Hyperscalers, Humanoids and the Attention Economy

And some good news from Skin2Neuron

We had some good news from Skin2Neuron, as one of the world’s largest pharmaceutical companies completed an investment this week, marking their first investment in an Australian biotech. This is strong validation of our thesis and guarantees plenty of offshore attention at the highest levels.

You can see our interview with CEO Brendon Boot here, or read the transcript here.

I recorded a podcast with Ellianna covering:

  • Nvidia’s Q1 performance

  • H20 chip ban, and why these bans are bad for America

  • Mary Meeker’s report on AI and capex

  • Autonomous driving takes significant market share off Uber in San Francisco

  • The Lidar debate - what we now know in 2025

  • SpaceX Starship and 2026 Mars launch window

  • Humanoid robots, what does this mean for global GDP?

  • The Attention Economy

  • ChatGPT takes significant query share from Google, how this might play out

  • Capex dynamics at Amazon, Google and Microsoft

Transcript

Michael  

How are you doing Ellianna?  

Ellianna 

I am good, thanks. How are you?  

Michael  

We had Nvidia reporting and they posted another set of astounding numbers. The stock had sold off after the Trump administration extended the ban of their H20s, which were chips designed specifically for the Chinese market, so people were expecting a potentially a weak quarter.  

But it came in strong, revenues were up 12% quarter-on-quarter, up 69% year-on-year, and though earnings came in a bit light there was just no sign of slowing demand.

Everyone’s waiting for the day that the hyperscalers slow down their purchases, perhaps maintaining capex but not growing it, which could cause a derating in Nvidia. But so far, that hasn’t happened. 

Ellianna 

I saw their data center revenue was $39 billion, with an interesting split between other main areas of the business. Why do you think that is?  

Michael 

Ultimately, it's still a data center business. That's their core set of products that they're selling for large language model training and inference. 

The auto sector has been kind of weak, roughly flat on the quarter but still up 72% year-on-year. Gaming is recovering at 42% year-on-year. But the growth has really been AI-related, and it’s not just Nvidia, it’s really been behind a big chunk of the Nasdaq 100's rally, led by Nvidia and the companies on the right side of AI.

At the index level the interesting thing about this round of disruption is that the new big thing has heavily favoured incumbents. It’s usually not like that.

The clearest winners are Microsoft, Amazon, Google, everybody with a scaled computing platform, because it's dramatically increased the demand for compute, and there’s very few companies that can operate at this scale.

This is very different to Web 2.0 where all the action was in an entirely new set of companies, who basically built web apps on databases and disrupted all kinds of traditional industries.

That is in process again here, with a lot of innovation at the product layer, with people using the models to create AI accountants, a bookkeepers, etc. But it’s harder to pick the winners, and everyone is building on top of the same large language models, so it’s not clear what’s defensible.

Even the open versions have to be hosted somewhere, and that's where the big tech companies are most comfortable. They want to provide that compute layer, the databases, manage the network traffic, that's where they're most comfortable. They will win at the base level whoever wins at the product level in each category. 

So it's been a moment of immense technological progress, but so far it has benefited incumbents the most, and that's been good for the Nasdaq and good for US equity indices. It’s not like a new set of companies are disrupting the hyperscalers, the same way the hyperscalers cut a swathe through global industry when they rose up.

Ellianna 

There was a really interesting slide chart from Mary Meeker (see presentation).  

Michael 

Yeah if you go through that deck you come away pretty bullish and excited about things.

LLMs are growing fast than the internet. One of the charts showed hyperscaler capex increasing by 63% year-on-year. And it's interesting that 63% number is very close to the 69% that Nvidia grew at. 

Another chart showed how autonomous driving has taken off in San Francisco. I mean, have you ever seen one of those cars? 

Ellianna 

No, I haven't. 

 Michael 

They must have come after, after you left the United States. 

So, they've taken 27% market share in San Francisco, with ride-sharing dropping from 34% down to 19%. So they knocked Uber's market share down from 34% to 19% in barely a year and a half, while taking more than a quarter of the market and growing super fast.

Which might indicate a pretty clear winner and loser. I think Uber's going to be challenged.

