Incumbents struggle to sell to startups because they're bound by the rules of P&L.
I think taking a more crypto-native approach is actually better, regardless of whether you want to offer the product to a crypto-native audience or to a more traditional audience.
Where you see autonomous labs and autonomous science being adopted first is probably more of a function of the market that it's operating in.
If you attract all of the new companies at Formation and then grow with them as your customers become big companies in their own right, so will you.
We'll see you next time.
How autonomy starts entering scientific research through labs and how distribution itself becomes a strategy when startups sell to other startups at formation.
We'll start with the rail that's already visible, stablecoins going mainstream.
Guy Willett argues that stablecoins are only the beginning.
The next phase is on-chain credit origination and new synthetic products that are easier to scale than simply copying traditional assets onto a blockchain.
Here's Guy.
My name is Guy Ouellette, and I'm a general partner on the crypto team here at A16Z.
This year, my big ideas are origination and purplification.
In 2025.
We've seen stablecoins go mainstream, with increasing outstanding issuance and payment volume.
And one of the things I've been thinking about is what the most important second order effects of outstanding stablecoin issuance will be.
I think a lot of the existing stablecoins look effectively like narrow banks today, where they hold user deposits in fiat or perhaps in treasury bills.
I think it's very unlikely in the long term that we scale on chain finance exclusively through narrow banks.
And so I've been thinking much more about how we can facilitate credit and capital formation on chain.
There will be a new role or entity that's very important which is something akin to a private credit fund helping to facilitate loans on-chain.
If you look at the way things are maturing on-chain for crypto, you see a very similar market structure to what's happening in the traditional financial world after the great financial crisis, in part because of Basel III.
There's a lot more non-bank lending from entities like large private credit funds that have significant equity capital.
So banks are doing much more lending to credit funds for making loans to end depositors.
This is very similar today in crypto to how I think stablecoins will effectively end up lending to curators or to asset managers, who will end up making loans to end users.
And one of the things I'm spending a lot of time considering is how we move credit origination on-chain.
Instead of originating a series of loans off-chain a credit card receivable, for example, then tokenizing that, potentially securitizing it.
And moving that on-chain is a copy of an off-chain asset, how we originate credit natively on-chain.
And I think this is important specifically because it can drastically reduce back office costs, like loan servicing, which in many cases can take 1 to 3 of the outstanding credit facility itself every year.
So it's incredibly expensive.
But I also think doing on-chain origination will allow for much more composability between different DeFi projects.
A very similar vein.
When we think about how to move traditional assets on-chain, lots of people focus on tokenizing those assets, meaning creating an on-chain record or copy of the existing asset that exists off-chain.
We've seen many more perpetual futures projects perpify or create a perp related to an off-chain price feed or an off-chain asset.
So instead of tokenizing something like an equity and putting it on chain, you could create a perpetual future for that equity.
This, I think, is interesting in the U.S. and in Europe and in robust capital markets.
But I think it's particularly interesting for emerging market equities things like the Indian equities market, for example, where existing derivatives, like zero day options, often trade more notional volume than the underlying spot.
So I think there's already pretty good product market fit for derivatives on these underlying assets.
And we could see a lot of those move on chain and in the coming years.
Stablecoins have gone mainstream today, but I would argue we need more than simple tokenization of fiat dollars or fiat currencies.
And I think there are a couple of reasons for this.
The first would be I think we need a better way to facilitate credit on-chain than minting stablecoins and then using those stablecoins for loans.
I think in many cases stablecoins look like narrow banks and we need some way to do credit on-chain.
And so I think there's an interesting opportunity for Curators that today operate on something like Morpho or private credit funds someone like Apollo that has put Acred on-chain take more of an active role in helping to manage existing stablecoin collateral.
An interesting form of this is there are lots of Call them synthetic dollars, because they're not, strictly speaking, stable coins.
Today,
But that means a dollar representation backed by off-chain traditional assets, where the normal token itself is fully collateralized by fiat dollars off-chain.
But the staked representation of it then would be collateralized by higher risk and higher yield credit assets.
And Athena popularized this idea.
Initially a synthetic dollar that is backed by a cash and carry trade, a basis trade for perpetual futures, but we're starting to see this proliferate into other asset classes and into other structures things like currency, cash and carry trades, synthetic dollars that are backed by physical infrastructure akin to solar panels or batteries or GPUs.
I think emerging market equities are perhaps a more interesting asset to bring on chain than, let's say, US equities, because they have a fundamental product value proposition as opposed to simply access.
The thing that most excites me about putting US equities on chain is allowing global access to traditional American financial services and assets.
But when we think about an example like the Indian equities market And derivatives market, in many cases zero day options trade higher notional volume than the underlying spot assets do.
And so I think there are many places where users and retail investors are actually more interested in trading derivatives.
And those derivatives could be made much more efficient and given much greater access by moving on chain.
So I think the idea of perpification is very interesting.
And just to say what that means very literally, that means to make a perp or a perpetual future out of what is today an existing spot asset.
