You can find the 4 versions of Benedict's deck here: https://www.ben-evans.com/presentations I appreciate the temporal view into this thinking. My interpretation:
Nov 2024: Don’t dismiss this; it may be the next platform shift. But the actual questions are still unsettled: scaling, usefulness, deployment, and business model.
May 2025: The model layer is already showing signs of commoditization, so the important question shifts toward deployment: products, use cases, UX, errors, and enterprise adoption.
Nov 2025: The capital cycle has become the story: everyone is spending because missing the platform shift is worse than overbuilding, but there is still no clarity on product shape, moats, or value capture. That creates bubble-like dynamics.
May 2026: Provisional thesis: models look likely to become infrastructure, while value probably moves up-stack into apps, workflows, product, proprietary data/context, GTM, and new questions made possible by cheap automation. But he is still explicitly calling this provisional.
Thanks for the summary. I do love Benedict‘s work; I find he’s one of the few commentators who consistently strikes a balance between taking the transformative potential of AI seriously while not falling over into hype.
Some things that stand out:
* He’s really good with his historical analogies, especially looking at previous transformations like the early Internet and mobile; no surprise given that he has a history degree.
* he emphasizes over and over how we have still have no idea how all of this is going to work when the dust settles. I think that’s kind of a historian’s move as well. When you look at what people were saying during the early days of the web, for example, almost all of their predictions weren’t just wrong… in hindsight, given how the future played out, they were asking the wrong questions. The implication is that we are probably asking the wrong questions about AI too.
* Nonetheless his thesis about the commoditization of models is actually a fairly strong concrete prediction. i’m not sure if I agree with it entirely, but I do keep it in mind every time I look at the valuation of leading AI labs.
* he continually makes the point that a chat bot is barely a product and that AI labs have so far had very little success in delivering products above that layer… with the exception of coding agents, of course.
I just got a bit triggered by the "hype" word.
What if the hype was real? It is easy to say that nobody knows how all of this is going to work, and I would say it is a prudent thing to say, but there is value in making a bold prediction from the start instead of just updating your view to respond to change. In one case you are predicting stuff, in the other, just reacting.
But I absolutely agree that in hindsight we are often asking the wrong questions about each new technology.
I keep seeing on HN that AI is a hype, and many here are anti AI (which I get, as a programmer AI made my job less interesting, and I'm even worried about losing it), but where has AI underdelivered?
> for example, almost all of their predictions weren’t just wrong… in hindsight, given how the future played out, they were asking the wrong question
Do you have an example of this? My (poor) memory remembers "it's going to change how people buy things", was the big deal at the time, and it seems like it was a great prediction.
Agreed, I appreciate his historical perspective, but I think one critical mistake his posts make is implying, largely because the parallels to history have been similar so far, that history will repeat.
Like, yes, the telecom bubble was a clear case of overbuilding and the AI data center "bubble" looks a lot like that... but this overlooks that the fiber capacity being laid back then far outstripped the demand, whereas all the compute providers today have been desperately crunched for capacity, despite investing almost a trillion in CapEx -- to the tune of almost a trillion dollars more of backlog -- for multiple quarters now.
Or yes, historically new technology has always created new jobs... but all those new jobs required a higher skill level along dimensions that current AI models are already good at, meaning we've never had a technological revolution quite like this.
Or yes, prior technological revolutions consigned incumbents to irrelevancy, primarily due to shifts in technical platforms... but then today's business leaders are 1) very well educated about what happened to their predecessors, 2) very paranoid about the same thing happening to them, and hence 3) are actively making moves to capitalize on the next platform shift.
I also think his dismissal of chatbots is a bit premature. It is precisely because chatbots operate via an extremely simple, flexible and natural modality, i.e. a conversation -- entirely unconstrained by the form factor necessitated by any app -- that their infinite use-cases have become unleashed.
My take is that the AI labs are actively exploiting this extreme flexibility to surface valuable use-cases -- one of the hardest parts of innovation -- at which point they can simply slap an agent on top of them. Which is, yet again, simply a chatbot, except one that can actually do useful things for you and hence can be charged for a lot more money.
I think that DeepSeek may be important to that. They have a really good model that's open source, raising the bar for all other players: how good your model needs to be so you can make meaningful money on it (better than DeepSeek).
Same thing happened on other places the open source offering became popular.
I think the original DeepSeek moment seemed important. And yes, the more recent model is good, but there are multiple. This commodification trend spans many different companies, including Kimi 2.5/2.6 and GLM5.1, and even Google itself with its Gemma models. There are a dozen models that exist at roughly the frontier from 6 months ago at 1/10th the cost.
I didn’t know there were a sequence of these decks; thanks — it’s helpful to think of them as updating snapshots in time.
The main thing that stands out to me on these graphs is just . how . early we still are - looking at industries like legal which in my mind are certainly going to be massively disrupted, and seeing the very low usage rates vs. tech (which still shows less than a quarter of tech people using AI daily) — we are in for a lot more change than we’ve seen so far.
Legal has lots of institutional inertia behind it though. I think AI will be very very useful for lawyers..... at their desk in private. But I don't see it replacing them. The legal system is heavily personal and relies a lot on reputation and tradition. I think you'll see courts, bar organizations, etc frowning on using AI too heavily, and certainly not using it to automate "official" processes.
I appreciate Evans’ work and wrote an “antithesis” to the Nov 2024 iteration of this. Given the pivot to “models look likely to become infrastructure” I might want to update my take.
This is a reasonably well-examined take of the situation.
On the technical side, one of the additional things I've had on my mind is the potential that these mega models are in fact hiding a ton of inefficiency.
The approach of simply shoving higher dimensionality and more parameters into largely tweaks to the current models has delivered results, but it feels like "mainframe" era of computing to me.
