READ MORE: a16z co-leads Kalshi’s Series D

In the mid-2010s, a new visual content format started popping up around elections, sports games and playoff races: the “probabilities changing over time” graph. These graphs were a compelling content format because they told a fascinating story: what was supposed to happen, and then what did.

You can tell amazing stories with these images. Just by looking at changing probabilities, you could tell a story about collapse, about redemption, or about an underdog beating the odds. (Kurt Vonnegut famously gave names to many of these stories: like “man in a hole”, “boy meets girl”, and “from bad to worse,” each with their own shape.) These images are a kind of meme: they compress a lot of information into a tiny space, and faithfully transmit the story intact when shared.

As engaging as these graphs were, they came with one major limitation: they didn’t really exist outside of politics, sports, or financial markets. There’s an obvious reason why: for these graphs to work, you needed predictive odds that were broadly accepted and legal to use. Finance has always had them; elections have polling numbers to work with, so you could construct these probability paths Nate Silver-style. And sports seasons (or even individual games) have well-understood structures and enough historical data to confidently predict a team’s midseason odds of reaching the playoffs. These aside, the format of “story shapes” could not spread any more deeply into popular culture.

READ MORE: Investing in Kalshi, by Chris Dixon

Prediction markets arrive, slowly

Prediction markets solve this problem in a fairly obvious way. As long as you can define a contract and the terms of its resolution, we now have a way for these “prediction shapes” to emerge about any story that’s going on in the world. Popular prediction – the required starter ingredient for this kind of story – goes from scarce to abundant.

In practice, these markets did not spring up overnight; not at first. In early 2024, the magazine Works in Progress published an article called “Why prediction markets aren’t popular.” The piece argued that there was “little natural demand for prediction market contracts,” since none of the three groups that classically make up market participants – savers (who seek to build wealth), gamblers (who make bets for thrills), and sharps (who seek to profit off the distortions caused by the former two groups) – have any particular reason to get involved in a prediction market. Savers who might buy a market index to build wealth over time have no reason to start betting on the outcome of a presidential election. Gamblers might be more inclined, but there are more interesting ways to speculate (day trading, memecoins, sports betting…) than making a prediction about the outcome of a state senate election. And with the other two groups having little involvement, sharps don’t see much money to be made by entering the market.

Subscribe now

With sparse involvement from those three groups, prediction markets were fated to remain illiquid and relatively useless for predicting the future. The poor performance of prediction markets in predicting the outcome of the 2022 midterm elections lent credence to this view.

But in the year-and-a-half since that article was published, something interesting has happened: prediction markets have surged into mainstream popularity. As one might predict from the massive volume bet on sports each week, the largest markets are in sports. But they’ve managed to surge into mainstream popularity – enough to be the subject of a South Park episode – while hosting markets on everything from the results of the New York City mayoral election to the path of Federal Reserve policy rates to the time period in which Taylor Swift will get married.

Read More: Prediction: the Successor to Postmodernism

The fourth wall breaks

What changed, if anything, over the past two years? There probably wasn’t one silver bullet. The 2024 election definitely helped: Americans have a long history of betting on elections, and prediction market volume grew 42-fold between early June and the week of the election. But it didn’t fade once the campaign was over.

The key actor in this positive feedback cycle was a new type of market participant, one that didn’t exist a few years ago but is now everywhere. This participant is similar to the promoters that you see in traditional betting events, like a boxing match in Las Vegas. It’s the humble, ordinary social media poster: and the new meme format of posting prediction paths as screenshots.

Prediction markets are now as much about social media-driven virality as they are about the classic dynamics of the market. The critical behavior mechanism is posting screenshots of the betting contract at the moment it becomes topically interesting, drawing awareness and liquidity into the contract.

For a great example, look at this Kalshi contract for the pop culture question of the year: “Will Taylor Swift and Travis Kelce get married in 2025?” If you look at the chart, you’ll see two important things happen on August 26th, when Swift and Kelce announced their engagement on Instagram. First, a spike in the odds; and second, a huge increase in liquidity as people start paying attention to the contract. While some amount of the liquidity jolt would’ve arrived regardless, there’s no doubt that these shared screenshots, at critical moments, constitute viral awareness of the contract itself, and a top-of-funnel draw into the bet. This kind of “fourth wall break,” where the broader audience suddenly gains awareness of the meme (or, more properly, of the reason to care about the contract at all), will give an interesting new meta element to future stories.

A new kind of Main Character for the timeline

Betting on the Pope was supposedly the “original prediction market”, and this most recent time around saw a glorious return to that tradition. It was a great moment for Catholics around the world, as Cardinal Robert Prevost assumed the first American Papacy, as Pope Leo XIV. And it was a great moment for betting markets too, as few people had considered him a viable candidate: most of the attention had been focused on frontrunners like Pietro Paolin and Luis Antonio Tagle.

A day after the white smoke cleared, @Domahhhh on X shared a real gem with the timeline: the play-by-play breakdown of his thought process and his bet sizing, in both the days leading up to conclave and in the critical minutes between the decision and the reveal.

In his words: “As a directional bet, I decided to just bet a ton that the next Pope would be someone other than [Parolin and Tagle, the frontrunners].

The white smoke went up after the fourth ballot. This is, relatively speaking, fast. And the logical conclusion (and one that I immediately jumped to) is that this means a strong vote getter from the 1st round has consolidated the vote and become Pope. Parolin went to ~65%. Tagle stayed around ~20%. These two were 85% to be Pope, and tbh even though that price was incredibly wrong in hindsight, it’s hard to think it’s THAT wrong in the moment. I was convinced I had lost a lot of money! I decided not to chase good money after bad, and not to bet more against Tagle/Parolin. I would accept my loss like a good little boy.

But what I DID do is scroll through the list of other options. When you have two people now trading at 85%+, everyone else is in the clearance aisle. And I went rummaging in the bargain bin, and found Turkson at 100-1 and Prevost at 200-1. In hindsight, I should’ve also bought Grech at whatever he was at.

I knew one piece of information: it was four ballots. It’s too quick for a longshot. Throw all the longshot lottery tickets straight into the garbage. You need someone with gravitas, who is capable of capatulting to 2/3rd in a relatively short time period. And those two were the ones I picked out. I bought thousands of shares of each of them as other traders were focused on Tagle/Parolin.

A few minutes later, my jaw dropped as Prevost -- the guy I had just amasssed shares in at 200-1 like 20 minutes earlier-- walked out onto the balcony as Pope.”

