Transcription
Today on the AI daily brief, the state of AI mid 2025. Welcome back to the AI daily brief and to discuss our show a little bit. It is a weekend, which means a long read/big big ideas episode. And today, we are combining a couple of recent presentations to get a sense of where we are at this halfway point of the year. We're also, it's worth noting, halfway through the first half of the century. We are officially closer to 2050 than we are to the year 2000. And there have been two big presentations recently that provide really interesting context to digging into exactly where we are with AI.
The first is this massive Mary Maker style presentation from investing giant Kowatu. Obviously, that's the one that's going to be a much more macro look. And then Menllo also released their State of Consumer AI 2025. So, the order we're going to take this in is private market trends first, public market trends second, and state of consumer AI last.
The TL;DR on the story of private markets is that they are all about AI. Now, this was a trend last year, but AI is over 50% of funding that's happened so far in 2025. Maybe a more significant part of the story, however, is the likelihood that private funding increases as exits actually rebound. You can see here from this chart, and by the way, this is a good moment to remind you that if you are a listener primarily, this is one that might be worth watching given the highly visual nature of some of these charts and slides. But if you look at this chart from a peak of exits in the height of the Zerco COVID era in 2021, we have been in an exit desert for the last few years, and that has caused some serious liquidity issues when it comes to funding and capitalization.
Now, however, we are seeing the very beginnings of the return of an IPO window. We're seeing more M&A activity, and we are also, although it's not really reflected yet in these charts, starting to see the return of SPACs as a path to going public as well. This is good news for investors because these companies are growing incredibly quickly. The example that Kowatu gives is Anthropic. It took that company around 21 months to go from 0 to a billion in annualized revenue. Then it took 3 months to go from 1 billion to 2 billion. 2 months to go from 2 billion to 3 billion. And actually, this slide is out of date because it only took an additional month to go from 3 billion to 4 billion. Now, this is of course an extremely notable example because of the rise of AI coding and coding assistants that are so driven by Anthropic's Claude models. But still, these companies are getting big incredibly fast. And the reality is that the better that these exits do, for example, a generally positive core IPO and some significant growth in M&A functions, the more capital is going to be unlocked. All of which is to say, everything continues to point in the direction of more capital for more AI spend across a variety of different themes. And for now, at least, the action remains in private markets.
The total market cap of the top 10 public AI companies grew just 10% between June of last year and June of this year, from $18 to $20 trillion, where the market cap of the top 10 private AI companies grew 130% from $283 billion to $658 billion. They sum up: in AI, the action is in the private markets. $10 billion of funding has become the norm for new model startups. Time to $10 million of revenue has accelerated from 10 years to 12 months. We are in a bonanza. Now, they do specifically call out the coding use case as one true inflection point. They show that there has been $1.3 billion of net new revenue generated in just a single year around vibe coding and AI co-pilot tools. And that trend just seems to be accelerating now across both private and public markets.
Acceleration seems to be a big theme. Kowatu is calling this the age of reasoning and basically points out how, in a number of different dimensions, token consumption has gone way up since the release of the reasoning models in Q4 of last year. In public markets, they point to Microsoft as an example of this phenomenon where, in their April 30th earnings call, Sachi Nadella said, "We processed over 100 trillion tokens this quarter, up 5x year-over-year, including a record 50 trillion tokens last month alone." Another example of this massive growth post-reasoning models is the share of US businesses with paid subscriptions to AI, according to Ramp estimates, which jumped from a little over 25% in Q4 of last year all the way up to 42% in Q1 of this year. They called it a positive inflection as reasoning models emerge. ChatGPT saw a similar explosion of growth after the launch of Deep Research, their reasoning model, and their latest image generation model.
Ultimately, as much as there have been a few scares in AI, which they identify as the AI ROI scare from July of last year (you'll remember the Goldman Sachs report that I talk about all the time), the Deep Seek scare in January of this year, and what they call the Microsoft Capex scare of February of this year, the reality, they argue, is that this is an AI super cycle. And one thing that they're not sure of is how the super cycle is going to reshape who's at the top of the market. They spend a lot of time wondering if the Magnificent 7's reign is coming to an end. Four of the MAG 7 are negative year-to-date, and they wonder if some of these companies, at least, will give way to new AI leaders. They group AI stocks into three categories: AI power, AI software, and AI semis. Semis have grown by 12%, with examples being Broadcom and TSMC, who are being driven by token growth's new inflection in computer demand. AI software, like Palantir, Oracle, and Snowflake, who are being driven by AI agents taking off with the new reasoning models. And AI power, like GE Vernova, Constellation, and Vistra, which is up 18% on the year, and who are being driven by electricity shortages and long-term deals with hyperscalers. Compare those 12%, 17%, and 18% year-to-date performances to the overall negative 1% performance for the MAG 7.
What could hurt the party? They identify three risks: the market being too expensive, tariffs leading to the end of US exceptionalism, and ballooning deficits. However, they point out that there is a difference between a crisis and a correction, and that right now there are a lot of tailwinds. Two examples they point to are the end of the AI diffusion rule, moving from tight export restrictions on chips to broader access for allies, indicated by $250 billion to $1 trillion in AI-related investment commitments from the Gulf States, and the energy and nuclear renaissance, as they call it, from no new builds and policy paralysis to major buildup of capacity. The broad shift they see is from autos today to semiconductors tomorrow, from oil today to electricity tomorrow, from manufacturing factories to AI factories, and from being a global leader of the industrial revolution to a global leader of the AI revolution.