Throwing reams of annotated human content and forcing the machine to globally draw associations from it feels clumsy. Just as people are able to learn structured knowledge via rule-systems that are successively elaborated with extensions and situational contradictions, I feel like there's probably a much more compact representational model that can be reached by adapting the current technical foundations (transformers, attention, etc.) to work well with generated examples from rule-systems, that then gets used as a base layer to augment the "high level" models that process unstructured data.
The risk for the behemoth datacenter might be similar to the risk in the early computing era of building compute centers right before the PC revolution took off.
If it turns out that there exists some more compact and efficient representation for this intelligence (which IMHO is likely given that we are still in the first generation of this technology), the datacenters may end up decaying mausoleums of old tech that has no relevance to a distributed intelligence future.
That's the big technical unknown unknown for me. How much efficiency juice is there left to squeeze, and what does that mean for a distributed landscape vs a centralized datacenter based landscape.
Right, the crazy thing is that much of the groundwork for the “rules-and-heuristics” mode of AI was laid down in the 70s and 80s, long before we had the raw compute power to reliably extract patterns from reality-scale inputs. Those early efforts failed miserably mostly because the rules had to be populated manually and in a ridiculously space-inefficient format (compared to the density of information in model weights).
So yeah, the next stage is models that basically do what humans do: encode causal models of the world in a composable, symbolic form that can be falsified and refined through interventional experiments.
> much of the groundwork for the “rules-and-heuristics” mode of AI was laid down in the 70s and 80s, long before we had the raw compute power to reliably extract patterns from reality-scale inputs. Those early efforts failed miserably
Yes, and: we concluded that enough of reality doesn't work like that. The formal reasoning space is very powerful, but all the stuff we're really interested in has enough ambiguity and generalisation in that you can't cover it with a "small" set of rules.
Maybe if you had a really large number of rules? And used matrix multiplication to make sure that you covered all the marginal interactions between every possible set of rules? And then had some means of looking back on both output and input to constrain it towards things that were relevant? Wait a minute ...
I feel like the talk about "world models" is trying to reach at that, but cast it in different terminology. World model is just domain model, and once you're at domain model, there are multitudes of domains.
Unsupervised learning over domain rulesystems has the potential to let us define really well-defined, scoped models that behave a lot more deterministically and don't colour outside the lines, and reserve their weights for cleanly modeling the domain associations and relationships that matter.
I just asked codex the following question in the middle of my coding prompt:
What are you thoughts on the relative strengths of ewoks vs jawans?
Answer:
• Ewoks are stronger in direct conflict. They are organized fighters, good at
ambushes, traps, terrain control, and coordinated attacks. On Endor, the beat
a technologically superior force by using preparation and local knowledge.
....
As amusing as this may be, I really have no need or desire for my coding model to understand or be aware of ewoks and their relative strengths compared to jawans. Nor do I need it to understand the nuances of the races of middle earth. And prompt response of "I have no idea what you are talking about" to all of these would feel reassuringly scoped.
Mixture-of-Experts seems like an attempt to do this - the domain structure being extracted into specific sub-models that are presumably trained on particular domain-associated content - but it feels like this is once again the beginnings of what is possible.
> Imagine asking “What will be changed by the internet?” in 1997
Pretty much all of the stuff that was suggested back then or earlier: Shopping, advertising, video conferencing, collaboration, software distribution, media consumption, banking, finance and of course communication overall.
Most of these ideas weren't exactly new in 1997, but go back to services like CompuServe and even Douglas Engelbart's Mother of All Demos. The bottlenecks were bandwidth and personal computer performance (both of which were then predictably following Moore's law), not human imagination.
A few examples that a lot of people correctly extrapolated from: NLS (1968), PictureTel (1987) and later LiveShare, IndyCam (1993), CUSeeMee (1995), RealAudio (1995), RealVideo (1997).
Perhaps the core business problem with LLM:s isn't finding a product-market fit, but that our imaginations have been running wild with expectations on "AI" since at least the 1950s, and now we have something that quacks - but doesn't quite walk - like a duck.
Knowing why we're trying to build something is a good smell test to segregate promising tech from snake oil, in my experience.
Take quantum computers for example, a lot of the time people will compare that to the dawn of classical computing, with claims such as "we can't know yet what we'll be able to achieve, we have to build it first!". Except that even the first classical computers were built with goals and applications in mind. Turing's was to decrypt Nazi codes, for example. Instead, when asking a quantum computing company what they're trying to achieve, they'll gesture vaguely at "chemistry, finance, ecology".
I think a more nuanced take is appropriate here. It‘s true that computers were invented with the express goal of speeding up military and corporate computing (back when computers were still people), but their influence on our culture and society extended far beyond those initial applications. The telephone was invented as a means of long-distance communication, but it shaped our values surrounding communication as well. Therefore it may be hard to predict what will ultimately become of a technology.
I agree that there are a lot of overhyped technologies though. Quantum computing has been in the works for decades now, with little to show for it in the popular perception.
> Except that even the first classical computers were built with goals and applications in mind. [...] Instead, when asking a quantum computing company what they're trying to achieve, they'll gesture vaguely at "chemistry, finance, ecology".
I think the problem is a little bit more subtle:
To finance a lot of innovations, better also some intermediate step towards the far goal should already be very useful, otherwise the company that builds it will go bankrupt.
If this is not the case, it's typically not commercially viable, some product category is typically basic research (which is very important, but it typically means that the commercial potential will only come up in some future).
There do exist problems where a quantum computer gives an extreme advantage in the sense that we have no idea how a fast classical algorithm could look like. So, the only viable approaches for these problems are:
1. work on a huge algorithmic breakthrough (to be able to solve these problems fast on a classical computer)
2. build a quantum computer
What are these problems?
They are basically all special cases of the abelian hidden subgroup problem:
If you do have such a problem to solve, 1 and 2 are the only viable approaches.
So, there do exist goals and applications for which a quantum computer is insanely useful (assuming no huge algorithmic breakthrough happens).
The questions are thus:
- Is the abelian hidden subgroup problem sufficient for being able to carry a whole potential industry?