There used to be this joke, “Every day on the timeline there’s one main character, and the goal is to never be it.” This kind of “victorious prediction poster” is a new kind of main character that gets to be a momentary certified hero.

Skin in the Game

The 2024 presidential election gave prediction markets a well-deserved redemption story. It started with the prolonged saga that ended with Biden dropping out, during which prediction markets offered a useful quantification of how various events affected the odds of the president leaving the race. Everyone from journalists to Wall Street traders began to lean on prediction markets along with the traditional tools of polls and punditry. Ultimately, the prediction markets – criticized throughout the campaign for the alleged influence of “whale” traders – beat the polls. And nearly a year later, Kalshi daily volume now exceeds the 2024 election (during football games, at least).

Prediction markets now stand for something important, beyond their utility as financial instruments or information sources. They stand for a kind of accountability, and for a new kind of character on the timeline: “the hero that made the brave call,” “the idiot that made the bad one”. These people, now brought to the forefront as Vonnegut caricatures, as if it were one of his short stories.

In all kinds of walks of life, from politics to business to culture, we ask that our leaders and our public figures actually lead our institutions to successful futures. And that means making gutsy calls that turn out right. For the past few decades, there’s a popular sense that we’ve slid into a culture of unaccountability among some of those leaders, and we appreciate all the more when individual people step up and buck that trend.

That, perhaps, is the main mechanism by which prediction markets are set to alter the course of popular culture: not only because betting itself is an information flow that directs attention where it’s needed, but because the path taken by a prediction from inception to finish brings new memes onto the timeline, and new characters into relief.

Share

Views expressed in “posts” (including podcasts, videos, and social media) are those of the individual a16z personnel quoted therein and are not the views of a16z Capital Management, L.L.C. (“a16z”) or its respective affiliates. a16z Capital Management is an investment adviser registered with the Securities and Exchange Commission. Registration as an investment adviser does not imply any special skill or training. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell — or a solicitation of an offer to buy — any securities, and may not be used or relied upon in evaluating the merits of any investment.

The contents in here — and available on any associated distribution platforms and any public a16z online social media accounts, platforms, and sites (collectively, “content distribution outlets”) — should not be construed as or relied upon in any manner as investment, legal, tax, or other advice. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Any charts provided here or on a16z content distribution outlets are for informational purposes only, and should not be relied upon when making any investment decision. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, posts may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein. All content speaks only as of the date indicated.

Under no circumstances should any posts or other information provided on this website — or on associated content distribution outlets — be construed as an offer soliciting the purchase or sale of any security or interest in any pooled investment vehicle sponsored, discussed, or mentioned by a16z personnel. Nor should it be construed as an offer to provide investment advisory services; an offer to invest in an a16z-managed pooled investment vehicle will be made separately and only by means of the confidential offering documents of the specific pooled investment vehicles — which should be read in their entirety, and only to those who, among other requirements, meet certain qualifications under federal securities laws. Such investors, defined as accredited investors and qualified purchasers, are generally deemed capable of evaluating the merits and risks of prospective investments and financial matters.

There can be no assurances that a16z’s investment objectives will be achieved or investment strategies will be successful. Any investment in a vehicle managed by a16z involves a high degree of risk including the risk that the entire amount invested is lost. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by a16z is available here: https://a16z.com/investments/. Past results of a16z’s investments, pooled investment vehicles, or investment strategies are not necessarily indicative of future results. Excluded from this list are investments (and certain publicly traded cryptocurrencies/ digital assets) for which the issuer has not provided permission for a16z to disclose publicly. As for its investments in any cryptocurrency or token project, a16z is acting in its own financial interest, not necessarily in the interests of other token holders. a16z has no special role in any of these projects or power over their management. a16z does not undertake to continue to have any involvement in these projects other than as an investor and token holder, and other token holders should not expect that it will or rely on it to have any particular involvement.

With respect to funds managed by a16z that are registered in Japan, a16z will provide to any member of the Japanese public a copy of such documents as are required to be made publicly available pursuant to Article 63 of the Financial Instruments and Exchange Act of Japan. Please contact compliance@a16z.com to request such documents.

For other site terms of use, please go here. Additional important information about a16z, including our Form ADV Part 2A Brochure, is available at the SEC’s website: http://www.adviserinfo.sec.gov

| America | Tech | Opinion | Culture | Charts |

Prediction markets have product-moment fit: they’re perfectly suited for the zeitgeist right now. And I think they carry a torch for a much bigger movement than what you see at face value. “Prediction” is now the entire game. It is the movement and meta-aesthetic that defines our main branch of society going forward; about how we interpret the world and participate in the AI century.

Prediction is as significant a concept as modernism and postmodernism. It’s the defining reframe of our generation, and we’d better understand it:

• Postmodernism was why we built businesses the way we currently do: why we treat innovation as “capital at risk”, why we have the “front end / back end” service model, and other themes that are so pervasive they’re invisible.

• “Prediction” is a distinct break from postmodernism. It is what comes next.

• It can be hard to tell, in the moment, whether you’re looking at the beginning of something or the end of something. Many early AI artifacts that we think are “the future” are actually the final form of the old thing, rendered perfectly at their moment of obsolescence.

• All value creation now is about prediction.

• Prediction markets give us a unit we can understand, like how patents represented progress a century ago, that organized our ambition and represented a stake in the system.

• Prediction helps us break free from postmodern malaise, and gives us a new purpose in the world: we create order in the universe by contributing information.

Postmodernism and its native business model

Our current nostalgia for a time “back when we built things” draws much of its aesthetic from a movement called Modernism, which was the main thing going on in the late 19th and early 20th centuries. We call them “movements” because they are truly grand ideas: multi-decade patterns of perceiving and thinking about art forms, business forms, and national forms; asking, “what is going on here?” at the highest level we can.

Peak modernism was about nation-states asserting their dominance as the new form of human identity and organization. The old multi-ethnic empires like the Habsburgs and Ottomans were dying; art movements like the Romantic period celebrated “unified” peoples, languages, aesthetics, and shared histories. This was a period of time where things in the world were the sum of their parts. We thought about one national heritage, or one body of science.

Nothing lasts forever, and Modernism certainly didn’t. After the pain of the World Wars, we retired that mindset for a subjective one, where identity is “rendered” case by case. Where modernism was about cause, postmodernism emphasized combinatorics: the thing you see is one arrangement of pieces; the thing I see may be another. That’s how we got Freud’s ego-self as a surface rendering of subconscious drives, defining art like Glenn Gould’s Goldberg Variations, and McKinsey. To experience peak postmodernism, watch Friends - first without the laugh track (the pieces themselves) and then with it (the rendered product). If you’d prefer, you can watch any episode of reality TV, or pro wrestling. It’s all the same.

