational company. Whether it’s able to fully execute on these plans in the next 6 months or 6 years is more of an open inquiry, but one that I’m excited to track nonetheless.

These platforms give consumers the ability to create apps by simply prompting LLMs with natural language. The result? Everyone is now a developer and software is a new form of content.

Today, these platforms are used primarily as tools. Consumers generate software for their own amusement and maybe post those vibe-coded apps publicly or share them with friends. The prompt-based apps aren’t good enough to top the app store (yet). But they are definitely good enough to have fun.

That single-player behavior acts as an important top-of-funnel for potential network effects. The more vibe-coded apps built and shared on the platform, the more that platform becomes a source of distribution and discovery. Come for the tool, stay for the network.

Let’s assume this hypothesis is indeed true. It provokes a bunch of interesting downstream questions:

• What does monetization for these platforms look like? If they are new social media networks, the logical conclusion would be that software generation should be free. Charging for credits would be analogous to charging people to post on Instagram; it’s antithetical in that it constrains engagement (with payment) whereas social media networks strive to optimize engagement.

• Ads are a tried and true mechanism for monetizing attention. What would advertisements look like on vibe coding platforms? This is really two questions: what is the form-factor of ads, and how are ads distributed?

• Are there better mechanisms for enabling content generators to earn? Everyone wants to get paid more for their content. Many want to be able to invest in the growth of their favorite creators, especially when they spot promising talent early. New platforms give us the ability to rethink economics (and utilize more modern payment rails) in really interesting and exciting ways.

• Are traditional feeds the best display for discovery and distribution? Feeds are great for relatively passive content. If apps require us to click in, what is the optimal format for discovery?

• How might these platforms cultivate network effects? Discovery and distribution of apps is one answer. The ability to remix pieces of content may be another. What else?

New social media networks emerge when we remove barriers to high-skilled creativity. Instagram made everyone a photographer. Twitter made everyone a short-form blogger. TikTok made everyone a video editor. Vibe coding platforms make everyone an app developer. Software is the new media, and vibe coding platforms are new social networks.

***********

Thank you to Charlie Blackstock, Bobby Blackstock, and Brexton Pham for conversations that indirectly influenced ideas in this post.

Two of my long-held investing principles are:

1. Invest in things that make the world more productive

2. Scarce resources accrue value

The second is particularly relevant in the post-DeepSeek world. Foundation models used to be a bottleneck: training was expensive so only a few large, well-capitalized players could afford to develop them. We’re seeing that change in real-time – from launches like DeepSeek, but also from advancements in distributed / parallelized training.

In that world, two of the scarcer resources are data and attention. Data is a well-discussed bottleneck in building LLMs. AI experts far more knowledgeable than me have discussed it here, here, and here.

Attention is scarce because humans have a limited number of hours in the day. Netflix famously expanded its addressable market by “competing with sleep.” But even competing with sleep can only go so far. Winning attention is valuable because consumers are some of the least margin-sensitive, most irrational customers.

Almost ironically, attention is the top-of-funnel for creating new data. There’s one company that sits at the intersection of both: Reddit.

Here’s why I think Reddit is a sleeping giant in a world dominated by AI.

Reddit Today

Reddit is one of the largest sites in the world. It consistently ranks in the top 10 most-trafficked websites in the world. The company’s app boasts over 120M installs.

Impressively, user numbers continue to grow. Company filings show DAUs have increased at an average rate of 35% YoY for the past couple years – with DAUs even surpassing 100M this past quarter! To put these user numbers in context, there are only about 30 companies with 100M monthly active users.

Those unique eyeballs are top of funnel for the content created and gathered on the platform. Reddit doesn’t directly report stats posting, but they did share that their Ask-Me-Anything feature grew 5x and conversion rates on ads improved by over 50%. This matters because passive eyeballs don’t create new data; active posters & clickers do.

Subscribe now

Embracing AI

Reddit appears to be barely scratching the surface with AI. That’s likely to change.

Here’s CEO Steve Huffman’s recent commentary on how Reddit is thinking about their advantage in producing data and incorporating machine learning:

Okay. I'll take it. Look, we're in the market, so there's basically kind of, call it, a barbell of customers. So we've got the big customers on the training side where we're still in conversations with folks, but there's a limited number of deals there. And then there's more customers. That would be smaller deals. We've done a number of deals there that are less training and more real-time access to information on Reddit, social listening, financial services. Those would be the customers we're targeting there.

Look, I think it's still early days for us here. We've built the kind of enterprise level kind of APIs to support this business. I'd say for us, it's a nice-to-have business, and we'll keep those conversations going. But if we're thinking about the future of Reddit, it's really our ads business, it's search, it's more of our kind of on-platform initiatives. But I think the story here is Reddit's data is extremely valuable. It's extremely valuable for a lot of players in the market, including to ourselves. And as I said in my opening remarks, Internet consumers broadly are trying to find Reddit's content for various reasons.

And so look, we're open and we're open for business. And I think finding the right balance between making this information accessible and building our own products on top of it is how we think about this.