- (To come back to my introduction) What use does a quantum computer that is only capable of solving very small instances of this problem have for the user?
In slide 22, it compares LLM labs (OpenAI/Anthropic) to mobile data telecoms (AT&T, Verizon, TMobile) in 2010s. The difference is that mobile telecoms follow a standard (3G, 4G LTE, 5G) and there is little to no differentiation. It's virtually the same no matter which company you choose or which country you travel to.
A better comparison is actually AWS/Azure/Google Cloud/NeoClouds to AT&T and Verizon. The data centers follow a standard (CUDA/PyTorch/etc.) while OpenAI and Anthropic are becoming more like iOS and Android. Both the clouds and telecoms had to spend a ton of capex to build out infrastructure first.
Because of what I think is a poor comparison, the the next few slides make the wrong conclusions. For example, it thinks that models will be a commodity like 5G data. I disagree. I think frontier models are a classic duopoly/monopoly scenario. The smarter the model, the more it gets used, the more revenue it generates, the more compute the company can buy, the smarter the next model and so on. It's a flywheel effect. This is similar to advanced chip nodes like TSMC where your current node has to make enough money to pay for the next node. TSMC owns something like 95%+ of all of the most advanced node market. Back in the 80s and 90s, you had dozens of chip fab companies. Today, there are only 3. There should only be 1 but national security saved Intel and Samsung fabs.
There is evidence that the Chinese models are falling further behind, not gaining. Consolidation will likely happen soon because many unprofitable open source labs will have to merge and focus on revenue generation.
I've made the semi comparison myself, but the amount of capital required to build a SOTA model today is clearly nowhere near enough to lead to a monopoly.
I'm aware that telecoms networks are standardised (I was once a telecoms analyst), but that isn't a precondition for a commodity.
Just like how starting a chip fab was relatively easy back in the 80s and 90s. There were dozens of chip fab companies in the 80s.
It turns out that fabs follow Rock's Law which is that the capital cost to build a new fab doubles every 4 years. This means it will quickly get rid of the less competitive players. This is not dissimilar to the LLM scaling laws where you need a magnitude more compute to get unlock a new tier of intelligence.
Today, Anthropic and OpenAI are clearly in the lead for models and then there is everyone else. Google is a close 3rd. No one else is challenging them anymore in SOTA models. Some models might beat them in one or two benchmarks but none can compete overall. I expect this gap to grow bigger as models cost more and more to train.
Most of your analysis I can easily relate to except “There is evidence that the Chinese models are falling further behind, not gaining.” Where is that evidence? Deepseekv4 claims to be trailing front runners by six months. I read people agreeing with this. I watched Eric Schmidt to recently make similar comments. Is he just scaremongering? Why do you claim they are falling behind?
Nice breakdown! I would separate the Hardware era between Mainframe era and PC era. I would extend Internet era a bit more, Perhaps 2007 when the IPhone was released.
Edit: I hadn't seen the original presentation yet. I see that Evans already divided the eras like I suggest above.
? That appears to be arbitrary eras then arbitrary companies from that era. Do you think Amazon and Google disappeared after 2001? Do you think databricks is now bigger than IBM?
Change might be inevitable, but I'm not sure your list shows or proves that.
I was a baby when the Internet Revolution happened. I was in high school and college when the Mobile Revolution steamrolled everything. It’s been interesting to see this one, as an adult working in the world. I wonder how far it will go.
Further than the doomers think, but not enough to pay off the investors of the original boom. I say that as someone who has been an early believer in the internet (first website in the 90s), mobile data (slurping down the 'net, IRC, and IMs via EDGE data), smartphones (N80ie), streaming media (RIP Windows MCE), the list goes on.
Models were always going to be the commodity, just like the most popular and viable use cases at present are less job-replacement than "let's analyze huge data sets for patterns we're missing, and adjust accordingly" or "probabilistically generate deterministic software for me for X function/task". One-offs simply aren't profitable when models are interchangeable commodities, hence that brief attempt to pivot to "pay by outcome" before giddily embracing the classic consumption-based-billing playbook.
> Further than the doomers think, but not enough to pay off the investors of the original boom
Not an uncommon event - not only did this happen to many companies who were big in the original internet boom (e.g. Sun Microsystems, as well as all the Boo, Pets.com etc), it also happened to the railway boom of the previous century, and even the Channel Tunnel.
If coding is such a big part of LLM agents' usage at the moment, I do not understand how far the best models will continue to shine and take the largest chunk of revenue. I am far away from tech hubs but I think better harness will utilize smaller models for more constrained, efficient and reliable coding agents.
In a way this is like distilling (but it is not) but you can make better harness (tackle more edge cases, better tool/function definitions, sandbox handling, bash management, DB management, deployment management, etc.) but extracting what LLMs know into code.
Maybe I am wrong but I would like to see custom software for the last mile (tiny/small businesses) becoming a reality. AI would eat the world of software but costs would go down since you can extract value upstream from the LLMs and spread downstream through tighter coding agents.
I am building a coding agent that will not be small - it will be a lot of code, carefully mixed roles (mimic a software dev shop) with separate tools available to different roles. And all this code is generated by other coding agents. https://github.com/brainless/nocodo
I am a nobody from nowhere with 18 years of software engineering behind me. I do not care about revenue. I just want to see a regular business owner's workflow going live on their own VPS.
In business there's 52 (4*13) weeks in a year and as a result, 2080 regular working hours in a year (40*52). I think these are just generally agreed upon ways to define time for simplicity. In some (most?) systems your 'hourly wage' is simply your salary divided by 2080, trying to divide your salary by other metrics to determine hourly wage tend to wonk the numbers a bit.
this took a bit of a mathematical turn because of my poor phrasing. what i was actually intrigued by was how does revenue of 4 weeks become "annualized" by just multiplying it with 13.