Postmodernism quickly figured out its business model: the “front of house / back of house” abstraction of cost-effective, personalized service at scale. My favourite essay on this is Venkatesh Rao’s The American Cloud, on the Great Back End (the enormous logistical enterprise, hidden from view, that grows our food and heats our homes) becoming abstracted out of view, while coastal consumers interact with pleasantly recreated “situations”, like Starbucks or Whole Foods.

This idea was so powerful that its best biographers were its Marxist critics. In Postmodernism: or, the Cultural Logic of Late Capitalism, Fredric Jameson wrote in a tone that’s so frustrated it betrays a hint of admiration: “Aesthetic production today has become integrated into commodity production generally: the frantic economic urgency of producing fresh waves of even more novel-seeming goods (from clothing to airplanes), at ever greater rates of turnover, now assigns an increasingly essential structural function and position to aesthetic innovation and experimentation.” This is almost a sentence you could find in Atul Gawande’s completely sincere article about why hospitals should operate more like The Cheesecake Factory. It could also describe mobile app development. There’s one coherent idea through it all: the meaning of things is how they’re rendered-for-purpose.

From “progress” to “innovation”

Modern and postmodern actors think about progress in very different ways. Modernism understood progress as something that simply happened; perpetually and unstoppably. And that way of thinking about the world became a self-fulfilling prophecy: the early 20th century delivered on those promises, as houses lit up with electricity and living standards shot up everywhere. Then in the middle of the 20th century, something interesting happened: we stopped talking about “progress”, and found a new word, “innovation”.

(Playing around on Google ngram - not to be taken too seriously)

In earlier times, “progress” came from either sheer force of will, or from luck (fortuna). But innovation is a postmodern concept: it reorients our goals into units of risk to be assessed, and prudently taken. Forward motion now happens in parallel expeditions, from measured acceptance of capital-at-risk. We rendered the potential out of a combination of bets, into a useful product. In 1957, Peter Drucker wrote in Landmarks of Tomorrow: “Innovation is… a new view of the universe, as one of risk rather than of chance or of certainty. It is a new view of man’s role in the universe; he creates order by taking risks. And this means that innovation, rather than being an assertion of human power, is an acceptance of human responsibility.”

Drucker was not predicting the future; he was trying to describe the present accurately. But he captured the mindset that led Bell Labs and the early Venture Capitalists to seed the beginnings of the future; and for startups to become a financeable endeavour. This is a big part of why Silicon Valley was so aesthetically compatible with what was going on in the rest of the country. Trying lots of small bets, searching for product-market fit, and getting rich at IPO were all postmodern quests. They fit with the rest of what was on TV.

These days in tech, we don’t use the words “progress” or “innovation” as much as we used to. Nowadays we talk about technological adoption. Adoption isn’t a new word; it certainly existed before, but it used to mean something specific about consumer requirements; whereas now it means the entire thing. I think there’s a clue here in where we’re going: that the important essence is not how you choose to engage (the postmodern mindset), but rather when you engage with the snowball rolling down the hill.

Prediction in culture: “I want to feel something, even if it hurts”

Now that our history basics are out of the way, let’s actually talk about prediction; the movement that’s happening now.

“Prediction” has two essential characteristics. The first is what we mean when we talk about prediction markets, or honestly any market. Prediction is the action that markets perform, as information moves around society and directs attention and resources to attractive destinations. Where they go constitutes a kind of prediction. The second essential characteristic of prediction is that computers are getting good at it. Computers now provide a broadly available superstructure for, “given this chain of things, predict a good next thing”, which will keep improving.

With that in mind, let’s go to the present moment and describe some of what we see.

The first observation I will make about popular society is how suddenly and decisively our culture moved towards speculation; not only in stocks and sports betting but also more broadly in life. Meme stocks might only be a few years old, but popular culture preceded them by a decade when Home & Garden TV programming became entirely about house flipping. Large swaths of the population now buy into the idea of “making your own luck” as a more literal virtue than it used to be.

This matters because there is now a meta-awareness (or meta-fantasy?) among the general public that our limited capital and attention could, in fact, find better places to go. There’s a driving energy in this world that everyone is doing this; and therefore, our attention each moment “predicts” the next one in a real sense. Not just forecasting it, but also setting the path dependence that leads us down the road and to a conclusion.

The second observation that I’ll make is how much social media now feels the same way that buying stocks does. Checking if your post is performing feels the same as checking whether your stock is up or down, and that tells you something important about your relationship to both activities.

At the forefront of culture and commerce, flash sales and popup stores are a revealing break from business convention where removing friction is always good. A sneaker drop is all about adding friction to the sales process. That’s why people like them: your place in line is the meaning of the product. Gaming predicted all of this, of course: that’s an industry that completely understands that friction is the product, and anticipation is consumption. The meaning of the game is the path you take through it, trying to feel something.

How early or late you are to something is now an essential component of your relationship to that thing. The timelines and reels that represent “what is going on” are increasingly about a single meta-topic: are you predicting it, or is it predicting you? This has become the main thing that you feel. And it is a complete break from the postmodern aesthetic, where your consumption was wrapped in an unthreatening fuzzy blanket. It doesn’t matter what time of year you arrive at Whole Foods to buy strawberries; the farmstand simulacra is recreated faithfully. The prediction aesthetic is a new thing, and rejects postmodernism: “I want to feel something, even if it hurts.”

“Is this art?”

It can sometimes be hard to tell whether you’re looking at the beginning of something, or at the end of something. In 2008 it sure felt like Obama was the beginning of something significant, but in hindsight he was actually the last president of the 20th century. Similarly, I think many of us thought the scrolling timeline feed was what came after newspapers; but in hindsight, the feed may be the last newspaper.

I remember the moment when AI generated art suddenly burst into public consciousness with Dall-E and Midjourney, and everyone gasped, “this is the beginning of a completely new kind of art!” I think with hindsight we will look at it with nostalgia, because it’ll actually be the last postmodern visual art form. It’s the final, perfect form of “everyone gets to consume the version they would like!”, at the moment it became culturally obsolete.