When it comes to technological change, companies can be bucketed as either drivers (those pushing forward the frontier) or riders (those that can incorporate the new technology in service of their same end mission). For AI, Reddit is both.

Reddit as a Driver

The mass amounts of data produced by Reddit are how the company can help drive AI: the data is valuable as an input to the training and fine-tuning of LLMs.

Until last week, it appeared that there might only be a few large customers in the market interested in leveraging Reddit’s data for training. DeepSeek may have changed that: if building foundation models becomes cheaper (and thus more accessible), the number of customers interested in data licensing agreements could balloon.

The other category of “data customers” are those interested in real-time access to Reddit’s data. For these customers, what matters is the frequency and amount of new data Reddit produces.

Historical data is useful for creating a baseline level of knowledge and facts within a model. It’s an important foundation. Real-time data helps inform models (and the applications of those models, e.g. agents) on what to do next.

Reddit is clearly leaning into products and features that promote the creation of real-time data – from exploring sports deals, to supporting more official government accounts, to doubling down on platform conversation components. These three categories (sports, news, and live entertainment) are ones where information ossifies quickly, so much of the value comes from being the most up to date. I expect we’ll see Reddit continue to build products that promote the creation and sharing of real-time information.

Reddit as a Rider

Reddit can also leverage LLMs to improve its core business – in effect, “riding” the technological wave.

We’re seeing some of this already:

• Reddit acquired Memorable AI, a company that utilizes generative AI to enhance ad campaigns

• Recent filings highlight significant improvements in ad conversion rates

• The company is leveraging machine learning to translate a greater amount of content on the platform

But that’s just the start. One playbook could be to follow what X did with Grok: Reddit could build its own, custom LLM trained on proprietary platform data. Or, taking inspiration from Midjourney’s Discord integration, Reddit could embed generative AI products directly into its feeds and use the existing communities and conversation features to supercharge the collaborative creation process. And yet another path could be to lean all the way into agents and create subreddits / tooling tailored for agents (and/or create a credentialing system for user-generated vs. agent-generated content).

The point is that there are many ways Reddit could ride the AI tsunami. As long as Reddit continues to own end user attention, its options for incorporating new technology should be near endless.

Subscribe now

Concluding Thoughts

Reddit certainly seems well-positioned for a world of exponentially more AI.

But the undiscussed elephant in the room is Reddit’s valuation. As of this writing, its market cap is $35B – a 31x P/S multiple. The company only just reported its first profitable quarter (a significant amount of which was driven by well-timed cryptocurrency sales). It definitely doesn’t look cheap.

“I find far greater intellectual satisfaction in trying to understand why a company might be undervalued at 65 times forward earnings than seeking comfort in a seemingly low valuation.”

- Christopher Tsai, Tsai Capital 2024 Year-End Update

Looking at Reddit today feels like an exercise in trying to understand how rapidly a TAM can change and expand. On the one hand, the valuation using backward-looking metrics feels sky-high. But on the flipside, if it is true that Reddit is set to both drive and ride the explosive innovation we’re seeing from AI, then maybe the current business is only a fraction of what’s to come.

For example, licensing is still a relatively small portion of Reddit’s revenue. Throughout the past few quarters, licensing represented approximately 10% of company revenue. As the company continues to iterate on that business model – and as the universe of potential customers grows – it’s very possible that licensing will increase as a share of revenue mix. At the very least, I expect the its contributions to total top line (in absolute dollars) will grow materially.

Whether that’s enough to justify or even propel the valuation? I don’t know. But it’s still fun to reason about. And there’s a lot we’ll probably learn about data, licensing, and the economics of attention as the company continues to evolve.

On the whole, Reddit is one of the companies I’m most excited to keep an eye on as AI increasingly impacts our lives.

1. President Trump won the election. It also appears we will have the most pro-crypto Congress ever.

2. An AI agent called Truth Terminal started posting about a token called $GOAT.

3. Hyperliquid airdropped its token.

The following are some of the ways I think the impacts may play out.

The New Administration

The win for crypto seems to be quite broad sweeping. We saw this sentiment reflected in the price action immediately following November 5th.

It’s hard to know what the new administration and new Congress will actually do. My hope would be some mix of the following: more ETFs, greater clarity on how traditional financial institutions can embrace digital assets (via legislation, agency-level rulemaking, or a combination), stablecoin legislation, and maybe some paths for decentralized revenue-generating protocols to redistribute earnings.

There are multiple ways this may manifest in the startup market:

• Several known categories of businesses will be de-risked, and we may see increased competition as a result. Institutional staking is a good example. If we get staking in the ETFs, the business risk will shift from being market-based (i.e. how big is the market and how fast will it grow?) to product and team oriented.

• The number and quality of customers for crypto projects may grow, stemming from traditional finance being able to more easily embrace the industry. For instance, consider the downstream effects of financial institutions being able to custody digital assets. In the U.S., there are over 2000 large commercial banks and over 3000 brokerage firms. Any and all that embrace digital assets will need custody and security services. It would represent a sea change in the addressable market for both crypto protocols and services-oriented software businesses.