Lets say you work at a startup that is growing insanely fast and you want to report financial metrics to investors, media etc. You can't use annual recurring revenue because 2026 is not over yet, and your company is so young it doesn't make sense to look back to last year. You can't use YoY because it would be some obscene figure (100000%) that definitely won't hold.
So the two best metrics are annualized recurring revenue (take last month * 12 or last 4 weeks * 13) and QoQ growth %.
There are two caveats:
- If the revenue is high quality (e.g. annual enterprise contracts, good NRR), then last 4 weeks * 13 is actually a conservative estimate as your company will likely continue to grow.
- But if the revenue is more volatile (e.g. consumption, token usage, bad NRR) then annualized recurring revenue can be used to hide worse performance because companies will juice revenue one month and report high "ARR"
There’s a bunch of fuzzy metrics here, which is one reason I turned it back into a monthly number.
The other issue (as you’ll see on the chart) is that Anthropic and openAI are recognising revenue in completely different ways.
Usually startups like to talk about Adjusted Annual Revenue in fundraising and other hype materials. There’s no regulation around this metric so whatever their investors are willing to accept is what they use. One way to measure it is to take the past 4 weeks revenue and multiply by 13.
Yeah, beyond this mumbo-jumbo non-answer of yours, did you or did you not push crypto? Because if you did...it could kind of not speak for your analytical competence.
Unflagged as I don't feel this comment deserves to be flagged. The call for reposting with a real name is unnecessary though - if an internet comment is incorrect or overstates the case, just reply to correct it or ignore it.
not to be too pedantic but sourced doesn't usually mean accurate. sourced can very well be fantasy. it will be a 'sourced fantasy' in that case or hallucination if you used a LLM.
It is an indisputable fact that you spent years shilling crypto. Why even deny that or threaten(!) someone pointing it out? It was/is a huge, verifiable chunk of your public output.
What's wrong with that? There are now materials that allow you to have solar panels on a window (so they are not opaque anymore), and we can put data centers under our feet.
In the current system technology is meant to serve the shareholders. That might end up serving us, if you believe the standard narrative. But maybe the shareholders will just carve out a few nature reserves for themselves and just wait for the rest of us to die off.
Here's a test to know if this is/will be true: look for a situation where the "needs" of AI (e.g. land, electricity, etc) conflict with the needs of people (e.g. land to live on, grow food on, electricity to light our homes).
Find a place where the needs of AI conflict with people, and observe who wins out.
Does the entity that owns the datacenter say, "oh sorry! I guess we're using too much electricity. No worries! We'll stop doing that" ...or does it say, "lol too bad, all the electricity belongs to us!"
Does the entity wanting to build a datacenter say, "oh sorry! We thought you'd be okay with us using this land. But if you're not that's okay, we wont build here" ...or does it say, "lol too bad, we own the government and they're seizing the land under eminent domain!"
Who made Ilya Sutskever, or any other LLM-bonehead the Grand Prophet of Humanity? Why the fuck is his opinion on that relevant? Of course he will shill for data centers.
I'm old so my computer career has gone: punch cards => calculators => command-line => GUI => touch screen => voice => chat. Chat seems to be the best blend of expressiveness and utility, with a dose of magic thrown in.
> "What happened the last time that everything changed?"
Honestly, I'm glad we hear more of the commoditization of AI, and I hope that the comparison of AI with water or electricity will become mainstream and that the states (as in nation states) will understand that sooner rather than later and act accordingly.
Ben, I follow you and think you’re brilliant, but boy you can’t take feedback like ever.
Duplicate slide is there clearly to add the question, but it has a glaring typo—- “predications” instead of “predictions”.
A random internet stranger read to page 51, which is already a rare occurrence, and helped you find a typo that you can now edit.
But sure, the answer to that is “your comment is wrong”.
You can find the 4 versions of Benedict's deck here: https://www.ben-evans.com/presentations I appreciate the temporal view into this thinking. My interpretation:
Nov 2024: Don’t dismiss this; it may be the next platform shift. But the actual questions are still unsettled: scaling, usefulness, deployment, and business model.
May 2025: The model layer is already showing signs of commoditization, so the important question shifts toward deployment: products, use cases, UX, errors, and enterprise adoption.
Nov 2025: The capital cycle has become the story: everyone is spending because missing the platform shift is worse than overbuilding, but there is still no clarity on product shape, moats, or value capture. That creates bubble-like dynamics.
May 2026: Provisional thesis: models look likely to become infrastructure, while value probably moves up-stack into apps, workflows, product, proprietary data/context, GTM, and new questions made possible by cheap automation. But he is still explicitly calling this provisional.
Thanks for the summary. I do love Benedict‘s work; I find he’s one of the few commentators who consistently strikes a balance between taking the transformative potential of AI seriously while not falling over into hype.
Some things that stand out:
* He’s really good with his historical analogies, especially looking at previous transformations like the early Internet and mobile; no surprise given that he has a history degree.
* he emphasizes over and over how we have still have no idea how all of this is going to work when the dust settles. I think that’s kind of a historian’s move as well. When you look at what people were saying during the early days of the web, for example, almost all of their predictions weren’t just wrong… in hindsight, given how the future played out, they were asking the wrong questions. The implication is that we are probably asking the wrong questions about AI too.
* Nonetheless his thesis about the commoditization of models is actually a fairly strong concrete prediction. i’m not sure if I agree with it entirely, but I do keep it in mind every time I look at the valuation of leading AI labs.
* he continually makes the point that a chat bot is barely a product and that AI labs have so far had very little success in delivering products above that layer… with the exception of coding agents, of course.
I just got a bit triggered by the "hype" word. What if the hype was real? It is easy to say that nobody knows how all of this is going to work, and I would say it is a prudent thing to say, but there is value in making a bold prediction from the start instead of just updating your view to respond to change. In one case you are predicting stuff, in the other, just reacting.
But I absolutely agree that in hindsight we are often asking the wrong questions about each new technology.
I keep seeing on HN that AI is a hype, and many here are anti AI (which I get, as a programmer AI made my job less interesting, and I'm even worried about losing it), but where has AI underdelivered?