On that note, you’ll remember a lot of freakout at that time, “In the future, everyone is going to get their own personalized echo chamber of AI-generated news!” which is 1) not what’s happening, and 2) clearly our projection of our anxiety onto the old form factor and its problems. What’s actually much closer to happening is, “Everyone is going to have their own personalized parlay on the same series of reference events.”

So what is an actually new art form? I don’t consider myself a contemporary art connoisseur, but I am pretty online. And there’s a definite art form emerging, which is public predictions that go really well or really badly. For example, this was art:

(If you don’t immediately recognize what this is - it’s a couple who dressed up as Taylor Swift and Travis Kelce for Halloween in 2020, who captioned their Instagram, “I don’t know any world in which Taylor Swift and Travis Kelce would be in the same room together. But apparently in this one they’re married?”) This post, now virally resurfaced, is certified pop art.

READ MORE: Prediction Path Screenshots: a New Kind of Meme

Another kind of art, which may make you mad, is NFTs and even more so Meme Coins. Because, 1) if suggesting something is art makes people Rite of Spring-level furious, you have something interesting; and 2), because they are a pure demonstration of public predictions as art, and the meaning of a thing being when you interact with it.

For example:

That’s art! Andy Warhol famously quipped of the postmodern era, “Everyone gets their 15 minutes of fame.” Today, I think it’s, “Everyone gets one 15-bagger.”

There also doesn’t necessarily need to be a prediction made, per se, to qualify as this kind of art. Just the prediction path itself is sometimes all you need:

Better luck next year, Steve Cohen.

Value creation in the Predictive Era

Every era finds a defining business model. For the modern period, it would be the vertically integrated giants that decided wars. For the postmodern period, it’s the front-of-house, back-of-house simulacrum. Our job today is to imagine what forms of business naturally resonate with where our cultural aesthetic is going.

READ MORE: Investing in Kalshi, by Chris Dixon

Crypto was really avant-garde here, because they had to argue with everyone, “Onchain isn’t in the front end or the back end, it’s this new third thing”, that was hard for us to put into words, especially for practical app-building purposes. Culturally speaking, the proposition remains: “for every piece of content, there is now a publicly viewable timeline for whether you own the content or not, and when you bought and sold it”, which produces very silly results when taken at face value, but contains a critical insight. Which is, “We’re all participating in one great online game, and your goal is to be predictive of the game, rather than predicted by the game.”

But the real party started when LLMs became a product in their own right, and suddenly this nebulous thing called “weights” became the most sought-after asset in the world. As AI models start contributing to business decisions, showing ads, and influencing the culture, we now have a tangible place where “everyone has their own personalized parlay on the same series of reference events” is now no longer about gambling; it’s about the whole timeline of your life. And in many cases, the house is actually trying to help you win.

“Gen” was the best word I could find at the time; “Prediction” is much better.

I think back on Packy’s Great Online Game piece a lot, because this was a piece that was basically about training LLMs, except for it was written in 2021, about ZIRP Online Citizenry. “Being good at the Internet” is now a highly practical and monetizable career - it used to be, too, but less legibly. Today, it’s easier to understand why being online is naturally monetizable: because prediction is it; prediction is the entire meta-game, from meme coins to model training, and the farther upstream you are in the information flow, the more likely you’ll be predictive of the game as opposed to being predicted by the game. It is suddenly not-so-ridiculous when Matthew Prince of Cloudflare asserts, “the way you’ll get paid on the internet now is by writing things that the AI does not yet know.” We’re not there yet, but there is a plausible path to it.

Meanwhile, the institutions of the future are getting built in a Venture Capital environment today that has completely embraced and assimilated into the new prediction movement. The old days where you ran a fixed fund size and accumulated early stakes in a single-stage portfolio may still work on rare occasions, but it’s definitely passé. Today, everything is about multi-stage continuity, and whether you’re on the predictive path.

Prediction contracts as the atomic unit

Every builder paradigm comes with its own financial instrument; its own way of capturing potential into an ownable, tradeable thing. The most important one of these in the modernist era was the patent. Patents did much more than legally protect an innovation and help them monetize: they also captured the imagination of the public, and became an “instrument of progress” that you could name and appreciate. They encapsulated the thinking of the time, defining an invention as literally the sum of its parts. “Parts” could mean physical components or mechanical processes or other contributing methods, but whatever they were, a patent encapsulated them all into a unitary invention.

Once the broad public came to associate patents with invention, they became what invention was, in a self-fulfilling prophecy. In time this became a bad thing, as universities and researchers overrotated on “produce and monetize patents” to the detriment of actual progress. But for a long time, having a single idea that you could associate with invention was almost certainly a net positive. Patents inspired people to think like inventors; they organized our abilities, and set a bar you needed to achieve. They were a stand-in for “progress” in our modern perception of the world.

I wonder if prediction contracts might serve a similar purpose: as a centrally organizing idea, and commonly accepted format, for the incredibly broad concept of “prediction” that’s so important to our main thread. Being a live player means making predictions. “Putting your money where your mouth is”, and actually committing to a prediction, feels like a new form of, “Oh so you think you invented something? Patent it!” and that’s a good thing. Especially considering that patents were only something a few people ever accomplished, whereas anyone can participate in a prediction contract, and in their own little way, contribute.

Our purpose in the world is to contribute information

We may soon arrive in a world where intelligent machines are conceiving, driving, and implementing a lot of our material progress. They’ll be making scientific discoveries, recursively designing new robots, and who knows what else. That’s a great thing. We’ll take all the progress we can get.

But it is essential that people in the world feel like they have a stake in the system; like they’ve contributed to our shared sense of progress. And I believe Predictions, as a movement, are the answer to our postmodern malaise. They give us something to do.

I see no better explanation for the rise of speculation, gambling, and public prediction in the last ten years of popular culture than as a mass revolt against the numbing apathy of late postmodernism. And even though I find some of it uncouth and crass, I don’t see it as a regression: it’s a step forward, in self-discovery. We have a new way to think about our purpose in the world: we create order in the universe by contributing information. And in the past few years, we’ve discovered the extent to which it’s a shared purpose; perhaps a universal one. I couldn’t imagine something more optimistic if I tried.

Share

Views expressed in “posts” (including podcasts, videos, and social media) are those of the individual a16z personnel quoted therein and are not the views of a16z Capital Management, L.L.C. (“a16z”) or its respective affiliates. a16z Capital Management is an investment adviser registered with the Securities and Exchange Commission. Registration as an investment adviser does not imply any special skill or training. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell — or a solicitation of an offer to buy — any securities, and may not be used or relied upon in evaluating the merits of any investment.