• All developers should benefit. In today’s world, figuring out what may or may not be compliant is resource-intensive; navigating the lack of regulatory frameworks consumes valuable time, headspace, and (often) capital. In some cases, this potential headache may discourage new developers from starting a project entirely. Regulatory clarity would bolster developer confidence, increase productivity, and even attract new entrepreneurs who find project inspiration from the guidelines / more legible market structure.

On the whole, we are entering what is likely to be the most friendly environment for crypto developers in the industry’s ~15 year history. Among builders and other investors I’ve talked to, the excitement is palpable. I can’t wait.

Subscribe now

Truth Terminal

The second of the major inflections occurred on October 20th with the following tweet:

For the full backstory on the Truth Terminal bot, I highly recommend this episode of the Ben & Marc Show.

Truth Terminal kicked off the current wave of interest and development around crypto-enabled AI agents and frameworks. To be honest, it does not yet seem like many agents are actually doing unique and interesting things with crypto. The most notable “action” is capital accumulation – in the form of trading and, more commonly, receiving airdrops from people trying to get the agent to post about their token(s).

As far as I can tell, there is (currently) little to no economic relationship between many of the agents and their tokens. (The one exception I can think of is Botto). It’s also not entirely clear how consumers can use these agents, save from buying their tokens. Contrast this to DeFi summer, where participants were actively using the protocols. The challenge is that it’s not clear exactly what it looks like to “use” an agent. My colleague Jesse wrote a great piece on what at least one compelling model could be:

With AI automating day-to-day objectives, the role of humans can be minimized to focus on providing capital, nudging guidance, long-term planning, and other resources the models do not have at their disposal.

Check out the full piece here.

One obvious next step for these agents is to start deploying that capital. That could be via trading, venture investing, providing liquidity to protocols, staking tokens to run validators, creating additional assets… the list goes on. Humans could help set the parameters or investment objectives.

Zooming out, there are a lot of factors that could continue to propel the crypto-enabled AI agent category. For one, the agents of today are fairly nascent. As they start performing more productive tasks and/or collaborating with each other, both excitement and fundamentals should grow. Second, some of the agent innovation is driven by forces external to crypto. The broader AI industry is full of talent dedicated to improving agents and the underlying models. As those researchers continue to expand the bounds of what agents can do, crypto developers can leverage that work as a foundation for then integrating crypto in unique and interesting formats.

Agents are just one part of the crypto x AI stack. Under the hood, these agents need to access compute, run inference, incorporate new training data, make payments, and more. Many teams have been working hard over the past year to build out performant infrastructure. Variant has invested in a number: Hyperbolic provides verifiable inference and compute, Sapien is scaling access to cost-efficient training data, and an unannounced project is working on decentralizing training. We’re actively investing in more.

Why is crypto-native infrastructure necessary for agent innovation? Agents should be able to sell shares of themselves, self-fund their operations and infrastructure needs, interact with other agents (and have verification mechanisms for instances where they’re deriving information from other agents, i.e. “agentic oracles”) – all without the risk of their operations being censored or shut down. Decentralized infrastructure plays a critical role in removing platform risk and giving these agents access to capital.

Hyperliquid Airdrop

The third inflection point was the Hyperliquid airdrop. It made a major splash in the industry. The project was entirely self-funded, so there were no non-team “insiders.” Nearly a third of the supply was airdropped to historical users of the project, creating a massive wealth generation event. And most importantly, the product itself is widely used and loved; it’s something users want to own a piece of.

As a result, Hyperliquid’s airdrop acted as a sort of lightning rod for illuminating other fundraising dynamics that have been emerging over the past half year. There are two themes the market has increasingly trended toward: 1) giving retail investors greater parity of access with institutional allocators, and 2) avoiding the adverse effects of unlocks (where there is a known date that a portion of the vesting insider tokens become liquid and dilute the existing circulating supply). Hyperliquid achieved both of these.

There had already been a decent amount of industry discussion bubbling about alternative funding and launch strategies. Hyperliquid brought a megaphone to the category. None of the following are apples-to-apples, but I think they fit well alongside a discussion of why Hyperliquid’s launch was so impactful. So with that caveat, some of the ways we’re seeing teams strive to achieve similar end goals include:

• Raise “community rounds” on platforms like Echo, where individual accredited investors are given the chance to invest in private fundraises. Such platforms enable a project’s community to put skin in the game while simultaneously expanding the set of capital options available to founders. These are becoming increasingly popular. Again, it’s not a direct comparison to the airdrop: private rounds often have vesting schedules while the Hyperliquid airdrop was liquid immediately. But on the whole, community rounds strive to achieve a similar end goal of giving community members valuations / access typically reserved for institutional allocators.