There was an idea that Uber would be the network and autonomous vehicles would plug into it and take advantage of their user base, but that hasn't happened here. 

Google's Waymo has their own app, and there's probably first mover advantage here, with all the data and customer preference that entails. Perhaps if we end up with four or five different competitive autonomous vehicle platforms it will make sense for a customer demand aggregator like Uber.

But at the moment, the only people in the game in town is Waymo, though of course you'd have to assume that Tesla and others come in soon.  

It will be really interesting to see what happens around lidar. There was a huge debate in the tech industry over whether normal cameras were enough, whether visible light was sufficient for autonomous driving (it’s enough for us). 

Other than Elon Musk, most thought you need lidar as well. Lidar gives you distance and speed, and you get all this extra data that you otherwise would have to infer from visual pixels.

If you just use the visible spectrum it comes to a compute problem, with enough data you can get the information you need, and also don’t have to pay for Lidar. But really lidar is not a huge chunk of the cost of a car. And even Tesla uses lidar when training their models.

I’m sure you could use vision and end up with a much better driver than a human. But there are situations where having that extra distance and speed data is helpful, and engineering redundancy is almost always safer, for example in low vision environments, if it's snowing or raining.

One of the Tesla’s drove into a white truck it couldn’t see or interpret properly. On Lidar, that would be extremely obvious. There was another crash where it went straight into one of those dividers, like a freeway split and just went bang straight into the middle, it was probably quite hard to see so maybe the visual model couldn't get a proper read on it. But lidar would've been able to. 

There were a huge number of companies that were set up to sell lidar in 2020 and forecast that they would win contracts the next year, or the year after. 

But ultimately the big auto manufacturers just put them off indefinitely. And today there's very little lidar being sold. None of those companies have achieved significant revenues, and it's 2025. If you look at their old forecasts, they were all assuming that they'd be making serious bank by now, and it just didn't happen. 

 It's kind of interesting to see it play out. 

Now, an autonomous platform with lidar is the first to be operational. And they’ve quickly taken 27% of the market in SF. 

 Ellianna 

As autonomous driving becomes more popular, do you think that more people would be willing to pay extra for a safer option?  

 Michael 

Maybe, if say, Waymo develops a better reputation for safety. 

I imagine also if there was a Tesla-specific app, there would be some people that just don't like Elon Musk and wouldn't use it. The problem with being so political is that by definition an alliance of half the country disagrees with you. It's not good business to be very political. 

I mean how much extra would you pay for a safer car? What's the cost of lidar? These are already expensive cars. Once they're souped up with all the extra equipment you're probably talking about a few percent extra to have a sensor that can measure the distance and speed of every object around it, I'd definitely pay it. There’s new solid state lidars that are a fraction of the cost of Waymo’s, and that’s probably the future.

Waymo’s large, expensive, but effective lidar system

If Waymo comes out with data that their car is safer, that will definitely drive consumer choice and that'd be very hard to fix if one company gets reputation for being safer.

Safety is the kind of thing that really does drive consumer demand, I mean look at the emphasis on safety in auto advertising. It's a key part of the sales pitch of new cars. Cars are designed around safety, and it’s driven the entire form factor. People worry about dying in a car crash.  Rationally or not, it looms large in everybody's mind. It probably is rational.

 Ellianna  

On the topic of Elon Musk, SpaceX had another Starship test flag that was successful this time. 

 Michael 

Yeah. And he said there's a launch window to Mars in 2026 that he's going to try and hit.  All of a sudden, that doesn't seem so far away. It won't be manned, but what he did allude to is that he's building these humanoid robots at Tesla and will ship these to Mars.

Like think of all the science fiction films where someone’s stuck on Mars and loses contact with Earth, like that Matt Damon film, it's a common trope in science fiction.

Now looks like there would be an army of humanoid robots there ready to help.

Ellianna 

Didn't they have applications a couple years ago for people who wanted to go to Mars non-return trip?  

 Michael 

Yeah, it's going to be great content. Whoever goes is going to be super famous. They'll get more than their 15 minutes, would you go to Mars? 