The contents in here — and available on any associated distribution platforms and any public a16z online social media accounts, platforms, and sites (collectively, “content distribution outlets”) — should not be construed as or relied upon in any manner as investment, legal, tax, or other advice. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Any charts provided here or on a16z content distribution outlets are for informational purposes only, and should not be relied upon when making any investment decision. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, posts may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein. All content speaks only as of the date indicated.

Under no circumstances should any posts or other information provided on this website — or on associated content distribution outlets — be construed as an offer soliciting the purchase or sale of any security or interest in any pooled investment vehicle sponsored, discussed, or mentioned by a16z personnel. Nor should it be construed as an offer to provide investment advisory services; an offer to invest in an a16z-managed pooled investment vehicle will be made separately and only by means of the confidential offering documents of the specific pooled investment vehicles — which should be read in their entirety, and only to those who, among other requirements, meet certain qualifications under federal securities laws. Such investors, defined as accredited investors and qualified purchasers, are generally deemed capable of evaluating the merits and risks of prospective investments and financial matters.

There can be no assurances that a16z’s investment objectives will be achieved or investment strategies will be successful. Any investment in a vehicle managed by a16z involves a high degree of risk including the risk that the entire amount invested is lost. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by a16z is available here: https://a16z.com/investments/. Past results of a16z’s investments, pooled investment vehicles, or investment strategies are not necessarily indicative of future results. Excluded from this list are investments (and certain publicly traded cryptocurrencies/ digital assets) for which the issuer has not provided permission for a16z to disclose publicly. As for its investments in any cryptocurrency or token project, a16z is acting in its own financial interest, not necessarily in the interests of other token holders. a16z has no special role in any of these projects or power over their management. a16z does not undertake to continue to have any involvement in these projects other than as an investor and token holder, and other token holders should not expect that it will or rely on it to have any particular involvement.

With respect to funds managed by a16z that are registered in Japan, a16z will provide to any member of the Japanese public a copy of such documents as are required to be made publicly available pursuant to Article 63 of the Financial Instruments and Exchange Act of Japan. Please contact compliance@a16z.com to request such documents.

For other site terms of use, please go here. Additional important information about a16z, including our Form ADV Part 2A Brochure, is available at the SEC’s website: http://www.adviserinfo.sec.gov

The impossible

AI is reshaping both soft power and hard power around the globe. The United States, to its credit, has an early lead with the former. The leading LLMs are trained on Western text, global training and inference are still dominated by American companies, and we are ahead in the global race for market share of total tokens generated.

But as it stands, China is running away with the hard power part of AI – robotics. As the incredible progress in AI continues, we start seeing intelligence embedded in the physical world – culminating in generalist robots that perform a wide variety of tasks across applications, from manufacturing to services to defense. This will redefine every aspect of our society and reshape daily life. The country betting on that future is China, not the US.

In the 10 years since the CCP released its “Made in China 2025” strategy, Chinese companies have leapfrogged the rest of the world’s density of robots per capita. They passed the United States in 2021, then the famously automated economics of Japan and Germany in 2024, and will soon eclipse Singapore and South Korea, their last remaining contenders. In short order China has become the world’s central robotics power. Entirely autonomous “dark factories,” like those of smartphone and automobile manufacturer Xiaomi, operate in complete darkness with no humans present.

China has successfully executed what we once thought impossible. Only ten years ago we scoffed that “China can copy, but they can’t innovate,” which we then revised to, “They can innovate, but they can’t make the upstream high-precision tooling.” Maybe we shouldn’t have been so comfortable, given how Chinese companies had outcompeted the rest of the world in industry after industry – from solar photovoltaics, where competition outside of China has been practically decimated, to 5G, whose global deployment was a massive success for China’s national champion Huawei. The same pattern is playing out now with robotics. China has built a playbook to dominate strategic industries, and has used that playbook to become the robot superpower.

Homegrown Chinese companies now design and fabricate precision parts like harmonic reducers at competitive quality, cheaper prices, and – most importantly – colocated with their customers in manufacturing superclusters. This is the part that should scare the West the most. The colocation of so many robot toolmakers, assemblers, and customers in nodes like Shenzhen or Shanghai is how new combinatorial use cases are discovered, how manufacturing sequences are optimized around that new potential, and how firms develop advanced process knowledge that is completely opaque to the West. In a few years, it will be Chinese companies that are making parts that we cannot replicate – not just at low cost, but at any cost. There are parallels from the past. In the 1970s, Japan shocked the world with Toyota’s lean production methods, just-in-time inventory, and ethic of kaizen, continuous improvement to eliminate waste. Initially dismissed, by the 1980s Japanese automakers had overtaken American and European giants and reshaped the global auto industry. If we do not act to avert it, this will be another Toyota moment, but on a much greater scale.

If we don’t act soon, the United States will find it extremely difficult to catch up: we are approaching a period of compounding improvement that threatens to make China’s advantage virtually insurmountable. As with LLMs, training advanced robotics systems requires pretraining data on the scale of the internet, along with reinforcement learning to train generalist policies that can reason across a wide range of distortions in environment, perception, and task. As data from real-world deployment comes online, the country with more robots gains flywheel momentum; more deployment means more high-quality data which underwrites further deployment. The United States isn’t entirely out of the game, and our lead in AI software carries over: American companies like World Labs are at the forefront of building frontier models that could allow robots to reason about 3D space. But as these capabilities mature, it will be action in the real world – from routing cable harnesses through chassis pathways in electronics assembly to simply doing laundry – that will unlock the economic and strategic promise of generalist robotics.

Micron tolerance

To understand what China has achieved in the past few years, let’s talk about the harmonic reducer – a simple manufactured part that’s deceptively hard to make.

Harmonic reducers are a type of gear system that looks almost like a shoulder or an elbow in its socket. They transfer rotational energy from one end (usually at high speed, from an electric motor) into a much slower gearing, at high torque. They do this by offsetting an inner and outer gear ring that are slightly offset from one another, paired with a rotating oval-shaped piece on the inside. When driven by an electric motor, this creates a waveform that slowly drives the outer socket with a high gear ratio and high torque – suitable for many robotic applications, including humanoid ones.

The challenge in manufacturing these tools comes from how sensitive they are to minute distortions in tooling and operating. They must be made micron-level precise, at low cost, to do their jobs correctly. Even more precision is required when these sockets are chained together into systems with multiple degrees of freedom, like the multiple joints on a robotic finger, hand, or limb. Achieving the strength and dexterity of a human hand, at non-prohibitive cost, requires true manufacturing excellence.