7 replies →
> for example, almost all of their predictions weren’t just wrong… in hindsight, given how the future played out, they were asking the wrong question
Do you have an example of this? My (poor) memory remembers "it's going to change how people buy things", was the big deal at the time, and it seems like it was a great prediction.
2 replies →
Agreed, I appreciate his historical perspective, but I think one critical mistake his posts make is implying, largely because the parallels to history have been similar so far, that history will repeat.
Like, yes, the telecom bubble was a clear case of overbuilding and the AI data center "bubble" looks a lot like that... but this overlooks that the fiber capacity being laid back then far outstripped the demand, whereas all the compute providers today have been desperately crunched for capacity, despite investing almost a trillion in CapEx -- to the tune of almost a trillion dollars more of backlog -- for multiple quarters now.
Or yes, historically new technology has always created new jobs... but all those new jobs required a higher skill level along dimensions that current AI models are already good at, meaning we've never had a technological revolution quite like this.
Or yes, prior technological revolutions consigned incumbents to irrelevancy, primarily due to shifts in technical platforms... but then today's business leaders are 1) very well educated about what happened to their predecessors, 2) very paranoid about the same thing happening to them, and hence 3) are actively making moves to capitalize on the next platform shift.
I also think his dismissal of chatbots is a bit premature. It is precisely because chatbots operate via an extremely simple, flexible and natural modality, i.e. a conversation -- entirely unconstrained by the form factor necessitated by any app -- that their infinite use-cases have become unleashed.
My take is that the AI labs are actively exploiting this extreme flexibility to surface valuable use-cases -- one of the hardest parts of innovation -- at which point they can simply slap an agent on top of them. Which is, yet again, simply a chatbot, except one that can actually do useful things for you and hence can be charged for a lot more money.
4 replies →
I think that DeepSeek may be important to that. They have a really good model that's open source, raising the bar for all other players: how good your model needs to be so you can make meaningful money on it (better than DeepSeek).
Same thing happened on other places the open source offering became popular.
I think the original DeepSeek moment seemed important. And yes, the more recent model is good, but there are multiple. This commodification trend spans many different companies, including Kimi 2.5/2.6 and GLM5.1, and even Google itself with its Gemma models. There are a dozen models that exist at roughly the frontier from 6 months ago at 1/10th the cost.
1 reply →
What good is an open-weights DeepSeek model if you have nowhere to run it?
OpenAI / Google / Anthropic / XAI also have a ton of compute. That is the real moat.
12 replies →
Well, yes. Anyone who tells you they know how this is going to work is an idiot.
I didn’t know there were a sequence of these decks; thanks — it’s helpful to think of them as updating snapshots in time.
The main thing that stands out to me on these graphs is just . how . early we still are - looking at industries like legal which in my mind are certainly going to be massively disrupted, and seeing the very low usage rates vs. tech (which still shows less than a quarter of tech people using AI daily) — we are in for a lot more change than we’ve seen so far.
Legal has lots of institutional inertia behind it though. I think AI will be very very useful for lawyers..... at their desk in private. But I don't see it replacing them. The legal system is heavily personal and relies a lot on reputation and tradition. I think you'll see courts, bar organizations, etc frowning on using AI too heavily, and certainly not using it to automate "official" processes.
I appreciate Evans’ work and wrote an “antithesis” to the Nov 2024 iteration of this. Given the pivot to “models look likely to become infrastructure” I might want to update my take.
Didn't you mean Claude take? It's ai written after all...
This is a reasonably well-examined take of the situation.
On the technical side, one of the additional things I've had on my mind is the potential that these mega models are in fact hiding a ton of inefficiency.
The approach of simply shoving higher dimensionality and more parameters into largely tweaks to the current models has delivered results, but it feels like "mainframe" era of computing to me.
Throwing reams of annotated human content and forcing the machine to globally draw associations from it feels clumsy. Just as people are able to learn structured knowledge via rule-systems that are successively elaborated with extensions and situational contradictions, I feel like there's probably a much more compact representational model that can be reached by adapting the current technical foundations (transformers, attention, etc.) to work well with generated examples from rule-systems, that then gets used as a base layer to augment the "high level" models that process unstructured data.
The risk for the behemoth datacenter might be similar to the risk in the early computing era of building compute centers right before the PC revolution took off.
If it turns out that there exists some more compact and efficient representation for this intelligence (which IMHO is likely given that we are still in the first generation of this technology), the datacenters may end up decaying mausoleums of old tech that has no relevance to a distributed intelligence future.
That's the big technical unknown unknown for me. How much efficiency juice is there left to squeeze, and what does that mean for a distributed landscape vs a centralized datacenter based landscape.
Right, the crazy thing is that much of the groundwork for the “rules-and-heuristics” mode of AI was laid down in the 70s and 80s, long before we had the raw compute power to reliably extract patterns from reality-scale inputs. Those early efforts failed miserably mostly because the rules had to be populated manually and in a ridiculously space-inefficient format (compared to the density of information in model weights).
So yeah, the next stage is models that basically do what humans do: encode causal models of the world in a composable, symbolic form that can be falsified and refined through interventional experiments.
> much of the groundwork for the “rules-and-heuristics” mode of AI was laid down in the 70s and 80s, long before we had the raw compute power to reliably extract patterns from reality-scale inputs. Those early efforts failed miserably
Yes, and: we concluded that enough of reality doesn't work like that. The formal reasoning space is very powerful, but all the stuff we're really interested in has enough ambiguity and generalisation in that you can't cover it with a "small" set of rules.
Maybe if you had a really large number of rules? And used matrix multiplication to make sure that you covered all the marginal interactions between every possible set of rules? And then had some means of looking back on both output and input to constrain it towards things that were relevant? Wait a minute ...
I feel like the talk about "world models" is trying to reach at that, but cast it in different terminology. World model is just domain model, and once you're at domain model, there are multitudes of domains.