 Ellianna 

No, I like it here! But I think the point about humanoid robots is so interesting, not just for exploration, but think about dangerous jobs like working on oil rigs. 

 Michael 

Yeah we're pretty dextrous. Adding a bit of strength from a mechanical frame and as they say, the world's designed for people, right? So a humanoid can immediately slip in to tasks that people do today.   

Something I’ve been trying to figure it out is what this means for GDP. Like let’s say you have an extra person, a new baby is born. That adds to global GDP. They’re going to live their life, pay taxes, consume, spend.

What happens if we’re spawning these humanoid robots, they’ll have their own economy as well, right? They need to be supplied, they need energy resources. They will consume, they would earn, they would do productive work.  Does that then  decouple global GDP from the size of the global population?   

Like, that'd be really interesting. If it plays out like that, basically everything is a buy right now. There's going to be significant growth ahead.

Ellianna

There’s some other interesting charts on the explosion of people using AI.

Michael

Yeah there's 800 million people using ChatGPT now, 20 million subscribers, and that's enough to get $8 billion of revenue, which has come from nothing only a couple of years ago. These business are big. Google said they had a billion plus people interacting with their AI, which I think is quite cheeky, because presumably they’re talking about those little AI summaries below search. If you clicked on it once in a month, you've then interacted with it that month and then they get that data.

Their original insight was that you could rank websites based on how many other websites linked to it. Then once people started using it, they could optimize it. It's just simple loop, like just improving and iterating the product based on what people like. 

Same with Instagram, Facebook, or Amazon, where if you go to Amazon and sign in, your front page will be completely different to mine.

They know what people are interacting with, which is a better indication of what they might like than what they say they might like. Now they know when their AI is being helpful.

The challenge is that with the AI summaries, and all the blue links that they’ve jammed in their to jam up revenue, it’s hard to get the actual search results you were after. Their financial performance has come at the cost of product quality.

Attention Economy

There's kind of like an attention economy. Say there’s 7 billion people on the planet and there's however many hours in the day, and you're only awake for maybe 17 or 18 of those hours.

There's only so much attention up for grabs.

The way we’re designed, we can’t actually multitask.

That means that every product is competing for that limited attention space.

Netflix isn’t competing with TV or Tik Tok, it’s competing with whether you go surfing, whether you go for a run, whether you like play a game, whether you are at work doing actual work instead of mucking around on the internet. 

It's a global market for attention and it's kind of interesting way of looking at it. You can see, okay, people are spending hours on Facebook. Clearly that's extremely valuable. The value of those companies generally tracked the amount of time people were spending on them.   

Which is why it’s interesting as a non-financial metric, the amount of time people were spending on these apps came way ahead of the market cap of those companies. And it was a really good way of identifying those companies earlier. And now with AI generation, the feedback loop is insane. There's people worrying about ‘AI slop’. 

Particularly for kids, if you've ever seen young children on an iPad, the things they listen to are nuts. You know? They're like, not speaking English. It's like weird sounds and flashing lights and images bouncing around.

At least in the past, that was based on what a team of people could experiment with, who you could try a few different things, create a few different videos, see what's working, optimize it. That limit - that team of people - is now gone. 

So you could create, a five an hour long children's cartoon, right? You could optimize that. You could run endless iterations of different things , you can monitor where when people are paying attention and then basically create something extremely hyper addictive. 

This has come first for children but also for adults. I think it’s easier to conceptualize it happening for kids because kids love cartoons whereas adults prefer more realistic content. But that’s changing, you’re already seeing these AI Instagram models, which can quickly attract followers, again, at the expense of something else, as there’s only one attention economy. Every time one of these fake AI kind of models goes up and people spend time on that, that comes directly out of time that they might have been swiping through, you know, a real person. And I guess that's just starting to take off now.   

Again, to link it back to investments, that’s going to take a lot of compute and the return on investment in compute is going to be extremely high, perhaps higher than ever.  There's probably a ton of opportunities around that. For an entrepreneur, you know, you could, you could just start this feedback loop and create content without knowing or caring what it is you’re making.  

 Ellianna 

It will be really interesting to see how kids grow up and how and the videos and attention span influence the way that they think. Creators in industry generating script visuals are already amazing.  I think so many industries are going to change. 