The precision required to manufacture harmonic reducers is well beyond the reach of most machine shops. Production has historically been dominated by highly specialized German and Japanese manufacturers: the Japanese company Sumitomo and the German-Japanese firm Harmonic Drive are the two dominant players in the space, together accounting for 95 percent of global market share. But in the last few years they’ve faced intensifying competition from new entrants from China. A firm called Green Harmonic, based in the city of Suzhou near Shanghai, offers harmonic reducers with performance comparable to products from Sumitomo and Harmonic Drive, but at roughly 30 to 50 percent cheaper price points. Green Harmonic now has more than 30 percent market share within China; and will soon look abroad. In the coming years, we can expect companies like Harmonic Drive to face their “Toyota moment,” with major strategic implications: there are countless cases of Chinese firms translating cheap, reliable manufacturing into global market share and eventually driving competitors out of business.

Harmonic reducers are just one illustrative part of the robotics hardware stack. Creating a fully functioning robot requires a huge variety of other small components – precision bearings that enable smooth joint rotation, custom printed circuit boards that route power and signals between subsystems, specialized connectors that maintain reliable communication in high-vibration environments, miniature encoders that provide millimeter-accurate position feedback, force-sensitive resistors embedded in fingertips for delicate manipulation, inertial measurement units that track orientation changes down to fractions of a degree, servo motors with sophisticated current control algorithms, shielding to prevent electromagnetic interference between tightly packed electronics, thermal interface materials that dissipate heat from high-performance processors, and countless fasteners, gaskets, and protective housings engineered to withstand the mechanical stresses of real-world operation. Each component must be carefully selected not just for its individual performance characteristics, but for how it integrates with the broader system: a single point of failure can render a sophisticated robot completely inoperable.

Chinese companies, from Siasun and Estun in controllers to AVIC Electromechanical in torque sensors, are rapidly entering and starting to win the market for every part of that system. Together, these firms and countless others constitute a sophisticated and mature ecosystem that has allowed Chinese firms to locally source practically the entire robot – not only from within China, but within a megacluster like Shenzhen.

We’re at the point today where Chinese domestic manufacturers and their suppliers contribute all of the parts necessary to bring robotic dreams to life, and iteratively learn from one another. The Chinese startup Unitree has captured the global imagination with highly advanced robots cheaper than anything else offered before – agile and LLM-integrated robot dogs for as little as $1,600, a humanoid for $5,900. Those costs will keep coming down; the robot dogs will keep getting stronger and more capable.

How the West lost “open vs. closed”

How has China converged on the global frontier so quickly? As with other parts of the Chinese industrial sector, state subsidies have helped. Many Chinese cities offer robotics companies complete tax deductions on their research expenses, generous subsidies, and preferential corporate income tax rate. Even though China offers more extensive state support for other strategic sectors like electric vehicles or batteries than for robotics, the cumulative economic assistance certainly helps domestic industry and keeps the Chinese market impenetrable to outside entrants.

But focusing on those subsidies misses the biggest part of the story. The Chinese robotics industry thrives off the unique advantages offered by the Chinese market. The scale of the Chinese industrial sector provides an endless variety of shopfloor scenarios to train and harden robots, and dense industrial agglomerations around Shanghai or Shenzhen provide endless opportunities for colocation and close collaboration at every layer of the supply chain – from component suppliers like Green Harmonic to end-users like BYD. Chinese factories adopt new technologies quickly and benefit from superfast digital infrastructure: they are now among the most networked in the world, filled with high-definition cameras and precision sensors collecting data over low-latency 5G connections. And below that bleeding-layer layer of Chinese industry there is a vast stratum of low-cost, highly-skilled labor that can churn out sophisticated hardware at a cost and speed virtually unmatched anywhere else in the world.

Most importantly, all of this occurs in a dynamic, competitive, and open ecosystem: new firms and technologies enter all the time, with few of the regulatory barriers that exist in other countries. The natural selection of market discipline removes uncompetitive firms rapidly, allowing those companies that emerge as global competitors, like Unitree, to be among the best companies in the world. Chinese entrepreneurs might enjoy state support in strategic sectors, but more importantly they benefit from the breakneck innovation and intense competition of an extremely dynamic open market – all while the CCP is able to keep that market opaque to outsiders and control access to it at will.

The United States was able to win the key industries of the twentieth century for reasons not so dissimilar to why China is winning now: a dynamic and open market, sometimes combined with strategic government support, created epochal American companies that unleashed immense economic value – from Boeing and Lockheed to IBM and Intel. But now we are undermining our historic advantage. Not only are we failing to thoughtfully modernize regulations created to address yesterday’s challenges – before we faced the prospect of a global competition with China – but some policymakers are busy adding new barriers, like administrative requirements for AI startups or restrictive copyright proposals at both the federal and state levels. China is doing everything they can to ensure that they win the industries of the future. The same cannot be said for the United States.

We know where this ends

When Marc Andreessen wrote that “software is eating the world,” few people understood at the time the extent to which every business was about to become a software business and every organization a software interface. The same thing is now happening in the physical world: robotics will simply become every manufacturing business, every part of the physical world.

But the transformations that the internet facilitated took place in a setting where governments recognized the benefits of innovation and encouraged a light-touch regulatory regime. Now they’ll be playing out in the regulation-heavy workplaces and factory floors of the United States. Our “permission-first” regulatory regime in everything from drones to sidewalk delivery robots means that any company interested in robotics will have to spend time and money dealing with lawyers, permits, and endless regulatory approvals. For more established players, agreements with labor unions make automation maximally difficult: dockworkers unions have been fighting against automation for years, helping ensure that American ports are some of the slowest and least efficient in the world. Simply put, we are not set up to win.

Given how difficult it is for companies to introduce automation of any kind, it shouldn’t be surprising that many sectors of the American economy are defined more by cost disease and rising prices than by increasing productivity. Construction labor productivity declined by more than 30 percent between 1970 and 2020. Even in manufacturing, which should normally be resistant to cost disease, the United States has seen its labor productivity decline over the last 15 years.

There are no “dark factories” in the United States. The closest that we have is Tesla’s Gigafactory Nevada, which is 90 percent automated. No other major manufacturer comes close. Because our highly restrictive regulatory approach constricts both demand and supply for robots – it’s hard for startups to build them and for enterprises to buy them – the American robotics industry has little global presence. There is no self-sufficient American robotics supply chain. There are robotics companies based in the United States, but they are almost always dependent on Chinese suppliers: products that are labeled “Made in America” will still have core components that originate in China. Even a component sourced from an allied nation will inevitably depend on China – CMOS image sensors, dominated by Japan’s Sony, depend not only on neodymium and dysproprium sourced from Chinese refiners but also on lenses produced by Chinese companies like Sunny Optical.