• Launch the token first and build the product second. The token then provides a source of capital for developing the product and attracting third-party contributors. Notably, it’s the opposite of what Hyperliquid did: build the product first, launch the token second. So again, it’s far from apples-to-apples – but it attempts to give early project users the chance to contribute, participate in the development, and speculate on the upside nonetheless.

• Reject unlocks altogether, and give investors the option to sell at token generation. Should early investors indeed choose to realize returns, the effect would likely be a temporary price decrease – in turn giving community members a chance to get in at attractive valuations. We’ve seen one or two of these; I think they’ll become increasingly common.

Why does this matter? The guiding theme seems to be that there are now more options than ever for projects to help their users become owners. Writing as an admittedly biased VC, I don’t think there’s any substitute for taking early-stage funding from venture investors – the legal help, strategic advice, and portfolio resources can be invaluable for going zero to one. Luckily, raising venture dollars is not mutually exclusive with leaning into some of these alternative fundraise / launch strategies. Thoughtfully embracing paths to giving communities true upside can and should be rewarded. But again, underpinning all of this – and the true lesson / reminder from Hyperliquid – is to build great products that users love and want to own :)

Subscribe now

Concluding Thoughts

The above considers major inflection points that occurred during the second half of 2024. It’s a bit different from my 2024 H1 reflection, in that the intent for the first was more focused on highlighting data-driven instances of what was working in crypto. Luckily, trying to replicate that here would take up many, many pages (there’s a lot that’s working!).

So instead, I’ll leave any readers who have made it this far with two of the most astonishing charts I’ve seen:

Crypto is becoming harder and harder to ignore.

And if you enjoyed this reflection, check out some of the other pieces I published over the last two quarters:

• What Credit Card Networks Can Teach Us About Stablecoin Opportunities: link

• What’s Happening at Coinbase? link

• What’s Happening at Robinhood? And Will It Become the Next Great Crypto Exchange? link

• We’re in the app phase: link

• DeFi super apps: link

• Distribution is all that matters: link

• Some ideas about lossless products: link

Subscribe now

Thank you to Jake Chervinsky, Hootie Rashidifard, Jesse Walden, Jack Gorman, and Derek Walkush for feedback on this piece.

Why might this be? And furthermore, is Robinhood actually well-positioned to become a leading regulated crypto exchange? Here’s my take.

Business Fundamentals

Robinhood’s stated mission is to serve the active trader. Few markets are more active than crypto: it has the greatest set of trading hours (24/7) and probably the fastest asset creation process.

Yet today, Robinhood supports trading in just 21 crypto assets. That’s fewer than what Coinbase offered in 2019! It seems reasonable that crypto would stand out as a bright spot for potential growth. This is now doubly true given the anticipated crypto friendliness of the new administration.

Robinhood’s distribution could provide a powerful lever for growing into this market. Approximately 24M customers have funded accounts on the platform. About half of those trade monthly. That’s several million more than Coinbase, which in recent quarters reported 7-8M monthly trading users.

Could Robinhood effectively activate its user base – e.g. get a third of their funded customers to trade crypto, or two thirds of their monthly active traders – they’d very quickly become one the largest crypto trading platforms (by users). Again, at face value, certainly seems like an attractive opportunity.

Subscribe now

Is Crypto Valuable for Robinhood to Pursue?

Robinhood’s business is structured such that each dollar deposited to the platform has the ability to impact multiple revenue lines. For instance, consider a user that deposits cash to buy a security. Robinhood makes money when the user both buys and sells that asset. They can also make money if a user opts to lend that security. In this way, Robinhood monetizes the entire lifecycle of the asset purchase.

In the example above, a dollar deposited to Robinhood impacts two revenue streams (equities transaction revenue and securities lending). Users may also take their share of revenue earned via securities lending and store it as cash (earning interest) or re-invest it in another asset. Both contribute to Robinhood’s top line.

Today, it’s not clear that crypto trading has the same amplification effect. As far as I can tell, a dollar deposited with the intent to trade crypto only generates revenue via transaction fees.

A first-order motivation for expanding the crypto business might simply be to get more dollars on the platform. Even just expanding the one revenue line is highly lucrative. In recent quarters, crypto trading drove nearly double the transaction revenue of equities – despite trading volumes being 10-20x smaller, and despite the small number of assets listed.

Looking toward the future, it stands to reason that if extending utilization of assets has historically been Robinhood’s bread & butter for growth, they’ll seek to do the same with crypto assets. There are clear playbooks for doing so (e.g. via staking those assets). Tenev has also become increasingly vocal about the potential of DeFi. I wouldn’t be surprised if we see Robinhood make dedicated efforts to move more and more onchain (especially now that best-in-class cybersecurity services are available to help protect users).

So, Will Robinhood be the Next Great Crypto Exchange?

Robinhood has the building blocks in place to become one of the largest venues for retail users to trade crypto. But the term “exchange” actually seems a misnomer. Robinhood is a platform that connects order flow to underlying exchange infrastructure, but (to my knowledge) it doesn’t actually operate the matching or settlement engines.