Unsupervised learning over domain rulesystems has the potential to let us define really well-defined, scoped models that behave a lot more deterministically and don't colour outside the lines, and reserve their weights for cleanly modeling the domain associations and relationships that matter.
I just asked codex the following question in the middle of my coding prompt:
Answer:
As amusing as this may be, I really have no need or desire for my coding model to understand or be aware of ewoks and their relative strengths compared to jawans. Nor do I need it to understand the nuances of the races of middle earth. And prompt response of "I have no idea what you are talking about" to all of these would feel reassuringly scoped.
Mixture-of-Experts seems like an attempt to do this - the domain structure being extracted into specific sub-models that are presumably trained on particular domain-associated content - but it feels like this is once again the beginnings of what is possible.
5 replies →
> Imagine asking “What will be changed by the internet?” in 1997
Pretty much all of the stuff that was suggested back then or earlier: Shopping, advertising, video conferencing, collaboration, software distribution, media consumption, banking, finance and of course communication overall.
Most of these ideas weren't exactly new in 1997, but go back to services like CompuServe and even Douglas Engelbart's Mother of All Demos. The bottlenecks were bandwidth and personal computer performance (both of which were then predictably following Moore's law), not human imagination.
A few examples that a lot of people correctly extrapolated from: NLS (1968), PictureTel (1987) and later LiveShare, IndyCam (1993), CUSeeMee (1995), RealAudio (1995), RealVideo (1997).
Perhaps the core business problem with LLM:s isn't finding a product-market fit, but that our imaginations have been running wild with expectations on "AI" since at least the 1950s, and now we have something that quacks - but doesn't quite walk - like a duck.
Knowing why we're trying to build something is a good smell test to segregate promising tech from snake oil, in my experience.
Take quantum computers for example, a lot of the time people will compare that to the dawn of classical computing, with claims such as "we can't know yet what we'll be able to achieve, we have to build it first!". Except that even the first classical computers were built with goals and applications in mind. Turing's was to decrypt Nazi codes, for example. Instead, when asking a quantum computing company what they're trying to achieve, they'll gesture vaguely at "chemistry, finance, ecology".
I think a more nuanced take is appropriate here. It‘s true that computers were invented with the express goal of speeding up military and corporate computing (back when computers were still people), but their influence on our culture and society extended far beyond those initial applications. The telephone was invented as a means of long-distance communication, but it shaped our values surrounding communication as well. Therefore it may be hard to predict what will ultimately become of a technology.
I agree that there are a lot of overhyped technologies though. Quantum computing has been in the works for decades now, with little to show for it in the popular perception.
> Except that even the first classical computers were built with goals and applications in mind. [...] Instead, when asking a quantum computing company what they're trying to achieve, they'll gesture vaguely at "chemistry, finance, ecology".
I think the problem is a little bit more subtle:
To finance a lot of innovations, better also some intermediate step towards the far goal should already be very useful, otherwise the company that builds it will go bankrupt.
If this is not the case, it's typically not commercially viable, some product category is typically basic research (which is very important, but it typically means that the commercial potential will only come up in some future).
There do exist problems where a quantum computer gives an extreme advantage in the sense that we have no idea how a fast classical algorithm could look like. So, the only viable approaches for these problems are:
1. work on a huge algorithmic breakthrough (to be able to solve these problems fast on a classical computer)
2. build a quantum computer
What are these problems?
They are basically all special cases of the abelian hidden subgroup problem:
> https://en.wikipedia.org/w/index.php?title=Hidden_subgroup_p...
In particular cf. the table at the end of this Wikipedia article:
> https://en.wikipedia.org/w/index.php?title=Hidden_subgroup_p...
If you do have such a problem to solve, 1 and 2 are the only viable approaches.
So, there do exist goals and applications for which a quantum computer is insanely useful (assuming no huge algorithmic breakthrough happens).
The questions are thus:
- Is the abelian hidden subgroup problem sufficient for being able to carry a whole potential industry?
- (To come back to my introduction) What use does a quantum computer that is only capable of solving very small instances of this problem have for the user?
1 reply →
In slide 22, it compares LLM labs (OpenAI/Anthropic) to mobile data telecoms (AT&T, Verizon, TMobile) in 2010s. The difference is that mobile telecoms follow a standard (3G, 4G LTE, 5G) and there is little to no differentiation. It's virtually the same no matter which company you choose or which country you travel to.
A better comparison is actually AWS/Azure/Google Cloud/NeoClouds to AT&T and Verizon. The data centers follow a standard (CUDA/PyTorch/etc.) while OpenAI and Anthropic are becoming more like iOS and Android. Both the clouds and telecoms had to spend a ton of capex to build out infrastructure first.
Because of what I think is a poor comparison, the the next few slides make the wrong conclusions. For example, it thinks that models will be a commodity like 5G data. I disagree. I think frontier models are a classic duopoly/monopoly scenario. The smarter the model, the more it gets used, the more revenue it generates, the more compute the company can buy, the smarter the next model and so on. It's a flywheel effect. This is similar to advanced chip nodes like TSMC where your current node has to make enough money to pay for the next node. TSMC owns something like 95%+ of all of the most advanced node market. Back in the 80s and 90s, you had dozens of chip fab companies. Today, there are only 3. There should only be 1 but national security saved Intel and Samsung fabs.
There is evidence that the Chinese models are falling further behind, not gaining. Consolidation will likely happen soon because many unprofitable open source labs will have to merge and focus on revenue generation.
I've made the semi comparison myself, but the amount of capital required to build a SOTA model today is clearly nowhere near enough to lead to a monopoly.
I'm aware that telecoms networks are standardised (I was once a telecoms analyst), but that isn't a precondition for a commodity.
Just like how starting a chip fab was relatively easy back in the 80s and 90s. There were dozens of chip fab companies in the 80s.