The trajectory is obvious. Chinese companies are slated to dominate and oversupply market niche after market niche until production is uneconomical for all competitors. Foreign suppliers are feeling the heat of Chinese competition; some have been acquired outright by Chinese firms. KUKA, a massive German manufacturer of industrial robots with roots in Bavaria dating back to 1898, was acquired by a Chinese company in 2016, with the German government changing its foreign direct investment policies only after the acquisition had been approved. There are a few bright spots of resilience – the United States, with the help of Taiwan’s TSMC, maintains an advantage in advanced logic chips, with Nvidia’s Orin the most popular chip powering advanced robotics. But Unitree’s G1, one of the most advanced humanoids on the market, doesn’t use Nvidia chips. It uses the Rockchip RK3588, made in China.

What we need to do

It’s time for the United States to get serious about American robotics. As generalized physical intelligence comes into view and robotics becomes a major force in reshaping countless critical industries, our absent robotics industry will emerge as a critical vulnerability for American strategy. It is not too different from importing all our critical networking and telecom equipment from Huawei.

Domestically, we need to stop shackling our robotics industry. Market discipline from competition with Chinese suppliers could easily be positive for American firms: steel sharpens steel. But in order for American companies to realize those benefits and actually compete, they need the freedom to experiment and iterate so they can truly compete with their Chinese counterparts. We need to go from permission-first to permissionless.

Even if we recalibrate our regulatory policy with an eye toward fostering innovation, we won’t be able to catch up to China in time. Ecosystems can’t be built in a day. But our domestic capacity, when combined with those of allied states like Germany, Japan, and South Korea, can still make up almost the entirety of a robust robotics supply chain, and those weaknesses can be identified and addressed.

As Chinese companies grab market share and displace European and Asian incumbents, our foothold is a wasting asset. The United States and our allies will need to align our interests, identify what we have that works, and let our entrepreneurial businesses get to work.

This is our wakeup call. The leadership of the United States has been bold in defining and defending the AI software stack; now in concert with its most relevant allies it must lay the groundwork for a defensible AI robotics stack. There is time, but not much. The most crucial challenge of the AI race is coming into view. We need to get in the running.

Views expressed in “posts” (including podcasts, videos, and social media) are those of the individual a16z personnel quoted therein and are not the views of a16z Capital Management, L.L.C. (“a16z”) or its respective affiliates. a16z Capital Management is an investment adviser registered with the Securities and Exchange Commission. Registration as an investment adviser does not imply any special skill or training. The posts are not directed to any investors or potential investors, and do not constitute an offer to sell — or a solicitation of an offer to buy — any securities, and may not be used or relied upon in evaluating the merits of any investment.

The contents in here — and available on any associated distribution platforms and any public a16z online social media accounts, platforms, and sites (collectively, “content distribution outlets”) — should not be construed as or relied upon in any manner as investment, legal, tax, or other advice. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Any charts provided here or on a16z content distribution outlets are for informational purposes only, and should not be relied upon when making any investment decision. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. In addition, posts may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein. All content speaks only as of the date indicated.

Under no circumstances should any posts or other information provided on this website — or on associated content distribution outlets — be construed as an offer soliciting the purchase or sale of any security or interest in any pooled investment vehicle sponsored, discussed, or mentioned by a16z personnel. Nor should it be construed as an offer to provide investment advisory services; an offer to invest in an a16z-managed pooled investment vehicle will be made separately and only by means of the confidential offering documents of the specific pooled investment vehicles — which should be read in their entirety, and only to those who, among other requirements, meet certain qualifications under federal securities laws. Such investors, defined as accredited investors and qualified purchasers, are generally deemed capable of evaluating the merits and risks of prospective investments and financial matters.

There can be no assurances that a16z’s investment objectives will be achieved or investment strategies will be successful. Any investment in a vehicle managed by a16z involves a high degree of risk including the risk that the entire amount invested is lost. Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by a16z is available here: https://a16z.com/investments/. Past results of a16z’s investments, pooled investment vehicles, or investment strategies are not necessarily indicative of future results. Excluded from this list are investments (and certain publicly traded cryptocurrencies/ digital assets) for which the issuer has not provided permission for a16z to disclose publicly. As for its investments in any cryptocurrency or token project, a16z is acting in its own financial interest, not necessarily in the interests of other token holders. a16z has no special role in any of these projects or power over their management. a16z does not undertake to continue to have any involvement in these projects other than as an investor and token holder, and other token holders should not expect that it will or rely on it to have any particular involvement.

With respect to funds managed by a16z that are registered in Japan, a16z will provide to any member of the Japanese public a copy of such documents as are required to be made publicly available pursuant to Article 63 of the Financial Instruments and Exchange Act of Japan. Please contact compliance@a16z.com to request such documents.

For other site terms of use, please go here. Additional important information about a16z, including our Form ADV Part 2A Brochure, is available at the SEC’s website: http://www.adviserinfo.sec.gov

While these pieces are broadly penned, a common and particularly puzzling source of the negativity is investors. We’ll never understand why an investor would be comfortable disparaging startups, or offering general pessimism about innovation waves. However, the pieces do tend to get a lot of visibility, and as a result, are a relatively cheap win for clicks. The problem is that they’re quite often wrong, or at best short-sighted, and using them as investment guidance would have resulted in passing on many of the largest winners in tech history.

In this post, we address some of the ill-informed chatter about gross margins in AI app companies. We strongly believe (and history provides ample evidence) that lower gross margins at a moment in time are not a long term indicator of a lack of a sustainable business model. Further, many of the critics completely miss the nuance of today’s AI companies by drawing comparisons to DTC (direct to consumer) subscription companies in prior cycles that suffered from low retention and paid customer acquisition. Those cohorts look dramatically different from today’s AI applications, which deliver deep customer value, strong usage and retention, and explosive net expansion via spreading to teams and the enterprise. We’ll do what we can to correct that as well.

In what follows are some of the common criticisms of app companies built on foundation models.