Robinhood’s announced acquisition of Bitstamp – a crypto exchange – suggests it may intend to move down the stack. But again, Robinhood’s bread & butter is a) distribution, and b) extending utilization of the assets on its platform. The atomic unit of growth is the funded account. Running an exchange is an entirely distinct business, with distinct operational capabilities, from the Robinhood of today.

Subscribe now

So in sum, Robinhood does seem primed to become one of the next great crypto trading platforms. Its existing distribution, track record of execution, and focus on serving the active trader all support this notion.

My guess is that the company will find the most success by leaning into consumer-facing crypto products rather than expending efforts to build down the stack. It’s my hope as well. In the broader crypto ecosystem, we are seeing innovation across all layers of the stack – from front-ends to orderbook infrastructure to institutional-grade staking products. Crypto is modular and composable by design. Should Robinhood choose to focus on the more consumer-facing elements of its crypto business, it could essentially plug and play the best of the underlying infrastructure products for servicing its active users. Whether that’s indeed the route they ultimately choose will, I think, signal a lot about the company’s potential success in crypto.

Irregardless of the path they choose, what is clear is that Robinhood is reinvesting in crypto in a big way. Overall, that seems poised to be a win-win-win – for the crypto ecosystem, for users, and for Robinhood.

This tends to be an excellent litmus test for screening whether a business can grow really big. And notably, it works well in both private and public markets.

What is Productivity?

Core to the question is the word “productivity.” I’m using the economic meaning of the word, where productivity refers to how much output can be produced with a given set of inputs. There are two ways to grow productivity:

1. Grow output while holding the number of inputs constant (or grow output at a faster rate than one increases inputs).

2. Reduce the number of inputs while maintaining the same level of output.

Both really mean the same thing: each input generates more output. But the framing slightly changes the characteristics one might look for. The former is more oriented toward new-market growth, whereas the latter puts the focus on cost declines. 

Applying this Framing

In practice, how does a company actually improve productivity? Keeping with the framing above, there are two heuristics that I’ve found especially valuable: 

1. For growing a market: the company (or the technology it utilizes) expands access to the good or service.

2. For reducing costs: the company has identified a cost center within an existing market and has found a way to improve upon it (e.g. via less expensive substitutes, better technology, etc.).  

These considerations don’t exist in isolation. Sometimes cost reductions make the market available to new participants who were previously priced out. Similarly, market growth can incentivize R&D into new technologies or techniques that, in turn, help drive down costs. Companies that build such flywheels often enjoy compounding competitive advantages.

There are many scenarios where this framework isn’t quite as cut-and-dry as it may sound. Answers often differ based on a function of one’s time horizon. Take re-launchable rocket companies. It took a lot of inputs a long time to generate any output. But if we get reusable rockets and can start to do things like space manufacturing, the size of the resulting market may be huge. 

Successful examples are almost always easier to rationalize in hindsight. That’s why I find the framing if it works far more helpful than will it work – at least for earlier-stage, venture-style investments. 

Subscribe now

Later-Stage / Public Markets Addendum

There are a couple variations of this question applicable in later stage (i.e. more mature) markets.

The initial question – will this meaningfully increase productivity? – points towards growth. But there are some companies that have already achieved this status. So the better question to ask might be If this company disappeared, would productivity fall? A “yes” answer to this question suggests not only that the business provides a valuable good or service (i.e. validation that it has indeed achieved “productive” status), but also that its market position is likely well-solidified. 

Early-stage companies tend to be hyper-oriented around growth – which frequently includes a plan for disrupting incumbents. The flipside is that mature, market leaders need to defend their business. There are many ways that a company can defend (or, in the case of regulatory moats, enshrine) its position. I’ve found this focus on productivity to be a good starting point: it’s both a litmus test for defensibility and a structured gateway to more market-specific analysis.  

So when Coinbase recently announced that Jesse Pollak – the internal leader for Base, Coinbase’s layer-2 blockchain – had been promoted to also lead Coinbase Wallet and join the Coinbase executive team, it struck me as meaningful. 

The broader picture suggests Coinbase may be reorganizing around a newly defined set of internal divisions:

1. Incumbent business lines – specifically, Coinbase’s exchange and institutional services. 

2. Onchain business opportunities, spearheaded by Jesse Pollak. 

3. Payments & commerce, enabled by the Coinbase Developer Platform. 

All three are growth opportunities for Coinbase, but the first is really their bread & butter while the latter two are more emergent. If the move is successful, it feels like the type of moment we’ll hear Brian Armstrong mention in talks a few years from now – similar to how Brian Chesky talks about Airbnb’s organizational restructuring during Covid.  

Here’s how I’m thinking about the different segments that encompass Coinbase.

Segment #1: Incumbent Products

At the core of Coinbase’s incumbent revenue streams lives its exchange product. The exchange services both retail and institutional customers. Historically, it’s also provided a strong platform on which Coinbase has built its product offerings. As users began to store more assets on the platform, Coinbase expanded into adjacent services that capitalize on those assets: blockchain rewards (staking), institutional custody, and stablecoin revenue.