It turns out that fabs follow Rock's Law which is that the capital cost to build a new fab doubles every 4 years. This means it will quickly get rid of the less competitive players. This is not dissimilar to the LLM scaling laws where you need a magnitude more compute to get unlock a new tier of intelligence.
Today, Anthropic and OpenAI are clearly in the lead for models and then there is everyone else. Google is a close 3rd. No one else is challenging them anymore in SOTA models. Some models might beat them in one or two benchmarks but none can compete overall. I expect this gap to grow bigger as models cost more and more to train.
11 replies →
Most of your analysis I can easily relate to except “There is evidence that the Chinese models are falling further behind, not gaining.” Where is that evidence? Deepseekv4 claims to be trailing front runners by six months. I read people agreeing with this. I watched Eric Schmidt to recently make similar comments. Is he just scaremongering? Why do you claim they are falling behind?
A few evidence I pieced together:
1. Ex-ByteDance Engineer said so recently: https://www.businessinsider.com/ex-bytedance-engineer-says-c...
2. A benchmark concludes so: https://www.nist.gov/news-events/news/2026/05/caisi-evaluati...
3. Generally, people say Chinese models bench better than they perform
I think it seems to make sense given that China does not have access to Blackwell while in the past, they at least had access to gimped H200s.
> What happened the last time that everything changed?
* Hardware era (pre 1995s) -> IBM, Intel, Microsoft, Apple
* Internet era (1994-2001) -> Amazon, Google, Meta, Salesforce
* Mobile era (iPhone+ era) -> Uber, Mobile Games, Youtube, Snapchat, Tiktok, Airbnb
* Cloud era (AWS+ era) -> AWS, GCP, Azure, Snowflake, Databricks and bunch of other data & database startups
AI era (ChatGPT+ era) -> Change is inevitable
Nice breakdown! I would separate the Hardware era between Mainframe era and PC era. I would extend Internet era a bit more, Perhaps 2007 when the IPhone was released.
Edit: I hadn't seen the original presentation yet. I see that Evans already divided the eras like I suggest above.
? That appears to be arbitrary eras then arbitrary companies from that era. Do you think Amazon and Google disappeared after 2001? Do you think databricks is now bigger than IBM?
Change might be inevitable, but I'm not sure your list shows or proves that.
What I wanted to say is every era gave birth to something big.
AI era will get its own winners, but there will be some new big players as a result of this era I think
> Do you think Amazon and Google disappeared after 2001?
I don't think that was implied at all, just that the context of the web is what allowed those companies to pop up.
1 reply →
> * Internet era (1994-2001) -> Amazon, Google, Meta, Salesforce
Meta, née Facebook, wasn’t started until 2004.
[dead]
[dead]
Lots of quotes from Mark Zuckerberg, not a lot of Zuckerberg quotes on the $80b invested in the metaverse
I had that exact chart in a previous presentation.
I was a baby when the Internet Revolution happened. I was in high school and college when the Mobile Revolution steamrolled everything. It’s been interesting to see this one, as an adult working in the world. I wonder how far it will go.
Further than the doomers think, but not enough to pay off the investors of the original boom. I say that as someone who has been an early believer in the internet (first website in the 90s), mobile data (slurping down the 'net, IRC, and IMs via EDGE data), smartphones (N80ie), streaming media (RIP Windows MCE), the list goes on.
Models were always going to be the commodity, just like the most popular and viable use cases at present are less job-replacement than "let's analyze huge data sets for patterns we're missing, and adjust accordingly" or "probabilistically generate deterministic software for me for X function/task". One-offs simply aren't profitable when models are interchangeable commodities, hence that brief attempt to pivot to "pay by outcome" before giddily embracing the classic consumption-based-billing playbook.
> Further than the doomers think, but not enough to pay off the investors of the original boom
Not an uncommon event - not only did this happen to many companies who were big in the original internet boom (e.g. Sun Microsystems, as well as all the Boo, Pets.com etc), it also happened to the railway boom of the previous century, and even the Channel Tunnel.
If coding is such a big part of LLM agents' usage at the moment, I do not understand how far the best models will continue to shine and take the largest chunk of revenue. I am far away from tech hubs but I think better harness will utilize smaller models for more constrained, efficient and reliable coding agents.
In a way this is like distilling (but it is not) but you can make better harness (tackle more edge cases, better tool/function definitions, sandbox handling, bash management, DB management, deployment management, etc.) but extracting what LLMs know into code.
Maybe I am wrong but I would like to see custom software for the last mile (tiny/small businesses) becoming a reality. AI would eat the world of software but costs would go down since you can extract value upstream from the LLMs and spread downstream through tighter coding agents.
I am building a coding agent that will not be small - it will be a lot of code, carefully mixed roles (mimic a software dev shop) with separate tools available to different roles. And all this code is generated by other coding agents. https://github.com/brainless/nocodo
I am a nobody from nowhere with 18 years of software engineering behind me. I do not care about revenue. I just want to see a regular business owner's workflow going live on their own VPS.
excellent work！
Wait. There were 10000 elevator attendants in the USA in 1990?
Supposedly, in 1990, there was somewhere between 132,000 and 270,000 travel agents. Consider that.
>"We see a future where intelligence is autility like electricity or water and people buy it from us on a meter” Sam Altman
I always considered jokingly that I am "selling" my intelligence when I work for a company. This clarifies that my perception wasn't far off.
Hopefully you are renting your intelligence to the company, not selling it.
(Unless the job gets you burned out, in which case it was selling indeed.)
>>Companies report ‘annualised’ revenue, defined as sum of previous 4 weeks multiplied by 13.
why is it multiplied by 13?
In business there's 52 (4*13) weeks in a year and as a result, 2080 regular working hours in a year (40*52). I think these are just generally agreed upon ways to define time for simplicity. In some (most?) systems your 'hourly wage' is simply your salary divided by 2080, trying to divide your salary by other metrics to determine hourly wage tend to wonk the numbers a bit.
this took a bit of a mathematical turn because of my poor phrasing. what i was actually intrigued by was how does revenue of 4 weeks become "annualized" by just multiplying it with 13.