 Michael 

Yeah the latest video models are a huge breakthrough, combining audio in videos from prompts. It was an inevitable breakthrough, but it's one of those ones that's like, oh, it's already here. There's so many business opportunities. It's a really exciting time.  

Ellianna 

On the topic of business opportunities, if this is changing the way that kids learn, the way that people work and the way that we consume entertainment, there would be some negative risks along with that in the sense of like fake news, kids cheating in school on tests and an assignment. 

What do you think the security side of this?

Michael 

There’s already an arms race between AI generators and software detecting AI content. There's open space for products that can immediately, identifies if something is AI or not. 

Kind of like antiviral software where bad actors try to stay one step ahead, similarly AI generators will try and make their content harder to identify as AI. AI detection software might soon always be one step behind. There's probably a whole class of valuable businesses there, the same way a bunch of people made money out of antiviral software. It’s probably worth studying how that industry played out and how a few companies took the lions share of the market, probably by partnering with major companies.

In terms of education and tests and exams, I think what's going to happen is that we go back to pen and paper. That's probably one example in a trend in this new AI world where there is a shift towards in-person things. It's starting to happen in the social space where after all this online dating, people are trying to foster in-person events. 

There's a group that does like Wednesday night dinners, which has just taken off massively around the world, where eight people have dinner who don’t know each other. They'll organize it every Wednesday. 

Which sounds like a kind of nice little niche thing. I think it's like taken off, and is huge around the world and is a massive business. There's clearly like that demand for that kind of thing, and so I think there's probably a lot of opportunities now in going in the opposite direction, now that we’re all hyper online and bombarded with AI content and all these frustrating things.

And you know, your wife or husband might just be the person who the algorithm showed you that particular day when you were single and swiping, and so your whole life trajectory depended on, which particular 10 people the algorithm showed you that day. 

So there’s probably opportunities in zigging instead of zagging with everyone else. But we'll see.

One big question is how much of search is going to ChatGPT? There's a number of different ways of calculating it. 

In this report it said that ChatGPT hit 365 billion annual searches in two years, which took Google 11 years. So ChatGPT is growing faster than Google did. Google is processing 8.5 billion searches a day. ChatGPT is processing 1 billion searches a day. So ChatGPT has taken more than 10% market share of searches. That's a huge, huge outcome and is why Google's share price has been under a bit of pressure lately. 

I actually think that Google is a long term or even short term winner. . Because they have the video in YouTube, because they have the compute, they have some of the best models, they have that billion users a month interacting with their AI. I think actually the market's probably got it wrong and they're a long term winner, but the stock has been, a pretty poor performer. I guess ultimately this is bad news for search, which has been one of the most successful products ever, it will take a lot to make up any stagnation there.

Google is down 11% this year, which amongst large companies is pretty bad. Then again, I'm probably using ChatGPT more than Google, and I can't be the only one. 

 Ellianna 

How are they calculating those searches? Is that one conversation with ChatGPT, or would that include all the queries?

Michael 

I think it’s actually each question. But then okay, what does that mean for Google's revenue? Now, the way I think about that is there’s a global advertising pool, a certain amount of global spending. 

When Google rose they did two things. Firstly, they wiped out every other advertiser. They took that wallet share directly from them. They hit the newspapers, billboards, TV. And obviously once companies shrink things become really tough. All companies have some level of fixed costs, so shrinking revenues is usually disastrous. Healthy businesses need at least some level of growth. And whole industries got effectively wiped out.

But there was a second part to it, it wasn't just, taking market share. It was the fact that it was so valuable to get a search that you could directly see your return on investment, and it was so much higher than say blanket untargeted advertising, so you could justify spending a lot more money.

The total market for advertising dramatically increased. And now every consumer related business has to spend money on Google and Facebook.  If they don't, their competitors will, and they'll get the additional market share. 

So effectively, Google and Facebook taxed every consumer business. Every successful startup that raises money from VC, the firs thing they do is pump a huge chunk of that into Google and Meta. Some don’t recoup that cost. But some get a high return on investment, multiples of what they spend.