Ultimately, they believe that we could see a virtuous flywheel set off by AI. AI productivity gains turn into lower unit labor costs, turn into lower inflation, lower interest rates, higher GDP growth, and that leads to higher tax receipts, lower debt-to-GDP ratio, ultimately circulating in a productivity deficit cycle that could redefine the economy in a positive way. The big impacts from AI on the economy they see is AI's productivity gains translating into higher GDP growth, increased revenues from incremental GDP turning into faster growing revenues from capital gains, with a key output of all this being a better debt-to-GDP ratio. Ultimately, this is a very optimistic look at how AI is going to impact not only public markets but government receipts. And if you want to dig in at all, especially around one controversial, at least very highly discussable piece of the story, their last slide is called the Kowatu Fantastic 40, and it's their list of companies that they anticipate being at the top of the public market charts in the year 2030. The only notable one is OpenAI at a $1.6 trillion market cap as the ninth biggest company in the world. My guess is that Sam Altman is certainly going for something a little bit bigger than that. And don't even talk to Elon about his placement of XAI down at the 35th rank.
Now, the other report that I wanted to talk about a little bit was this Menllo State of Consumer AI report. Nothing in here is shocking, but it is confirmation of just how insanely fast everything is moving. 61% of American adults have used AI in the past 6 months, and 20% use it every day. They write, "This is no longer experimentation. It's habit formation at an unprecedented scale." Now, the report comes from a study of 5,000 US adults. And beyond that, this big banner headline statistic also shows some really interesting things about, for example, who's using AI. Menlo writes, "While Gen Z leads overall AI adoption as expected, millennials emerge as power users, reporting more daily usage. At the same time though, nearly half of baby boomers have used AI in the past 6 months, with 11% still using it daily. Those who work use AI more. 75% of employed adults use AI as compared to 52% of unemployed adults. And the higher income you are, the more you use it as well. 53% of households earning under $50,000 use AI as compared to 74% of households earning $100,000 or more. 85% of students 18 and older reported using AI, and 15% of students are liars."
One of the unexpected findings about users they had was that parents are power users. While 54% of non-parents have used AI, 79% of parents have. 29% report using AI every day, which is almost twice the rate of non-parents. And 34% of them are using it to manage child care. This resonates because one of the most common gateway experiences that I've seen over the last 2 years is parents who do something with AI vis-à-vis their kids. They tell stories with ChatGPT and choose-your-own-adventure style, or they have their kids make images with them on Midjourney, or they create specialized coloring books that have their kids' images and favorite characters together. These really capture the wonder and joy and capital F fun of AI in a way that some of the work cases don't, and then translate to people experimenting with other more practical use cases in other parts of their life as well.
When it comes to which tools they use, there are really two parts of the story: convenience and first-mover advantage. ChatGPT has name recognition, representing 28% of the total general AI assistant usage among adults. And then after that, it's all about what's in your face and what you're already using. Gemini had 23%, Meta had 18%, Alexa had 18%, and even awful Siri had 16%. And what are people using AI for? I will caveat that this is maybe the least consistent question if you look across all AI studies. Harvard reported a bunch of use cases that look totally different from some other use case reports. So, you have to take all of these with a big grain of salt. However, the top 10 most common ways that people use AI in everyday life were writing emails, researching topics of interest, managing to-do lists, general writing support, meal planning, managing expenses, taking and organizing notes, creating images, researching purchases, and researching health questions. All of those had between 14% and 19% of US adults using them for that purpose.
Now, while in general AI saw the highest penetration, where it fit into activities that people were already doing, the biggest example of where AI is opening up behavior that wasn't there before seems to be in vibe coding. They found 47% of respondents applying AI to coding for work or school, and 41% using it for personal projects. Menllo concludes, "Much like photo editing or slide design before it, AI-assisted coding is becoming foundational digital literacy."
Now, if this is all about the people who have adopted AI, what about the 39% of Americans who aren't using it? The most common answer is the simplest. 80% say that they prefer people over AI. 71% say they're worried about data privacy. 63% just don't see a need. 58% don't trust AI information. 48% said they don't know how to use AI. 40% said that they think it's biased, which is another version of not trusting it. And 27% said that they lack access to tools.
So what do they think happens next? Their first prediction is a second wave, moving from generalized to specialized tools. Basically, now that nearly 2/3 of Americans are using AI, but it's mostly still in the realm of ChatGPT and generalized assistance like that, there may be room now for more specialized tools that expand and get into specific use case opportunities, like enterprises. They anticipate we move from assistance to full automation and agentic workflows. They think that we're likely to see some multiplayer modes that integrate social experiences with AI. Menllo is guessing that voice AI is going to be a much bigger part of the ecosystem, and that physical AI will enter the home, just like the experimentation we're seeing in the enterprise. Revenue models are likely to diversify beyond subscriptions as well.
To me, this report, I think, is extremely confirming of what many our expectations would be, but in some ways with even higher numbers than you might expect. It wasn't very long ago that we were talking about how AI was the fastest-growing technology ever because it had only taken it 2 years to go from 0 to 40% penetration of US households. We've grown another 50% then, all the way up to over 60% penetration. And there are no signs that this is slowing down. So friends, that is going to do it for our midyear recap of where we are with AI.