So a perpetual future is a derivative that allows an end user, an investor, to tune their leverage limits in a very intuitive and efficient way, simple way.
And there is a perp price that trades against a mark price.
And when the two prices diverge, there are fees or a funding rate that are charged to the holder of the asset.
And I think purplification is interesting for a wide swath of traditional assets because it's much easier to create a synthetic representation on chain that can scale to very high notional volume than trying to tokenize existing assets, basically copying them on chain.
And so, to say it in briefer terms, I think the idea of creating synthetic representations of traditional assets on chain is more easily scalable than creating literal copies of those assets on chain today.
So, specifically when we think about the credit markets, lots of people are interested in tokenization today.
But I don't know that tokenization brings as many benefits for credit assets as people would imagine.
I'm much more interested in native on-chain loan origination because I think it can significantly reduce the back office costs that are traditionally associated with creating basically asset-backed securities or other forms of credit assets.
I think taking a more crypto-native approach is actually better, regardless of whether you want to offer the product to a crypto-native audience or to a more traditional audience, because you're going to have much greater back office efficiency in many cases.
And a lot of these details are still being figured out, but I think there is great potential there.
When we talk about synthetic dollars on chain, meaning a token, a blockchain that is pegged to a dollar denominated representation but is collateralized by some underlying set of financial assets or, in many cases, a specific structured product.
I think there are a lot of opportunities to take things like currency, cash and carry trades or traditional infrastructure financing for things like GPUs or solar panels or batteries and reflect those on-chain.
And really this looks very similar in a crypto-native sense, to how traditional trading strategies have been turned into ETFs in existing markets, providing greater access and legibility to users.
I think that will happen much more on-chain.
I think there's a real opportunity for builders to create, let's call it, a synthetic dollar factory to take these existing, these interesting technologies, trades or investing products and peg them to a dollar, offer them to existing investors and help collateralize stable coins with those synthetic dollars.
Guy's point is that once the new rails exist, you can find financial products that are more native, more composable and cheaper to operate.
Now we shift to another domain where new rails can change the speed of progress, science.
Oliver Hsu explains how advances in AI reasoning and robot learning move us toward autonomous labs.
Not fully self-driving science yet, but real collaboration between scientists, AI systems and lab automation, where interpretability and traceability matter, because research requires understanding.
Why not just what?
Here's Oliver.
My name is Oliver Xu.
I'm a partner on the American Dynamism team here at A16Z.
And my big idea is that advances in AI reasoning capabilities and in robot learning will help accelerate scientific progress by moving us closer towards autonomous labs.
Laboratory automation is something that's existed for a long time.
Like, that is not new.
This idea of having robots that you can pre-program to assist in some of the motions involved in a lab.
What is new and what is emerging right now is the combination of reasoning capabilities and experiment planning and the physical element of lab automation.
So what that might look like in the near term is collaboration between a scientist and a system that involves both an AI application and a robot, and having that be a much more collaborative process in the near term in many different kinds of labs and many different kinds of scientific processes, whether that's in the life sciences, in the chemicals industry, in the material science research sphere, and so on and so forth.
One of the things that I think is important in the near term though is around interpretability.
So you know, if you think about AI systems as non-deterministic computers, one of the things that really matters for research is you want to really understand why the system is doing what it's doing, why it's planning on iterating on an experiment in a given way, why it's planning on doing this particular thing.
And I think you know, Systems that are purpose-built for scientific research are probably going to focus a lot on that on the interpretability, on recording what exactly is happening throughout each step of the process as it collaborates with a human scientist.
I think this concept of fully self-driving science right like a closed loop where you have AI that iterates on itself and then carries out an experiment and then continues to iterate without human intervention.
I think this is further out.
This is what I would consider the destination for this idea of autonomous science.
I think where we are right now is that there's a lot of work being done to form the foundations of autonomous science.
And, if you consider science as broadly speaking, some combination of theory, of computation and of experimentation.
There's work being done in the ai ecosystem across areas like mathematical reasoning, physical reasoning, simulation and world models and robot learning and all of these things.
Eventually, as these capabilities improve, can be applied to closing this loop.
But progress across all these fields is, of course, uneven and you kind of have to wait for the capabilities to get to the point where they're ready to be applied to this closed loop.
And i think that's the destination.
And in the near term incrementally, we'll make progress on, you know, lab automation, on the reasoning pieces of this, but ultimately the final destination, i think, would be around this idea of a self-driving lab or of autonomous science.
Part of this is going to be driven by the market dynamic in which the research is conducted.
So I think there are certain categories of science where there is just a much more mature demand side market for the outputs of research.
And examples include, of course, life sciences and pharma, the chemicals industry, facets of the material science industry.
I think these are areas where there is a ready and willing buyer for a lot of the outputs of this research and the increase in speed and capability, as well as any cost advantage that might accrue.
All of these things are going to matter more for markets where there is a well-established buyer of research output.
And so I think Where you see autonomous labs and autonomous science being adopted first is probably more of a function of the market that it's operating in.