So every successful business ends up spending more and more with those companies. If you have a good riff, a good marketing line, a good ad, you can just pour money into those companies and get multiples back. For Google and Meta it didn't matter who won in each industry, they were the beneficiary. 

This is kind of like how hyperscalers are winning by providing the compute layer to anyone who wants to compete in AI. The winners will just spend more and more.

Now, where does that sit with ChatGPT and other LLMs now?

The ChatGPT is by far the dominant player for consumers. But the real question is does that take advertising dollar share off Google?

Possibly. They’re not doing it now, but if you start saying to ChatGPT, I’d like to buy an SUV for my family, help me think through this and give me the options. They’re the most valuable kinds of searches, where you have a targeted buyer.

Is there space for advertising there?

Probably somewhere, but it's probably not as valuable as it would be as a Google search, as you’re expecting advice not ads.

And then go back to that idea about that global market for attention. Maybe just taking attention, that means that those Google searches are less valuable. You know, people are less likely to buy, the return on ad spend goes down. It could destroy the value of those searches without actually creating equivalent or more advertising value elsewhere.

The bull case for Google is that the highest value searches are still going to be done there.

Like if you want to buy a car, or book a flight, or find a toaster, though that is already going to Amazon, which took a decent chunk of search queries. And think of how much more valuable a search query like that is in Amazon, when you’re ready to buy with one click. Those are the valuable queries. 

Compare that to information questions like, how big is the population of Australia? That kind of search is just information based. Very little advertising value. It’s the targeted searches by a qualified buyer that are the most valuable. But how this all plays out remains to be seen.

Like does the shift of attention away from Google into ChatGPT, does that actually shrink the market? Unlikely, but it's certainly possible. If there’s less people seeing and clicking on ads, they’re less valuable. But then again, everyone needs a toaster, so that buyer can probably be targeted somewhere. I don’t have a strong view about it, but this is the way to look at it. Taking or losing market share, and whether searches at Google, Amazon, or ChatGPT are more or less valuable, does the total market and return on ad spend increase or decrease .

I think that's a useful lens to look at the future of that industry. 

And remember, ChatGPT is monetized by subscriptions. They might find it’s better to optimise their product for that, and then the attention they pull away from Google will end up just damaging Google.

Like, is ChatGPT going to show ads next to your coding question? It's probably the wrong way to monetize. You'd be better off just charging a bit for high users, who presumably are getting the most value and willing to pay the most.

Another chart we saw was that the number of Google developers has gone up five times from May, 2024 to May 2025, so a fivefold increase in developers in that ecosystem in a year, which is probably a good proxy for the increase in developers. There were similar numbers for Nvidia CUDA developers.

 This is probably due to the ease of coding right now, these large language models have lowered the barrier to coding and massively increased the market of developers. Which again translates into demand for all those big tech companies and probably increases the entire rate of technological progress, which is quite interesting, there’s five times as many people coding and developing software. 

There was also this stat where there’s been a huge increase in the number of patents issued. So if that’s a proxy for technological progress, that's a huge increase. It's not a six times increase, it's a six times 18 increase. So almost over a hundred fold increase in the rate of patents. Now maybe some of it's AI slop - I mean, we know a lot of it probably is.

But there's no reason not to assume that a big chunk of it is actually real technological progress. There's another interesting stat here that says that years it took to reach a hundred million users, 0.2. I remember when TikTok was the one that was, that everybody was talking about, and that was probably four or five times longer to get to a hundred million users. So just gives another example of that growth. 

This was an pretty interesting chart on AWS that shows that capex, when they were building out their cloud platform, was 27% of revenue, which decreased to 4% in 2018. So clearly they built all this infrastructure, then decided it was harvest time, let's make some money off all this infrastructure that we've built, but then changed course, rapidly increased capex and now it's 49% of revenue. 

They're spending an enormous amount, so the money that they're making is half every dollar that they get in revenue. Half of it is going straight into capex with the bulk of that, certainly the bulk of the profits of that going to one company, Nvidia. 

Why don't we wrap up there. Ellianna, thanks so much for joining. 

 Ellianna 

Sounds great. Thanks for having me again.