These business lines are still growing. Revenue from transactions (read: trading) was over 2x higher in Q2 2024 than in Q2 2023. The same was true of both custodial fee revenue and blockchain rewards. 

Stablecoin revenue comes via cash deposits. Coinbase acts as a distributor for Circle’s USDC stablecoin, meaning that it’s one of several partners that can convert fiat into USDC (an individual can’t just go to Circle and mint USDC). For Coinbase, funds come in the form of fiat deposits to the exchange. Once a customer has US dollars in their account, there’s zero cost to swap them into USDC to start earning yield. Circle monetizes the fiat deposits via investing in short term treasuries and shares that yield with their distributors. For Coinbase (a distribution partner), this added over $240M to its top line in Q2 2024. 

From a scan of the executive page, it seems like Gregory Tusar and Max Branzburg are probably in charge of the institutional and consumer product lines, respectively. Tusar has been at Coinbase since 2020 (and has been working on institutional blockchain products since 2018). Branzburg has been at Coinbase since 2018. Both are industry vets. 

Segment #2: Onchain Opportunities

Putting Base and Coinbase Wallet under the same leadership shows Coinbase is taking a more integrated approach to building onchain. It’s a move that I feel may be underappreciated.

Base’s mission is to bring more users onchain. Wallets are the port of entry for those users. Yet many wallets today (including Coinbase Wallet) still present users with significant friction, both when onboarding and trying to transact onchain. There’s too many clicks. Discovery is subpar. The list goes on.

Similarly, application developers are often acutely aware of where users drop off in the conversion funnel – whether it’s onboarding funds into the actual wallet, having to bridge funds from one chain to another, signing the message to approve the transaction, or something else. It’s possible and even likely that many of the app developers have ideas about how to design a better flow, but have previously lacked the time or connections to get those thoughts to the wallet builders.

Unifying the Base and Wallet teams under the same leadership should create a tighter feedback loop. It facilitates smoother information sharing from both sets of stakeholders (users and developers) and will be well-positioned to build a more integrated product, improving the experience for all parties.

My guess is that, as a result, Coinbase will become even more vocal about what users can do onchain. The wallet product improving makes it easier to showcase it in marketing materials; a one-click experience is much more appealing to consumers than a five-click one. More eyeballs on the wallet also makes it a meaningful discovery point for applications and assets within the Base ecosystem. Altogether, this reorganization creates a stronger foundation for Coinbase to grow and experiment with its onchain businesses.

Subscribe now

Segment #3: Payments & Commerce

When I look at CDP, I see a suite of services for everything one would need to run a business entirely on crypto rails. Products include embedded wallet SDKs, Payment APIs, blockchain data, and commerce tools. The sum total is a package that enables merchants to easily accept payment in any token, custody and monetize those deposits, and off-ramp securely into fiat.

CDP also offers tools beyond pure payments infrastructure. This includes trading bots, exchange access, and connected credentials. It looks a lot like Coinbase asked “what infrastructure and access would AI agents need to engage in commerce and interact (autonomously) onchain?” and built a product suite around those needs. In other words, the aspiration seems to be Stripe for AI agents. 

Coinbase is Still One Company

In reality, none of these divisions are cut-and-dry. The lines very clearly blur. For instance, should CDP-enabled trading bots that transact primarily on Base be considered part of segment #2 or #3? Similarly, if a merchant accepts USDC on Base as its primary source of payment, is that part of segment #1, #2, or #3?

The answer is that it doesn’t actually matter. Coinbase is one company, not three. All of the divisions likely talk to each other (and Pollak’s promotion to the executive team further supports the notion that Coinbase strives for strong internal communication between these businesses). I find the segment categorization useful primarily for thinking about where unsuspecting growth opportunities may lay for Coinbase. The incumbent business is well-studied and well-analyzed. The latter two – onchain opportunities and payments & commerce – feel more like high-slope startups within Coinbase and I’m excited to see them grow.

Today, global stablecoin supplies exceed $150B. There are five stablecoins that have at least $1B in circulation: USDT (Tether), USDC (Circle), DAI (Maker), First Digital USD (Binance), and PYUSD (PayPal). I believe we’re headed toward a world with many more – a world in which every financial institution will offer its own stablecoin. 

I’ve been trying to reason about the opportunities that will emerge as a result of this growth. Where I’ve landed is that looking toward the maturation of other payment systems – specifically, credit card networks – may hold some of the answers. 

Subscribe now

How Similar Are Credit Card and Stablecoin Networks?

To consumers and merchants, all stablecoins should simply feel like a dollar. But in reality, each stablecoin issuer treats a dollar differently – stemming from distinct issuance & redemption processes, the reserves backing each stablecoin supply, varying regulatory regimes, the frequency of financial audits, and more. Reconciling these complexities will be big business. 