Claim: AI apps must be permanently low margin at $X/month

A primary fixation of critics is to point out that introductory plans that offer unlimited usage tend to be low gross margin. And yes, that’s generally the case with most product-led companies. However, the standard approach that follows is for the company to introduce a pro plan that will offer limits which then convert to usage based billing. Or they’ll abstract away exactly which model is being used allowing them to direct a subset of traffic to smaller / cheaper models. Many of the leading gen AI apps companies continue to see strong continued growth after this tiering, underscoring that actual product value rather than subsidies drive usage.

Furthermore, it’s important to realize that, in many domains, the cost to serve a query is highly variable. If a subset of the more expensive queries can be identified and serviced by a cheaper model, the margin problem is mollified. Many of the leading app companies we’re aware of are quite sophisticated at routing high value queries to margin optimized inference.

The argument that users simply want the frontier model at all times suggests a blinkered understanding of how these apps are used in practice. Model preference varies by task, and in many use cases where performance differences are limited, a lower TCO (total cost of ownership) is preferred.

Claim: AI app companies’ unprofitable cohorts drive overall business economics indefinitely

For many AI subscription businesses, a small minority of users are driving the bulk of the usage and therefore underlying costs. This means that rate limiting a small percentage of total users (who tend to be loud on X) can meaningfully reduce costs without impacting revenue growth significantly. Even frontier models face this issue, typically seeing limited revenue impact but meaningful cost reduction after introducing rate limits for the top 5% of users.

Moreover, most app companies have profitable segments at higher level tiers (teams, enterprise, etc.) that are not visible in external data. In the AI era, pricing and packaging are more strategic margin levers than ever, and professional users continue to surprise even our more optimistic internal models with their willingness to pay for high value, heavily utilized AI products.

Claim: AI app companies don’t control model pricing and are at the mercy of monopolistic rent seeking

To date, there is no model monopoly. And there is unlikely to be in the foreseeable future, given the level of investment behind the big players and the effectiveness of the open source community. Model drops see massive movements of users, in part because apps are a valid distribution channel and can effectively steer users. Very recently, we’ve seen a massive swing of developers away from Claude to GPT-5 after OpenAI’s recent launch. And while drafting this piece, DeepSeek released its V3.1 API, which appears to again dramatically push the performance cost boundary, likely causing yet more shifts.

Further, there is little indication that the foundation model market will naturally consolidate into a monopoly. Being first to SOTA (state of the art) in a model modality has not guaranteed long-term leadership. There was a nearly year-long period where model providers couldn’t catch up to GPT-4. Midjourney wasn’t the first to image. Google and Kling weren’t first to video. In fact, one could argue leadership positions are historically short for AI models. And in the absence of a dominant position for a sustained period, pricing pressure increases in what has been a savage fight for market share over the past 18 months.

But that’s not the only driver of cost reductions. Inference optimizations are still in early innings, and the trend to date on inference costs is indisputable. Depending on what benchmark you’re looking at, for the same cost per million tokens, inference costs have dropped anywhere from 10x to 100x+ in the last 18 months. While we acknowledge that it’s too simplistic to tie dropping overall cost of inference to increased margins in apps, and we recognize the frontier models are typically more expensive than prior generations, it’s similarly short sighted to assume we’ve reached equilibrium given the ongoing competition and optimization at the model layer. Moreover, given the utilization of multiple models as previously discussed, drops in the cost of prior model generations still drive opportunity for margin expansion.

Claim: Subsidies are clouding true product market fit

At the moment, there is a lot of tired talk about “selling $1 for 50 cents” to describe AI app companies (particularly those touching code) and using VC subsidies to buy behavior from transient users that will immediately dry up when the music stops.

Drawing this conclusion based on short term gross margins is dangerous, as it ignores the more important metrics to gauge whether a free tier or prosumer strategy is working: 1) conversion to paid, 2) conversion to enterprise 3) cohort usage as well as revenue retention and expansion. If the top of funnel that a company is attracting converts to paid, walks the product into enterprises (often with record fast sales cycles), and not only retains but also expands revenue, the strategy is working, full stop.

The playbook is not a new one by any means, but we’re starting to see gen AI companies execute on this successfully, penetrating the enterprise faster than anticipated, and even passing incumbents with incredible distribution. AI companies are bootstrapping their own distribution by building a large user base that fuels their sales funnel, leading to higher-margin teams and enterprise contracts that will account for an increasingly larger share of revenue over time. As anyone who has been in enterprise long enough knows, the top 20% of contracts generally drive 80% of revenue. Having a small segment of revenue where a company trades off margin for pipeline into higher margin segments is a feature, not a bug.

Claim: AI apps are thin and add no value beyond access to the underlying frontier model

Another prevailing misconception is that gen AI apps are predominantly thin wrappers around frontier models. While this may be true of a subset, we’ve also observed that applications have a strategic control point that can quickly lead to differentiation. This happens primarily in two ways: 1) high-velocity teams use the initial entry point to rapidly expand their product surface area, and 2) leading applications compose multiple models (including some of their own) offering a service at either a capability or price point no single model could offer. Both contribute to higher long term margins, yet require winning the user in the near term.

The reality is that models themselves simply aren’t products or platforms. AI code gen companies have created massive value by adding capabilities such as hosting, custom domains, and security scans. High-value (and often high-margin) upsells increase ARPUs and retention.

The rise of product specific models creates room for even further potential differentiation, as products often have far more contextual data on user behavior and preference. As a result, they are in a position to create models, or fine tunes of models, that are uniquely effective in a way no model provider without deep app layer visibility could achieve. Take coding for example,from the vantage point of an API, it’s difficult to know to what extent generated code was accepted. And even if it was accepted, there is no visibility into whether it survived long enough to be submitted as a PR or merged into main. In the constant race to find ever more meaningful sources of data, few are closer to the truth than product data. And as much as the apps teams are training or fine tuning their own models, this is a clear path for model quality differentiation, another potential lever to drive longer term pricing power and cost reduction.

Conclusion

Investors looking for visibility at the expense of startups will continue to write these formulaic and generally ill informed critiques on short term startup margins. There is little we can do about that. But for those looking for a broader view on a complicated topic, we hope this piece adds positively to the discourse. Not all startups will succeed. Low margins will, at times, be a fatal aspect of the business. And of course, systemically low or negative margin businesses should be valued accordingly. That said, margins alone don’t tell the full story. We strongly believe that fixating on margins while ignoring a company’s value to its customers, retention and ease of acquisition misses the mark. Just as important, there are many effective strategies for AI app companies to drive healthy margins in the long term.

Our friend Nat Friedman said it best: “Being a pessimist makes you sound smart. Being an optimist makes you money”.