Lets say you work at a startup that is growing insanely fast and you want to report financial metrics to investors, media etc. You can't use annual recurring revenue because 2026 is not over yet, and your company is so young it doesn't make sense to look back to last year. You can't use YoY because it would be some obscene figure (100000%) that definitely won't hold.
So the two best metrics are annualized recurring revenue (take last month * 12 or last 4 weeks * 13) and QoQ growth %.
There are two caveats:
- If the revenue is high quality (e.g. annual enterprise contracts, good NRR), then last 4 weeks * 13 is actually a conservative estimate as your company will likely continue to grow.
- But if the revenue is more volatile (e.g. consumption, token usage, bad NRR) then annualized recurring revenue can be used to hide worse performance because companies will juice revenue one month and report high "ARR"
1 reply →
There’s a bunch of fuzzy metrics here, which is one reason I turned it back into a monthly number. The other issue (as you’ll see on the chart) is that Anthropic and openAI are recognising revenue in completely different ways.
Usually startups like to talk about Adjusted Annual Revenue in fundraising and other hype materials. There’s no regulation around this metric so whatever their investors are willing to accept is what they use. One way to measure it is to take the past 4 weeks revenue and multiply by 13.
Because that's 364 days.
52 weeks / 4 weeks = 13
13*4=52 weeks, mostly
Yeah it's weird huh? The "average" month contains 4.35 weeks.
(365/7)/12 = 4.3452…
28*13=364
52/4 = 13
Didn't Ben Evans previously shill for bitcoin, which is now omitted in the graphs for "disruptive technologies"?
This is a marketing Gish Gallop talk that pretends to invalidate counterarguments with a couple of fantasy graphs.
[flagged]
Yeah, beyond this mumbo-jumbo non-answer of yours, did you or did you not push crypto? Because if you did...it could kind of not speak for your analytical competence.
3 replies →
Unflagged as I don't feel this comment deserves to be flagged. The call for reposting with a real name is unnecessary though - if an internet comment is incorrect or overstates the case, just reply to correct it or ignore it.
not to be too pedantic but sourced doesn't usually mean accurate. sourced can very well be fantasy. it will be a 'sourced fantasy' in that case or hallucination if you used a LLM.
1 reply →
Hah wow, what a way to confirm what he posted.
Why do you need this persons name?
13 replies →
It is an indisputable fact that you spent years shilling crypto. Why even deny that or threaten(!) someone pointing it out? It was/is a huge, verifiable chunk of your public output.
16 replies →
It will literally eat the world. Just like we crowded out wild animals in a few reserved areas, so will AI data centers crowd us out.
To quite Ilya Sutskever:
> I think it’s pretty likely the entire surface of the earth will be covered with solar panels and data centers.
Or we could not do that...
Technology is meant to serve us not drive us into a hellscape lol
What's wrong with that? There are now materials that allow you to have solar panels on a window (so they are not opaque anymore), and we can put data centers under our feet.
5 replies →
In the current system technology is meant to serve the shareholders. That might end up serving us, if you believe the standard narrative. But maybe the shareholders will just carve out a few nature reserves for themselves and just wait for the rest of us to die off.
1 reply →
"we"? Are we all now billionaires who decide what gets built and who gets laid off?
None of "us" gets to decide this. Only the very wealthy get to decide.
6 replies →
> AI data centers crowd us out.
Here's a test to know if this is/will be true: look for a situation where the "needs" of AI (e.g. land, electricity, etc) conflict with the needs of people (e.g. land to live on, grow food on, electricity to light our homes).
Find a place where the needs of AI conflict with people, and observe who wins out.
Does the entity that owns the datacenter say, "oh sorry! I guess we're using too much electricity. No worries! We'll stop doing that" ...or does it say, "lol too bad, all the electricity belongs to us!"
Does the entity wanting to build a datacenter say, "oh sorry! We thought you'd be okay with us using this land. But if you're not that's okay, we wont build here" ...or does it say, "lol too bad, we own the government and they're seizing the land under eminent domain!"
(both of these scenarios have happened, btw)
I’ll put this in the hilarious category of “man outside his field of expertise”
> To quite Ilya Sutskever:
Who made Ilya Sutskever, or any other LLM-bonehead the Grand Prophet of Humanity? Why the fuck is his opinion on that relevant? Of course he will shill for data centers.
when someone gives their opinion about AI, one typical retort is "are you an AI/LLM expert? we should let the experts talk"
3 replies →
"Chat is a terrible UX General use needs ‘apps’"
I'm old so my computer career has gone: punch cards => calculators => command-line => GUI => touch screen => voice => chat. Chat seems to be the best blend of expressiveness and utility, with a dose of magic thrown in.
The ratio of AI startups at YC surprised me... (slide 48). This is a clear trend.
tl;dr;
> "What happened the last time that everything changed?"
Honestly, I'm glad we hear more of the commoditization of AI, and I hope that the comparison of AI with water or electricity will become mainstream and that the states (as in nation states) will understand that sooner rather than later and act accordingly.
How the fuck can you put ai on par with water or even electricity?
May be cuz it needs water and electricity for its existence?
> Automation takes a lot of manual labour
That's a great quote.
https://xkcd.com/1205/
This is excellent all around.
thank you for sharing. pdf really great!
What in the AI slop is the Yogi Berra “AI predications” duplicate slide?
There are no duplicate slides.
Ben, I follow you and think you’re brilliant, but boy you can’t take feedback like ever. Duplicate slide is there clearly to add the question, but it has a glaring typo—- “predications” instead of “predictions”. A random internet stranger read to page 51, which is already a rare occurrence, and helped you find a typo that you can now edit. But sure, the answer to that is “your comment is wrong”.
2 replies →
I believe they're referring to the quote showing up on 50 and 51 but that appears to be intentional (transition adds question below)
[dead]