We’ve seen a version of this before with credit cards. Consumers spend using almost-but-not-quite fungible assets that proxy the dollar (they’re loans on dollars, but those loans aren’t fungible because people have varying credit scores). There are networks – like Visa and Mastercard – that handle orchestration of payments throughout the system. And the stakeholders in both systems are (or will likely end up) looking similar: the consumer, the consumer’s bank, the merchant’s bank, and the merchant.  

An example might help unpack the similarities in network structure. 

Let’s say you go out to dinner and pay the bill using your credit card. How does your payment make it into the restaurant’s account? 

1. Your bank (the credit card issuer) will authorize the transaction and send funds to the restaurant’s bank (called the acquirer).

2. An interchange network – such as Visa or Mastercard – facilitates this exchange of funds and takes a small fee.

3. The acquirer then deposits the funds into the restaurant’s account, minus a fee.

Now say you instead want to pay using stablecoins. Your bank, Bank A, issues the AUSD stablecoin. The restaurant’s bank, Bank F, uses FUSD. These are two different stablecoins, even though both represent dollars. The restaurant’s bank will only accept FUSD. How does the AUSD payment convert into FUSD?1 

It should end up looking pretty similar to the flow through credit card networks:

1. The consumer’s bank (which issues AUSD) authorizes the transaction.2 

2. An orchestration service facilitates the exchange from AUSD to FUSD and likely takes a small fee. This exchange could happen in at least a few ways: 

   1. Path 1: utilize stablecoin-to-stablecoin swaps on a decentralized exchange. For instance, Uniswap offers a number of such pools with fees as low as 0.01%.3

   2. Path 2: redeem AUSD for the dollar deposit, and then deposit that dollar with the acquirer to issue FUSD. 

   3. Path 3: orchestration services could net flows through the network against each other; this would likely only come with scale. 

3. FUSD is deposited into the merchant’s account, maybe minus a fee.

Where the Analogy Starts to Diverge

The above draws what I see as clear parallels between credit card and stablecoin networks. It also provides a useful framing for thinking about where stablecoins start to meaningfully upgrade and outperform elements of card networks.

The first is in cross-border transactions. If the scenario above had instead been an American consumer paying at a restaurant in Italy – where the consumer wanted to pay in US dollars, and the merchant wanted to receive euros – existing credit cards would charge upwards of 3%. That same conversion between stablecoins on a DEX could be as little as 0.05% (a 60x difference). Apply that magnitude of fee reduction to cross-border payments broadly and it becomes clear just how much productivity stablecoins could add to global GDP.

The second is in payment flows from businesses to individuals. The time between when the payment is authorized and when the funds actually leave the payor’s account is fast: as soon as the funds are authorized, they can leave the account. Instant settlement is both valuable and desired. Additionally, many businesses operate with global workforces. The frequency and size of cross-border payments may be much higher and larger than that of the average individual consumer. Workforces continuing to globalize should provide a strong tailwind for this opportunity. 

Subscribe now

Thinking Ahead: Where Might There Be Opportunity?

If the comparison between network structures is even directionally true, it helps shed some light on where there could be startup opportunity. Within the card ecosystem, major businesses have emerged from orchestration, issuance innovations, form factor enablement, and more. The same could be true for stablecoins.

The earlier example primarily describes the role of orchestration. That’s because moving money is big business. Visa, Mastercard, American Express, and Discover are all worth at least tens of billions. In aggregate, they constitute over $1T in value. That multiple card networks exist in equilibrium suggests that competition is healthy and the market is large enough to support major businesses. It seems reasonable to believe that similar competition for stablecoin orchestration will exist in a mature market. We’re really only 1-2 years into having sufficient underlying infrastructure for stablecoins to succeed at scale. There is still plenty of time for new startups to pursue this opportunity. 

Issuance is another area for innovation. Akin to how business charge cards have grown in popularity, we may see a similar trend of corporations wanting their own white-labeled stablecoins. Owning the unit of spend provides greater control over end-to-end accounting, from expense management to dealing with foreign taxes. It’s possible that these become direct business lines for stablecoin orchestration networks, but they could also be opportunities for entirely new startups (e.g. similar to Lithic). The byproducts of that corporate demand could lead to even more new businesses. 

There are also a lot of ways issuance can become increasingly specialized. Consider the emergence of tiering. With many credit cards, customers can pay an upfront fee for better reward structures. Think: the Chase Sapphire Reserve or the AmEx Gold. Some companies (often airlines and retailers) even offer their own specialized cards. I wouldn’t be surprised if similar experimentation in stablecoin reward tiering follows suit.4 This, too, might be an inroad for a startup.

In many ways, all of these trends feed each others’ growth. As more variation in issuance emerges, demand for orchestration services increases. As orchestration networks mature, they’ll lower the barrier for new issuers to compete. All of these are massive opportunities and I’d love to see more startups in the space. Long-term, these will be trillion-dollar markets that should support many large businesses.

Thank you to Daniel Barabander for comments and feedback on this piece.

1. Incumbent applications are becoming more siloed and self-contained.

2. New startups are trying to make the internet more open and interoperable.

I’d summarize the tension as: incumbents are trying to strengthen their margins and moats, while new startups see that margin as opportunity. 