I think Periodic Labs is a great example of a team taking a swing at autonomous science.
I think, when you look at the early stage startup landscape, there's companies like Medra that are focused on the life sciences and pharma market.
There's companies like Chemify and Yoneda Labs that are focused on the chemistry industry.
And then, zooming out a bit, there's collaborations between government and industry that are really focused on this intersection of AI and science.
There is the Genesis mission, led by the Department of Energy that brings together academia, government and the national labs, as well as leading AI companies, to pursue AI-driven science.
I think just today DeepMind announced a partnership with the UK government to collaborate on areas of scientific discovery.
So I think there's startups that are working on lab automation.
There's startups that are working on building an AI scientist.
And that work is happening against the backdrop of a broader collaboration between both the public sector and the private sector and academia to really accelerate AI-driven scientific discovery.
Oliver shows what happens when tools become systems.
Research gets faster when planning, execution, and iteration begin to close the loop.
To close, we focus on a different kind of primitive, distribution.
James DaCosta introduces the greenfield strategy AI-native startups selling to other AI-native startups early when there are few stakeholders, no switching costs and the chance to grow alongside customers as they scale into major companies.
Here's James.
My name is James DaCosta.
I'm a partner on the apps investing team.
My big idea for 2026 is the greenfield strategy.
We're going to see AI native startups selling to other AI native startups reach scale.
One of the biggest issues for startups is whether they can reach distribution before incumbents reach innovation.
And in this software cycle, incumbents are also adding AI as well.
So what can the startup do?
One of the most powerful and underrated ways for startups to win distribution is actually to serve companies at formation or greenfield companies.
The battle between every startup and incumbent ultimately comes down to whether the startup can get to distribution before the incumbent innovates.
And in this software cycle, incumbents aren't asleep on AI either.
But one of the best and underrated ways startup can win the game of distribution is actually selling to other startups.
Those startups have far fewer stakeholders.
You just need to convince a CEO and a couple of founders.
Those startups don't need a complete solution.
And those startups don't have any switching costs because they don't even have a solution in place today.
If you attract all of the new companies at Formation and then grow with them As your customers become big companies in their own right, so will you.
This is actually the playbook that Stripe used over a decade ago.
Many of Stripe's first customers did not exist when Stripe was founded.
But as Stripe's customers grew to become big customers in their own right, Stripe used them as a case study to then sell to other enterprises outside of Silicon Valley.
Incumbents struggle to sell to startups because they're bound by the rules of PL.
New startups represent very little in the form of new revenue for incumbents, but they cost money to serve in terms of marketing spend or sales or building out a new product.
As a new startup, you're still figuring it out, and you just need to get your initial product in customers' hands.
One of the most important things if you're following this strategy is to find a constant source of new customers.
And accelerators like Y Combinator or Speedrun or Entrepreneur First offer a perfect opportunity for that.
Mercury, for instance, works with 50 of every single YC batch and then grows with those companies over time.
Every single new startup still needs a CRM.
Every single new startup still needs a HR and workforce system.
If you can fundamentally find a narrow wedge and build a better solution for those AI startups, you can use that as a way to get into the market.
Once you've found a wedge and made that single wedge much better with AI, one of the really important things is that you have to keep shipping useful features to your customers to actually be able to grow with them and reduce the risk of churn over time.
Once you've actually got these new customers, you have to rapidly iterate and ship features and build out a wider product as your customers grow.
Graduation moments also offer the perfect opportunity for startups to apply the Greenfield strategy.
This is the moments where maybe startups have outgrown a simple solution that they have in place, like moving from QuickBooks to NetSuite.
And AI companies like Rillet are doing exactly this.
You also don't have to be constrained by the existing categories of enterprise software.
CRM, customer success, and finding pipeline all used to be different software systems.
But now, with AI, maybe the next new AI native CRM can find your leads, can track those leads and actually ensure your customers are successful once they're onboarded to your platform.
These three ideas fit together as a single thesis.
New rails create new compounding.
Guy shows programmable money evolving from stablecoins into on-chain credit origination in synthetic products that scale with lower operational friction.
Oliver shows how autonomy entering science through lab collaboration, with interpretability as the near-term requirement that makes it usable.
James shows a distribution strategy that compounds.
Win customers at formation and grow with them before incumbents can catch on.
This is what new infrastructure primitives really means here.
Not a buzzword, but the rails that let entirely new systems emerge and scale.
Thanks for listening to this episode of the A16Z podcast.
If you liked this episode, be sure to like comment subscribe, leave us a rating or review and share it with your friends and family.
For more episodes, go to YouTube, Apple Podcasts, and Spotify.
Follow us on X, A16Z, and subscribe to our Substack at a16z.substack.com.
Thanks again for listening, and I'll see you in the next episode.
As a reminder, the content here is for informational purposes only.
It should not be taken as legal business tax or investment advice or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund.
Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast.
For more details, including a link to our investments, please see a16zcom forward slash disclosures.