The core premise of a more interoperable internet is that one application should be able to read a user’s data from another application. Think: SeatGeek giving you priority bidding for season tickets based on your DraftKings betting history, Resy recommending restaurants based on the type of food in your most recent Amazon Fresh order, or Fila giving unique discount codes to folks who watched Emma Navarro’s awesome US Open run and consequently became interested in purchasing gear from the sponsor. 

I’ve seen two categories of startups going after this interoperable internet opportunity. 

The first approach meets the internet where it is today, in that it embraces the fact that the vast majority of existing user data lives in closed applications. Its foundation is a technology called web proofs: cryptographically signed data from a web server that proves a user had taken a certain action online. Web proofs allow users to choose how their data is leveraged. Putting this in context of the examples mentioned earlier, a user could choose to prove to SeatGeek that they are active DraftKings users. Consumers could choose to let Resy access details about their most recent Amazon Fresh orders, but they don’t have to. (Check out this piece if you’re interested in learning more about the legal considerations).

The second approach is an attempt to rebuild core categories of internet applications with persistently and enforceably open data. A core distinction from the first approach is that no one has control over the data – so instead of a third party developer needing to ask users to verify that they’ve taken a certain action (i.e. in the form of a user providing a web proof), developers can simply verify the action themselves. Farcaster is one of the main companies building in this space. It’s a new social network, where everything from user social graphs, to posts, to likes are stored in a distributed set of hubs. It’s become a flourishing developer hub and we’re starting to see some cool applications leverage its network.

I’m excited about both approaches. And when considered in tandem with growing adoption of embedded crypto wallets, we can start to envision how new economic flows might emerge.

On the former: we could see microeconomies and new information marketplaces emerge on the back of this enriched access. To date, one of the challenges with the notion of users selling their data is each piece of data isn’t actually worth that much. Micropayments might be one way that consumers can start to monetize their data. But the reverse might happen, too. I could envision a world in which consumers are willing to pay for priority access to certain features (e.g. better slots in ticketing queues), and that the process for granting that exclusive access involves a web proof.

On the latter: advertisements underpin a significant amount of economic flows on the internet today. What if that market could be made more efficient, by better matching consumers with products they want, and ensuring dollars spent by advertisers actually convert into dollars spent by consumers? I wrote this in an earlier reflection on the state of crypto:

• A company could send coupons directly into their target customers’ wallets (because every account has an associated wallet). 

• The coupons might be based on some mention of a similar product in a post the consumer created, or a post that the user liked.

• The business can operate with the confidence that data will remain open and accessible forever (i.e. no concern about APIs being shut off or prices being jacked up), which allows it to then invest in refining the effectiveness of this distribution channel. 

• The budget is only spent if the consumer converts to a sale (i.e. the coupon is used).

Open data graphs are a core step toward this vision. 

On the whole, both seem like very promising approaches toward a similar end vision: a more interoperable internet. That they take very different approaches to how the data is accessed, what data is available, and whether to meet users where they’re at or pull them into the future seems like a net good thing. The more experimentation on the internet, the better.

If you’re a developer reading this and interested in learning more, some good resources I’ve seen for the web proofs approach are on a company called Pluto’s website. For the latter, download Farcaster and start building cool stuff today.

The view was breathtaking. Seeing the skyline lit up at night — and driving toward it knowing that what awaited at the end of the bridge was a bustling city full of people enjoying late dinners, pedestrians walking home, and bikers racing by — made me feel both inspired and grateful. Inspired by the energy. Grateful that I got a peek at this version of New York’s 10pm city.

It also made me think: ride sharing apps should offer “scenic route” options. I’d easily pay a few extra dollars to route through prettier parts of a city, especially when I’m traveling somewhere new. I imagine other folks would, too.

Scenic routes would expand the set of premium options available to consumers. Today, riders can pay more for luxury cars and/or (in some apps) faster pickup times. Luxury services (e.g. Uber Black) usually go for a meaningful markup to regular ride share prices, in part because the operating vehicle is more expensive to purchase and maintain. Faster pickup times have less of a premium assigned, likely because the incremental costs involved are lower.

Scenic routes would fall closer to the latter cost structure. These rides would presumably be longer than a standard ride, so the core incremental costs should be driver time and extra gasoline (and maybe car depreciation, if one wants to get really specific). Uber and Lyft also allocate a portion of each ride’s payment towards a pool of incentives that helps keep drivers on the road longer and more consistently. Lengthier rides might make it harder for drivers to complete the incentivized challenges (e.g. complete 7 rides for an extra payout), so it seems reasonable that a portion of scenic routes’ premiums would go toward driver compensation for the opportunity cost of the longer rides. My guess is that all of the above would only add a few extra dollars to the average ride.

All in all, scenic routes strike me as a desirable feature that could be implemented without breaking the bank. For now, I’m resigned to occasionally asking the driver for a scenic route and promising to give them a nice tip. But hopefully in the future, it’ll just be an option available within ride share apps.